check-parallel: Running tests without check
cover check-parallel: Running tests without check
Hi folks,
This set of patches is intended to move check-parallel away from using check to execute tests.
To do this, we need to share a bunch of check code between check and check-parallel. This is mainly the code that parses and builds the test list, the config section parsing and iteration, and the test execution loop itself.
To do this, test list parsing and building is factored out of check into common/test_list. check is converted to use the test list functions at the same time, and then check-parallel is converted to use the factored code to directly build it's test list rather than the open coded grep hack it currently uses. This allows check-parallel CLI to use the same group selection interface as check.
The next change is to factor the config section parsing out of common/config and move it to common/config-section. This allows check-parallel to parse and implement section iteration itself without needing to run all the environment setup code in common/config. This also allows check-parallel to implement it's own config section to define the device sizes that it will use independently of the sections that run tests.
Next, we change check-parallel to use a global test list that runner scripts can safely dequeue the next test to run. This uses a test list file and a lock file to serialise access to the file. Hence a runner can dequeue the next test and remove it from the test list file without racing with any other runner trying to dequeue the next test to run. This means we get rid of the static per-runner test lists that result in many runners finishing and going idle while other test runners have pending tests still to run. i.e. all test runners keep executing tests until there are no tests left in the queue, hence keeping utilisation as high as possible across the test run.
Then we factor the test execution loop out of check and put it in common/test_exec. This abstraction makes the results array part of the test execution, as well using a context defined helper "_run_seq" to do the actual execution of the test. This allows the test execution loop to be completely generic, whilst allowing check and check-parallel to do completely independent things with individual test execution and overall results reporting.
Finally, we change check-parallel to run tests directly via the common/test_exec infrastructure rather than executing them via check. This requires a new helper function that does the test environment setup in the private mount+pid namespace, but this is much simpler and faster than using check itself to execute individual tests. This last bit of functionality is still a work in progress, so this specific patch is still tagged with [RFC].
There are lots of other bits of changes. The way common/rc and common/config are used is changed. common/config only sets up the execution environment now, and should not contain any code that needs to be executed outside of environment setup. It should only be sourced once at the highest level to set up the environment, and never called again.
common/rc is similar - all directly executed code has been removed from it, and that is now called from the high level code that needs initialisation work done. It no longer sources common/config, either. The test preamble does not need to run init_rc() any more; they just need to source the generic and fs specific functions the tests may run. Also, because check does some weird things and lots of _requires....() functions assume the TEST_DEV is mounted without first running _require_test(), it also needs to ensure the TEST_DEV is mounted...
check-parallel can now take a "-t N" parameter to specify how many execution threads it will use. If this is not specified, it will default to the number of CPUs in the machine. Testing with 4p restrictions show that check-parallel will run the quick group 3.5x faster on a 4p system with 8 execution threads than it will with a single execution thread. IOWs, even on small test systems, check-parallel can result in dramatic reductions in test runtime over check.
On a 64 p machine, testing XFS with the quick group drops from 61 minutes to just under 4 minutes. Testing XFS with the auto group drops from 246 minutes to just under 8 minutes.
Other miscellaneous stuff in the series:
- kill non-numeric test name support
- creating common/exit for all the general test exit functions to fix circular dependencies between common/rc and common/config
- fix iscratch_mkfs_sized to make USE_EXTERNAL on XFS work the same as ext4.
- dm-logwrites devices are now created by check-parallel
- several test conversions from sync() to syncfs()
- removal of a could of stale .c test source files.
- address poor CPU count scaling in a couple of tests
I have tried not to cause any regressions for people running plain check. I've tested that a bit with XFS and ext4, but I can't guarantee that there aren't issues I haven't uncovered. e.g. btrfs, as yet, is untested. It is unfortunate that the problem I seek to address - running exhaustive check testing across many filesystem types and configurations is prohibitively expensive in terms of time
- is the very reason I can't really adequately test check for regressions as I develop check-parallel functionality...
Thoughts, comments and code review all welcome!
-Dave.
.gitignore | 1 - check | 727 ++++-------------------------------- check-parallel | 351 ++++++++++++++--- common/config | 612 +----------------------------- common/config-sections | 461 +++++++++++++++++++++++ common/dmlogwrites | 5 +- common/exit | 48 +++ common/preamble | 19 +- common/rc | 253 +++++++++++-- common/report | 2 +- common/test_exec | 377 +++++++++++++++++++ common/test_list | 308 +++++++++++++++ common/test_names | 8 +- new | 24 -- src/Makefile | 4 +- src/bulkstat_unlink_test.c | 12 +- src/bulkstat_unlink_test_modified.c | 193 ---------- src/fsync-tester.c | 2 +- src/open_by_handle.c | 6 +- src/scaleread.c | 224 ----------- src/scaleread.sh | 64 ---- src/stale_handle.c | 15 +- tests/generic/531 | 8 +- tests/xfs/259 | 1 - tests/xfs/271 | 2 - tools/run_test.sh | 116 ++++++ 26 files changed, 1954 insertions(+), 1889 deletions(-)
PATCH 1/28 fstests: remove support for non-numeric test names
From: Dave Chinner <[email protected]>
We haven't had any tests using the "999-the-mark-of-fstests" name format for a long time. Th eonly test that used this format was xfs/191-input-validation, and that got removed in 2022 by commit c1941d6f5 ("xfs/191: remove broken test").
However, the infrastructure for this naming convention still exists, so lets get rid of that dead code so we don't have to carry it anymore.
--- check | 15 --------------- common/test_names | 8 +------- new | 24 ------------------------ 3 files changed, 1 insertion(+), 46 deletions(-)
check +0 -15
@@ -856,21 +856,6 @@ function run_section() for ((ix = 0; ix < ${#_list[*]}; !${#loop_status[*]} && ix++)); do seq="${_list[$ix]}" - if [ ! -f $seq ]; then - # Try to get full name in case the user supplied only - # seq id and the test has a name. A bit of hassle to - # find really the test and not its sample output or - # helping files. - bname=$(basename $seq) - full_seq=$(find $(dirname $seq) -name $bname* -executable | - awk '(NR == 1 || length < length(shortest)) { shortest = $0 }\ - END { print shortest }') - if [ -f $full_seq ] && \ - [ x$(echo $bname | grep -o "^$VALID_TEST_ID") != x ]; then - seq=$full_seq - fi - fi - # the filename for the test and the name output are different. # we don't include the tests/ directory in the name output. export seqnum=${seq#$SRC_DIR/}
common/test_names +1 -7
@@ -2,11 +2,5 @@ # Valid test names start with 3 digits "NNN": # "[0-9]\{3\}" -# followed by an optional "-": -# "-\?" -# followed by an optional combination of alphanumeric and "-" chars: -# "[[:alnum:]-]*" -# e.g. 999-the-mark-of-fstests -# VALID_TEST_ID="[0-9]\{3\}" -VALID_TEST_NAME="$VALID_TEST_ID-\?[[:alnum:]-]*" +VALID_TEST_NAME="$VALID_TEST_ID"
new +0 -24
@@ -50,30 +50,6 @@ export AWK_PROG="$(type -P awk)" echo "Next test id is $id" shift -read -p "Append a name to the ID? Test name will be $id-\$name. y,[n]: " -r -if [[ $REPLY = [Yy] ]]; then - # get the new name from user - name="" - while [ "$name" = "" ]; do - read -p "Enter the name: " - if [ "$REPLY" = "" ]; then - echo "For canceling, use ctrl+c." - elif echo "$id-$REPLY" | grep -q "^$VALID_TEST_NAME$"; then - if [ -e "$tdir/$id-$REPLY" ]; then - echo "File '$id-$REPLY' already exists, use another one." - echo - else - name="$REPLY" - fi - else - echo "A name can contain only alphanumeric symbols and dash!" - echo - fi - done - - id="$id-$name" -fi - echo "Creating test file '$id'" if [ -f $tdir/$id ]
Signature · 1 line
2.45.2
Re: [PATCH 01/28] fstests: remove support for non-numeric test names
Quoted 10 lines · Dave Chinner
From: Dave Chinner <[email protected]>
We haven't had any tests using the "999-the-mark-of-fstests" name format for a long time. Th eonly test that used this format was xfs/191-input-validation, and that got removed in 2022 by commit
c1941d6f5("xfs/191: remove broken test").However, the infrastructure for this naming convention still exists, so lets get rid of that dead code so we don't have to carry it anymore.
Any other reason why we are planning to remove this convention apart from the fact that it is not being used for a long time? But yes,I agree that only numeric names are easier to refer and we can also use some one liner shell script tricks to run several tests - something like
./check xfs/{1...100} to run all the tests from xfs/1 xfs/2 ... xfs/100 (ofcourse assuming all these tests with these numbers exist).
Quoted 99 lines
Signed-off-by Dave Chinner <[email protected]>--- check | 15 --------------- common/test_names | 8 +------- new | 24 ------------------------ 3 files changed, 1 insertion(+), 46 deletions(-)check +0
-0@@ -856,21 +856,6 @@ function run_section() for ((ix = 0; ix < ${#_list[*]}; !${#loop_status[*]} && ix++));do
check +0
-2seq="${_list[$ix]}" - if [ ! -f $seq ]; then - # Try to get full name in case the usersupplied only
check +0
-1- # seq id and the test has a name. A bit ofhassle to
check +0
-1- # find really the test and not its sampleoutput or
check +0
-5- # helping files. - bname=$(basename $seq) - full_seq=$(find $(dirname $seq) -name $bname* -executable | - awk '(NR == 1 || length <length(shortest)) { shortest = $0 }\
check +0
-3- END { print shortest }') - if [ -f $full_seq ] && \ - [ x$(echo $bname | grep -o"^$VALID_TEST_ID") != x ]; then
check +0
-4- seq=$full_seq - fi - fi - # the filename for the test and the name output aredifferent.
# we don't include the tests/ directory in the name output.check +0
-0export seqnum=${seq#$SRC_DIR/}common/test_names +1
-7@@ -2,11 +2,5 @@ # Valid test names start with 3 digits "NNN": # "[0-9]\{3\}" -# followed by an optional "-": -# "-\?" -# followed by an optional combination of alphanumeric and "-" chars: -# "[[:alnum:]-]*" -# e.g. 999-the-mark-of-fstests -# VALID_TEST_ID="[0-9]\{3\}" -VALID_TEST_NAME="$VALID_TEST_ID-\?[[:alnum:]-]*" +VALID_TEST_NAME="$VALID_TEST_ID"new +0
-1@@ -50,30 +50,6 @@ export AWK_PROG="$(type -P awk)" echo "Next test id is $id" shift -read -p "Append a name to the ID? Test name will be $id-\$name.y,[n]: " -r
new +0
-8-if [[ $REPLY = [Yy] ]]; then - # get the new name from user - name="" - while [ "$name" = "" ]; do - read -p "Enter the name: " - if [ "$REPLY" = "" ]; then - echo "For canceling, use ctrl+c." - elif echo "$id-$REPLY" | grep -q "^$VALID_TEST_NAME$";then
new +0
-2- if [ -e "$tdir/$id-$REPLY" ]; then - echo "File '$id-$REPLY' already exists,use another one."
new +0
-6- echo - else - name="$REPLY" - fi - else - echo "A name can contain only alphanumericsymbols and dash!"
new +0
-7- echo - fi - done - - id="$id-$name" -fi - echo "Creating test file '$id'" if [ -f $tdir/$id ]
Re: [PATCH 01/28] fstests: remove support for non-numeric test names
Quoted 8 lines · Nirjhar Roy (IBM)
Quoted 7 lines · Dave Chinner
From: Dave Chinner <[email protected]>
We haven't had any tests using the "999-the-mark-of-fstests" name format for a long time. Th eonly test that used this format was xfs/191-input-validation, and that got removed in 2022 by commit
c1941d6f5("xfs/191: remove broken test").
Quoted 5 lines
Quoted 3 lines
However, the infrastructure for this naming convention still exists, so lets get rid of that dead code so we don't have to carry it anymore.
Any other reason why we are planning to remove this convention apart from the fact that it is not being used for a long time?
It hasn't been used because nobody has ever really seen much value in trying to describe the test in the test filename. It would be used if people valued it, right? Mostly, though, people complained about that one test with a weird name....
Quoted 6 lines
But yes,I
agree that only numeric names are easier to refer and we can also use some one liner shell script tricks to run several tests - something like
./check xfs/{1...100} to run all the tests from xfs/1 xfs/2 ... xfs/100 (ofcourse assuming all these tests with these numbers exist).
Right, that becomes more complex as soon as names have free-form components.
If you want to know what all the tests do, use the lsqa.pl to extract the initial comment in the test that describes what the test is exercising.
-Dave.
Signature · 2 lines
Dave Chinner [email protected]
Nirjhar Roy (IBM)
Reviewed-byRe: [PATCH 01/28] fstests: remove support for non-numeric test names
Quoted 12 lines · Dave Chinner
Quoted 11 lines · Nirjhar Roy (IBM)
Quoted 10 lines · Dave Chinner
From: Dave Chinner <[email protected]>
We haven't had any tests using the "999-the-mark-of-fstests" name format for a long time. Th eonly test that used this format was xfs/191-input-validation, and that got removed in 2022 by commit
c1941d6f5("xfs/191: remove broken test").However, the infrastructure for this naming convention still exists, so lets get rid of that dead code so we don't have to carry it anymore.
Quoted 6 lines
Quoted 2 lines
Any other reason why we are planning to remove this convention apart from the fact that it is not being used for a long time?
It hasn't been used because nobody has ever really seen much value in trying to describe the test in the test filename. It would be used if people valued it, right? Mostly, though, people complained about that one test with a weird name....
Yeah.
Quoted 7 lines
Quoted 7 lines
But yes,I
agree that only numeric names are easier to refer and we can also use some one liner shell script tricks to run several tests - something like
./check xfs/{1...100} to run all the tests from xfs/1 xfs/2 ... xfs/100 (ofcourse assuming all these tests with these numbers exist).
Quoted 6 lines
Right, that becomes more complex as soon as names have free-form components.
If you want to know what all the tests do, use the lsqa.pl to extract the initial comment in the test that describes what the test is exercising.
Thanks. I wasn't aware of this. This is useful.
I feel this change is useful. Feel free to add
--NR
Quoted 3 lines
-Dave.
Signature · 3 lines
Nirjhar Roy Linux Kernel Developer IBM, Bangalore
PATCH 2/28 _scratch_mkfs_sized: obey USE_EXTERNAL for XFS filesystems
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
_scratch_mkfs_sized is failing on my check-parallel setup because it has an RTDEV defined but USE_EXTERNAL is not set. Hence _scratch_mkfs_sized sees the RTDEV, adds a size parameter for it, then passes that to _try_scratch_mkfs_xfs() which looks at USE_EXTERNAL and so doesn't add a RTDEV to the filesystem.
The result is that _scratch_mkfs_sized -always- fails with this sort of error:
.....
** mkfs failed with extra mkfs options added to "-m rmapbt=1 -i exchange=1 " by test 300 ** ** attempting to mkfs using only test 300 options: -d size=536870912 -r size=536870912 -b size=4096 ** size specified for non-existent rt subvolume
Usage: mkfs.xfs
.....
Make the XFS code in _scratch_mkfs_sized look at USE_EXTERNAL like the ext4 code does.
--- common/rc | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
common/rc +1 -1
@@ -1264,7 +1264,7 @@ _try_scratch_mkfs_sized() xfs) local rt_ops - if [ -b "$SCRATCH_RTDEV" ]; then + if [ "$USE_EXTERNAL" = yes -a -b "$SCRATCH_RTDEV" ]; then local rtdevsize=`blockdev --getsize64 $SCRATCH_RTDEV` if [ "$fssize" -gt "$rtdevsize" ]; then _notrun "Scratch rt device too small"
Signature · 1 line
2.45.2
Re: [PATCH 02/28] _scratch_mkfs_sized: obey USE_EXTERNAL for XFS filesystems
Quoted 7 lines · Dave Chinner
From: Dave Chinner <[email protected]>
_scratch_mkfs_sizedis failing on my check-parallel setup because it has an RTDEV defined butUSE_EXTERNALis not set. Hence_scratch_mkfs_sizedsees the RTDEV, adds a size parameter for it, then passes that to_try_scratch_mkfs_xfs()which looks atUSE_EXTERNALand so doesn't add a RTDEV to the filesystem.
Okay so the function flow is like:_try_scratch_mkfs_xfs() --> _scratch_mkfs_xfs_opts --
Quoted 1 line
_scratch_xfs_options()-->_scratch_optionsmkfs -->
_scratch_xfs_options()
In _scratch_xfs_options() we have
[ "$USE_EXTERNAL" = yes -a ! -z "$SCRATCH_RTDEV" ] && \
SCRATCH_OPTIONS="$SCRATCH_OPTIONS
${rt_opt}rtdev=$SCRATCH_RTDEV"
_scratch_mkfs_xfs_opts() adds $mkfs_opts which already holds "rt_ops="-
r size=$fssize" but _scratch_xfs_options() doesn't add "-r rtdev=<realt ime_device>" because USE_EXTERNAL is not set and hence the error. This makes sense. We should fill up rt_ops only when USE_EXTERNAL is set.This looks good to me.
Quoted 32 lines
The result is that
_scratch_mkfs_sized-always- fails with this sort of error:.....
** mkfs failed with extra mkfs options added to "-m rmapbt=1 -i exchange=1 " by test 300 ** ** attempting to mkfs using only test 300 options: -d size=536870912 -r size=536870912 -b size=4096 ** size specified for non-existent rt subvolume
Usage: mkfs.xfs
.....Make the XFS code in
_scratch_mkfs_sizedlook atUSE_EXTERNALlike the ext4 code does.Signed-off-by Dave Chinner <[email protected]>--- common/rc | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)common/rc +1
-1@@ -1264,7 +1264,7 @@ _try_scratch_mkfs_sized() xfs) local rt_ops - if [ -b "$SCRATCH_RTDEV" ]; then + if [ "$USE_EXTERNAL" = yes -a -b "$SCRATCH_RTDEV" ]; then local rtdevsize=`blockdev --getsize64 $SCRATCH_RTDEV` if [ "$fssize" -gt "$rtdevsize" ]; then _notrun "Scratch rt device too small"
PATCH 3/28 fstests: move test exit functions to common/exit
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
Functions like _fatal() and _exit() need to be called from common/config as well as general common and test code. Having some of these test exit functions defined in common/config requires all contexts to source this file and run all the environment setup just to gain access to these functions.
There is a catch-22 with these functions - they cannot be defined in common/rc for all contexts to pick up automatically, because common/config needs them and some functions in common/rc depend on common/config being sourced first. And we can't define them all in common/config, because there are contexts where that hasn't been sourced that need a specific test exit function.
Solve this by move all the test exit functions to a new common/exit file and source that explicitly in the contexts that need it, hence removing the circular dependency between common/config and common/rc for defining these functions...
--- check | 2 ++ common/config | 16 +--------------- common/exit | 48 ++++++++++++++++++++++++++++++++++++++++++++++++ common/preamble | 1 + common/rc | 22 ---------------------- 5 files changed, 52 insertions(+), 37 deletions(-) create mode 100644 common/exit
check +2 -0
@@ -52,6 +52,8 @@ rm -f $tmp.list $tmp.tmp $tmp.grep $here/$iam.out $tmp.report.* $tmp.arglist SRC_GROUPS="generic" export SRC_DIR="tests" +. ./common/exit + usage() { echo "Usage: $0 [options] [testlist]"'
common/config +1 -15
@@ -40,6 +40,7 @@ # . common/test_names +. common/exit # all tests should use a common language setting to prevent golden # output mismatches.
@@ -96,15 +97,6 @@ export LOCAL_CONFIGURE_OPTIONS=${LOCAL_CONFIGURE_OPTIONS:=--enable-readline=yes} export RECREATE_TEST_DEV=${RECREATE_TEST_DEV:=false} -# This functions sets the exit code to status and then exits. Don't use -# exit directly, as it might not set the value of "$status" correctly, which is -# used as an exit code in the trap handler routine set up by the check script. -_exit() -{ - test -n "$1" && status="$1" - exit "$status" -} - # Handle mkfs.$fstyp which does (or does not) require -f to overwrite set_mkfs_prog_path_with_opts() {
@@ -121,12 +113,6 @@ set_mkfs_prog_path_with_opts() fi } -_fatal() -{ - echo "$*" - _exit 1 -} - export MKFS_PROG="$(type -P mkfs)" [ "$MKFS_PROG" = "" ] && _fatal "mkfs not found"
common/exit new file +48 -0
48 lines not on this page · raw message
common/preamble +1 -0
@@ -49,6 +49,7 @@ _begin_fstest() _register_cleanup _cleanup + . ./common/exit . ./common/rc init_rc
common/rc +0 -22
@@ -1798,28 +1798,6 @@ _do() return $ret } -# bail out, setting up .notrun file. Need to kill the filesystem check files -# here, otherwise they are set incorrectly for the next test. -# -_notrun() -{ - echo "$*" > $seqres.notrun - echo "$seq not run: $*" - rm -f ${RESULT_DIR}/require_test* - rm -f ${RESULT_DIR}/require_scratch* - - _exit 0 -} - -# just plain bail out -# -_fail() -{ - echo "$*" | tee -a $seqres.full - echo "(see $seqres.full for details)" - _exit 1 -} - # # Tests whether $FSTYP should be exclude from this test. #
Signature · 1 line
2.45.2
PATCH 4/28 check-parallel: report how many tests were _notrun
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
We already count tests "run" and tests that failed, but there is no indication of how many tests were _notrun. With tests actually failing to run because of things like bugs in _scratch_mkfs_sized being reported as "_notrun" instead of failing, visibility of changing numbers of _notrun tests is needed.
--- check-parallel | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)
check-parallel +4 -1
@@ -188,7 +188,10 @@ done; wait echo -n "Tests run: " -grep Ran /mnt/xfs/*/log | sed -e 's,^.*:,,' -e 's, ,\n,g' | sort | uniq | wc -l +grep Ran $basedir/*/log | sed -e 's,^.*:,,' -e 's, ,\n,g' | sort | uniq | wc -l + +echo -n "Tests _notrun: " +grep "^Not run" $basedir/*/log | uniq | sed -e 's,^.*:,,' -e 's, ,\n,g' -e 's,^\n,,' | wc -l echo -n "Failure count: " grep Failures: $basedir/*/log | uniq | sed -e "s/^.*Failures://" -e "s,\([0-9]\) \([gx]\),\1\n \2,g" |wc -l
Signature · 1 line
2.45.2
Re: [PATCH 04/28] check-parallel: report how many tests were _notrun
Quoted 7 lines · Dave Chinner
From: Dave Chinner <[email protected]>
We already count tests "run" and tests that failed, but there is no indication of how many tests were
_notrun. With tests actually failing to run because of things like bugs in_scratch_mkfs_sizedbeing reported as "_notrun" instead of failing, visibility of changing numbers of_notruntests is needed.
This looks okay to me. 2 questions below. Apart from that, this looks good to me.
Quoted 11 lines
Signed-off-by Dave Chinner <[email protected]>--- check-parallel | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)check-parallel +0
-0
check-parallel done; · line 191
Quoted 5 lines
check-parallel +1
-1@@ -188,7 +188,10 @@ done; wait echo -n "Tests run: " -grep Ran /mnt/xfs/*/log | sed -e 's,^.*:,,' -e 's, ,\n,g' | sort | uniq | wc -l +grep Ran $basedir/*/log | sed -e 's,^.*:,,' -e 's, ,\n,g' | sort | uniq | wc -l
Nit: /mnt/xfs --> $basedir is not related to this change, right? Should we move it to a different commit/patch?
Not related to this change - Can we change the above statement to something like it is done below for _notrun i.e
grep "Ran $basedir/*/log" | uniq | sed ... | wc - l
in this way we can get rid of the sort command?
--NR
Quoted 6 lines
check-parallel +3
-0+ +echo -n "Tests _notrun: " +grep "^Not run" $basedir/*/log | uniq | sed -e 's,^.*:,,' -e 's, ,\n,g' -e 's,^\n,,' | wc -lDon't we need to sort before uniq? echo -n "Failure count: " grep Failures: $basedir/*/log | uniq | sed -e "s/^.*Failures://" -e "s,\([0-9]\) \([gx]\),\1\n \2,g" |wc -l
Re: [PATCH 04/28] check-parallel: report how many tests were _notrun
Quoted 26 lines · Nirjhar Roy (IBM)
Quoted 7 lines · Dave Chinner
From: Dave Chinner <[email protected]>
We already count tests "run" and tests that failed, but there is no indication of how many tests were
_notrun. With tests actually failing to run because of things like bugs in_scratch_mkfs_sizedbeing reported as "_notrun" instead of failing, visibility of changing numbers of_notruntests is needed.This looks okay to me. 2 questions below. Apart from that, this looks good to me.
Reviewed-by Nirjhar Roy (IBM) <[email protected]>Quoted 13 lines
Signed-off-by Dave Chinner <[email protected]>--- check-parallel | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)check-parallel +0
-0@@ -188,7 +188,10 @@ done; wait
Quoted 5 lines
check-parallel done; · line 191
Quoted 3 lines
check-parallel +1
-1echo -n "Tests run: " -grep Ran /mnt/xfs/*/log | sed -e 's,^.*:,,' -e 's, ,\n,g' | sort | uniq | wc -l +grep Ran $basedir/*/log | sed -e 's,^.*:,,' -e 's, ,\n,g' | sort | uniq | wc -lNit:
/mnt/xfs--> $basedir is not related to this change, right? Should we move it to a different commit/patch?
I don't see much point in separating it out into a different patch. That makes more work for everyone and it's a simply, obvious fix to reporting code this patch is adding to...
Quoted 4 lines
Not related to this change - Can we change the above statement to something like it is done below for
_notruni.e
grep "Ran $basedir/*/log" | uniq | sed ... | wc - l
in this way we can get rid of the sort command?
Don't think so - uniq only filters adjacent lines. You have to sort the input to put all the duplicates together in the stream before uniq will detect and remove them. I suspect that it is possible to use 'sort -u' rather than 'sort | uniq' but I've always used the latter because not all implementations of sort support the '-u' option.
-Dave.
Signature · 2 lines
Dave Chinner [email protected]
Re: [PATCH 04/28] check-parallel: report how many tests were _notrun
Quoted 26 lines · Dave Chinner
Quoted 25 lines · Nirjhar Roy (IBM)
Quoted 7 lines · Dave Chinner
From: Dave Chinner <[email protected]>
We already count tests "run" and tests that failed, but there is no indication of how many tests were
_notrun. With tests actually failing to run because of things like bugs in_scratch_mkfs_sizedbeing reported as "_notrun" instead of failing, visibility of changing numbers of_notruntests is needed.This looks okay to me. 2 questions below. Apart from that, this looks good to me.
Reviewed-by Nirjhar Roy (IBM) <[email protected]>Quoted 12 lines
Signed-off-by Dave Chinner <[email protected]>--- check-parallel | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)check-parallel +0
-0@@ -188,7 +188,10 @@ done; wait
Quoted 8 lines
Quoted 5 lines
check-parallel done; · line 191
Quoted 3 lines
check-parallel +1
-1echo -n "Tests run: " -grep Ran /mnt/xfs/*/log | sed -e 's,^.*:,,' -e 's, ,\n,g' | sort | uniq | wc -l +grep Ran $basedir/*/log | sed -e 's,^.*:,,' -e 's, ,\n,g' | sort | uniq | wc -lNit:
/mnt/xfs--> $basedir is not related to this change, right? Should we move it to a different commit/patch?I don't see much point in separating it out into a different patch. That makes more work for everyone and it's a simply, obvious fix to reporting code this patch is adding to...
Okay.
Quoted 11 lines
Quoted 5 lines
Not related to this change - Can we change the above statement to something like it is done below for
_notruni.e
grep "Ran $basedir/*/log" | uniq | sed ... | wc - l
in this way we can get rid of the sort command?Don't think so - uniq only filters adjacent lines. You have to sort the input to put all the duplicates together in the stream before uniq will detect and remove them. I suspect that it is possible to use 'sort -u' rather than 'sort | uniq' but I've always used the latter because not all implementations of sort support the '-u' option.
Yes, uniq considers adjacent lines. So for the getting number of _notruns we are using:
"grep "^Not run" $basedir/*/log | uniq | sed -e 's,^.*:,,' -e 's, ,\n,g' -e 's,^\n,,' | wc", why aren't we using sort? After looking at this command, I suggested above change.
--NR
Quoted 3 lines
-Dave.
Signature · 3 lines
Nirjhar Roy Linux Kernel Developer IBM, Bangalore
PATCH 5/28 check: factor out test list building code
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
Factor out all the test list parsing and building code to common/test_list so that it can be used by both check and check-parallel.
This also namespaces all the test list code to use _tl_ prefixes, and adds wrappers to set up test list parsing parameters.
Note: there is still some future work to convert externally visible parameters like SRC_DIR to the new namespace.
--- check | 270 +++---------------------------------------- common/report | 2 +- common/test_list | 295 +++++++++++++++++++++++++++++++++++++++++++++++ 3 files changed, 314 insertions(+), 253 deletions(-) create mode 100644 common/test_list
check +18 -252
@@ -15,19 +15,13 @@ notrun=() interrupt=true diff="diff -u" showme=false -have_test_arg=false -randomize=false -exact_order=false export here=`pwd` -xfile="" -subdir_xfile="" brief_test_summary=false do_report=false DUMP_OUTPUT=false iterations=1 istop=false loop_on_fail=0 -exclude_tests=() # This is a global variable used to pass test failure text to reporting gunk _err_msg=""
@@ -49,8 +43,9 @@ timestamp=${TIMESTAMP:=false} rm -f $tmp.list $tmp.tmp $tmp.grep $here/$iam.out $tmp.report.* $tmp.arglist -SRC_GROUPS="generic" -export SRC_DIR="tests" +# We need to include the test list processing first as argument parsing +# requires test list parsing and setup. +. ./common/test_list . ./common/exit
@@ -126,153 +121,12 @@ examples: exit 1 } -get_sub_group_list() -{ - local d=$1 - local grp=$2 - - test -s "$SRC_DIR/$d/group.list" || return 1 - - local grpl=$(sed -n < $SRC_DIR/$d/group.list \ - -e 's/#.*//' \ - -e 's/$/ /' \ - -e "s;^\($VALID_TEST_NAME\).* $grp .*;$SRC_DIR/$d/\1;p") - echo $grpl -} - -get_group_list() -{ - local grp=$1 - local grpl="" - local sub=$(dirname $grp) - local fsgroup="$FSTYP" - - if [ -n "$sub" -a "$sub" != "." -a -d "$SRC_DIR/$sub" ]; then - # group is given as <subdir>/<group> (e.g. xfs/quick) - grp=$(basename $grp) - get_sub_group_list $sub $grp - return - fi - - if [ "$FSTYP" = ext2 -o "$FSTYP" = ext3 ]; then - fsgroup=ext4 - fi - for d in $SRC_GROUPS $fsgroup; do - if ! test -d "$SRC_DIR/$d" ; then - continue - fi - grpl="$grpl $(get_sub_group_list $d $grp)" - done - echo $grpl -} - -# Find all tests, excluding files that are test metadata such as group files. -# It matches test names against $VALID_TEST_NAME defined in common/rc -get_all_tests() -{ - touch $tmp.list - for d in $SRC_GROUPS $FSTYP; do - if ! test -d "$SRC_DIR/$d" ; then - continue - fi - ls $SRC_DIR/$d/* | \ - grep -v "\..*" | \ - grep "^$SRC_DIR/$d/$VALID_TEST_NAME"| \ - grep -v "group\|Makefile" >> $tmp.list 2>/dev/null - done -} - -# takes the list of tests to run in $tmp.list, and removes the tests passed to -# the function from that list. -trim_test_list() -{ - local test_list="$*" - - rm -f $tmp.grep - local numsed=0 - for t in $test_list - do - if [ $numsed -gt 100 ]; then - grep -v -f $tmp.grep <$tmp.list >$tmp.tmp - mv $tmp.tmp $tmp.list - numsed=0 - rm -f $tmp.grep - fi - echo "^$t\$" >>$tmp.grep - numsed=`expr $numsed + 1` - done - grep -v -f $tmp.grep <$tmp.list >$tmp.tmp - mv $tmp.tmp $tmp.list - rm -f $tmp.grep -} - _timestamp() { local now=`date "+%T"` echo -n " [$now]" } -_prepare_test_list() -{ - unset list - # Tests specified on the command line - if [ -s $tmp.arglist ]; then - cat $tmp.arglist > $tmp.list - else - touch $tmp.list - fi - - # Specified groups to include - # Note that the CLI processing adds a leading space to the first group - # parameter, so we have to catch that here checking for "all" - if ! $have_test_arg && [ "$GROUP_LIST" == " all" ]; then - # no test numbers, do everything - get_all_tests - else - for group in $GROUP_LIST; do - list=$(get_group_list $group) - if [ -z "$list" ]; then - echo "Group \"$group\" is empty or not defined?" - exit 1 - fi - - for t in $list; do - grep -s "^$t\$" $tmp.list >/dev/null || \ - echo "$t" >>$tmp.list - done - done - fi - - # Specified groups to exclude - for xgroup in $XGROUP_LIST; do - list=$(get_group_list $xgroup) - if [ -z "$list" ]; then - echo "Group \"$xgroup\" is empty or not defined?" - continue - fi - - trim_test_list $list - done - - # sort the list of tests into numeric order unless we're running tests - # in the exact order specified - if ! $exact_order; then - if $randomize; then - if type shuf >& /dev/null; then - sorter="shuf" - else - sorter="awk -v seed=$RANDOM -f randomize.awk" - fi - else - sorter="cat" - fi - list=`sort -n $tmp.list | uniq | $sorter` - else - list=`cat $tmp.list` - fi - rm -f $tmp.list -} - # Process command arguments first. while [ $# -gt 0 ]; do case "$1" in
@@ -287,48 +141,20 @@ while [ $# -gt 0 ]; do export OVERLAY=true ;; - -g) group=$2 ; shift ; - GROUP_LIST="$GROUP_LIST ${group//,/ }" - ;; + -g) _tl_setup_group $2 ; shift ;; + -e) _tl_setup_exclude_tests $2 ; shift ;; + -E) _tl_setup_exclude_file $2 ; shift ;; + -x) _tl_setup_exclude_group $2; shift ;; + -X) _tl_setup_exclude_subdir $2; shift ;; + -r) _tl_setup_randomise ;; + --exact-order) _tl_setup_ordered ;; - -x) xgroup=$2 ; shift ; - XGROUP_LIST="$XGROUP_LIST ${xgroup//,/ }" - ;; - - -X) subdir_xfile=$2; shift ; - ;; - -e) - xfile=$2; shift ; - readarray -t -O "${#exclude_tests[@]}" exclude_tests < \ - <(echo "$xfile" | tr ', ' '\n\n') - ;; - - -E) xfile=$2; shift ; - if [ -f $xfile ]; then - readarray -t -O ${#exclude_tests[@]} exclude_tests < \ - <(sed "s/#.*$//" $xfile) - fi - ;; -s) RUN_SECTION="$RUN_SECTION $2"; shift ;; -S) EXCLUDE_SECTION="$EXCLUDE_SECTION $2"; shift ;; -l) diff="diff" ;; -udiff) diff="$diff -u" ;; -n) showme=true ;; - -r) - if $exact_order; then - echo "Cannot specify -r and --exact-order." - exit 1 - fi - randomize=true - ;; - --exact-order) - if $randomize; then - echo "Cannnot specify --exact-order and -r." - exit 1 - fi - exact_order=true - ;; -i) iterations=$2; shift ;; -I) iterations=$2; istop=true; shift ;; -T) timestamp=true ;;
@@ -346,13 +172,13 @@ while [ $# -gt 0 ]; do -*) usage ;; *) # not an argument, we've got tests now. - have_test_arg=true ;; + _tl_setup_cli $* esac # if we've found a test specification, the break out of the processing # loop before we shift the arguments so that this is the first argument # that we process in the test arg loop below. - if $have_test_arg; then + if $_tl_have_test_args; then break; fi
@@ -392,51 +218,6 @@ if [ -n "$FUZZ_REWRITE_DURATION" ]; then fi fi -if [ -n "$subdir_xfile" ]; then - for d in $SRC_GROUPS $FSTYP; do - [ -f $SRC_DIR/$d/$subdir_xfile ] || continue - for f in `sed "s/#.*$//" $SRC_DIR/$d/$subdir_xfile`; do - exclude_tests+=($d/$f) - done - done -fi - -# Process tests from command line now. -if $have_test_arg; then - while [ $# -gt 0 ]; do - case "$1" in - -*) echo "Arguments before tests, please!" - status=1 - exit $status - ;; - *) # Expand test pattern (e.g. xfs/???, *fs/001) - list=$(cd $SRC_DIR; echo $1) - for t in $list; do - t=${t#$SRC_DIR/} - test_dir=${t%%/*} - test_name=${t##*/} - group_file=$SRC_DIR/$test_dir/group.list - - if grep -Eq "^$test_name" $group_file; then - # in group file ... OK - echo $SRC_DIR/$test_dir/$test_name \ - >>$tmp.arglist - else - # oops - echo "$t - unknown test, ignored" - fi - done - ;; - esac - - shift - done -elif [ -z "$GROUP_LIST" ]; then - # default group list is the auto group. If any other group or test is - # specified, we use that instead. - GROUP_LIST="auto" -fi - if [ `id -u` -ne 0 ] then echo "check: QA must be run as root"
@@ -597,21 +378,6 @@ _check_filesystems() return $ret } -_expunge_test() -{ - local TEST_ID="$1" - - for f in "${exclude_tests[@]}"; do - # $f may contain traling spaces and comments - local id_regex="^${TEST_ID}\b" - if [[ "$f" =~ ${id_regex} ]]; then - echo " [expunged]" - return 0 - fi - done - return 1 -} - # retain files which would be overwritten in subsequent reruns of the same test _stash_fail_loop_files() { local seq_prefix="${REPORT_DIR}/${1}"
@@ -719,7 +485,7 @@ _run_seq() { } _detect_kmemleak -_prepare_test_list +_tl_prepare_test_list fstests_start_time="$(date +"%F %T")" if $OPTIONS_HAVE_SECTIONS; then
@@ -793,7 +559,7 @@ function run_section() # common/rc again with correct FSTYP to get FSTYP specific configs, # e.g. common/xfs . common/rc - _prepare_test_list + _tl_prepare_test_list elif [ "$OLD_TEST_FS_MOUNT_OPTS" != "$TEST_FS_MOUNT_OPTS" ]; then # Unmount TEST_DEV to apply the updated mount options. # It will be mounted again by init_rc(), called shortly after.
@@ -854,13 +620,13 @@ function run_section() loop_status=() # track rerun-on-failure state local tc_status ix - local -a _list=( $list ) + local -a _list=( $_tl_tests ) for ((ix = 0; ix < ${#_list[*]}; !${#loop_status[*]} && ix++)); do seq="${_list[$ix]}" # the filename for the test and the name output are different. # we don't include the tests/ directory in the name output. - export seqnum=${seq#$SRC_DIR/} + export seqnum=$(_tl_strip_src_dir $seq) group=${seqnum%%/*} if $OPTIONS_HAVE_SECTIONS; then REPORT_DIR="$RESULT_BASE/$section"
@@ -882,7 +648,7 @@ function run_section() echo -n "$seqnum" if $showme; then - if _expunge_test $seqnum; then + if _tl_expunge_test $seqnum; then tc_status="expunge" else echo
@@ -908,7 +674,7 @@ function run_section() rm -f $seqres.out.bad $seqres.hints # check if we really should run it - if _expunge_test $seqnum; then + if _tl_expunge_test $seqnum; then tc_status="expunge" _stash_test_status "$seqnum" "$tc_status" continue
common/report +1 -1
@@ -196,7 +196,7 @@ _xunit_make_testcase_report() echo -e "\t\t<skipped/>" >> $report ;; "fail") - local out_src="${SRC_DIR}/${test_name}.out" + local out_src="${_tl_src_dir}/${test_name}.out" local full_file="${REPORT_DIR}/${test_name}.full" local dmesg_file="${REPORT_DIR}/${test_name}.dmesg" local outbad_file="${REPORT_DIR}/${test_name}.out.bad"
common/test_list new file +295 -0
295 lines not on this page · raw message
Signature · 1 line
2.45.2
Re: [PATCH 05/28] check: factor out test list building code
Quoted 11 lines · Dave Chinner
From: Dave Chinner <[email protected]>
Factor out all the test list parsing and building code to common/test_list so that it can be used by both check and check-parallel.
This also namespaces all the test list code to use _tl_ prefixes, and adds wrappers to set up test list parsing parameters.
Note: there is still some future work to convert externally visible parameters like
SRC_DIRto the new namespace.
I have gone through this patch. I did some basic testing too with ./check and I havent' found any obvious issues. There are some minor feedback that I have given below:
Quoted 416 lines
Signed-off-by Dave Chinner <[email protected]>--- check | 270 +++---------------------------------------- common/report | 2 +- common/test_list | 295 +++++++++++++++++++++++++++++++++++++++++++++++ 3 files changed, 314 insertions(+), 253 deletions(-) create mode 100644 common/test_listcheck +18
-252@@ -15,19 +15,13 @@ notrun=() interrupt=true diff="diff -u" showme=false -have_test_arg=false -randomize=false -exact_order=false export here=`pwd` -xfile="" -subdir_xfile="" brief_test_summary=false do_report=false DUMP_OUTPUT=false iterations=1 istop=false loop_on_fail=0 -exclude_tests=() # This is a global variable used to pass test failure text to reporting gunk _err_msg=""@@ -49,8 +43,9 @@ timestamp=${TIMESTAMP:=false} rm -f $tmp.list $tmp.tmp $tmp.grep $here/$iam.out $tmp.report.* $tmp.arglist -SRC_GROUPS="generic" -export SRC_DIR="tests" +# We need to include the test list processing first as argument parsing +# requires test list parsing and setup. +. ./common/test_list . ./common/exit@@ -126,153 +121,12 @@ examples: exit 1 } -get_sub_group_list() -{ - local d=$1 - local grp=$2 - - test -s "$SRC_DIR/$d/group.list" || return 1 - - local grpl=$(sed -n < $SRC_DIR/$d/group.list \ - -e 's/#.*//' \ - -e 's/$/ /' \ - -e "s;^\($VALID_TEST_NAME\).* $grp .*;$SRC_DIR/$d/\1;p") - echo $grpl -} - -get_group_list() -{ - local grp=$1 - local grpl="" - local sub=$(dirname $grp) - local fsgroup="$FSTYP" - - if [ -n "$sub" -a "$sub" != "." -a -d "$SRC_DIR/$sub" ]; then - # group is given as <subdir>/<group> (e.g. xfs/quick) - grp=$(basename $grp) - get_sub_group_list $sub $grp - return - fi - - if [ "$FSTYP" = ext2 -o "$FSTYP" = ext3 ]; then - fsgroup=ext4 - fi - for d in $SRC_GROUPS $fsgroup; do - if ! test -d "$SRC_DIR/$d" ; then - continue - fi - grpl="$grpl $(get_sub_group_list $d $grp)" - done - echo $grpl -} - -# Find all tests, excluding files that are test metadata such as group files. -# It matches test names against $VALID_TEST_NAME defined in common/rc -get_all_tests() -{ - touch $tmp.list - for d in $SRC_GROUPS $FSTYP; do - if ! test -d "$SRC_DIR/$d" ; then - continue - fi - ls $SRC_DIR/$d/* | \ - grep -v "\..*" | \ - grep "^$SRC_DIR/$d/$VALID_TEST_NAME"| \ - grep -v "group\|Makefile" >> $tmp.list 2>/dev/null - done -} - -# takes the list of tests to run in $tmp.list, and removes the tests passed to -# the function from that list. -trim_test_list() -{ - local test_list="$*" - - rm -f $tmp.grep - local numsed=0 - for t in $test_list - do - if [ $numsed -gt 100 ]; then - grep -v -f $tmp.grep <$tmp.list >$tmp.tmp - mv $tmp.tmp $tmp.list - numsed=0 - rm -f $tmp.grep - fi - echo "^$t\$" >>$tmp.grep - numsed=`expr $numsed + 1` - done - grep -v -f $tmp.grep <$tmp.list >$tmp.tmp - mv $tmp.tmp $tmp.list - rm -f $tmp.grep -} - _timestamp() { local now=`date "+%T"` echo -n " [$now]" } -_prepare_test_list() -{ - unset list - # Tests specified on the command line - if [ -s $tmp.arglist ]; then - cat $tmp.arglist > $tmp.list - else - touch $tmp.list - fi - - # Specified groups to include - # Note that the CLI processing adds a leading space to the first group - # parameter, so we have to catch that here checking for "all" - if ! $have_test_arg && [ "$GROUP_LIST" == " all" ]; then - # no test numbers, do everything - get_all_tests - else - for group in $GROUP_LIST; do - list=$(get_group_list $group) - if [ -z "$list" ]; then - echo "Group \"$group\" is empty or not defined?" - exit 1 - fi - - for t in $list; do - grep -s "^$t\$" $tmp.list >/dev/null || \ - echo "$t" >>$tmp.list - done - done - fi - - # Specified groups to exclude - for xgroup in $XGROUP_LIST; do - list=$(get_group_list $xgroup) - if [ -z "$list" ]; then - echo "Group \"$xgroup\" is empty or not defined?" - continue - fi - - trim_test_list $list - done - - # sort the list of tests into numeric order unless we're running tests - # in the exact order specified - if ! $exact_order; then - if $randomize; then - if type shuf >& /dev/null; then - sorter="shuf" - else - sorter="awk -v seed=$RANDOM -f randomize.awk" - fi - else - sorter="cat" - fi - list=`sort -n $tmp.list | uniq | $sorter` - else - list=`cat $tmp.list` - fi - rm -f $tmp.list -} - # Process command arguments first. while [ $# -gt 0 ]; do case "$1" in@@ -287,48 +141,20 @@ while [ $# -gt 0 ]; do export OVERLAY=true ;; - -g) group=$2 ; shift ; - GROUP_LIST="$GROUP_LIST ${group//,/ }" - ;; + -g) _tl_setup_group $2 ; shift ;; + -e) _tl_setup_exclude_tests $2 ; shift ;; + -E) _tl_setup_exclude_file $2 ; shift ;; + -x) _tl_setup_exclude_group $2; shift ;; + -X) _tl_setup_exclude_subdir $2; shift ;; + -r) _tl_setup_randomise ;; + --exact-order) _tl_setup_ordered ;; - -x) xgroup=$2 ; shift ; - XGROUP_LIST="$XGROUP_LIST ${xgroup//,/ }" - ;; - - -X) subdir_xfile=$2; shift ; - ;; - -e) - xfile=$2; shift ; - readarray -t -O "${#exclude_tests[@]}" exclude_tests < \ - <(echo "$xfile" | tr ', ' '\n\n') - ;; - - -E) xfile=$2; shift ; - if [ -f $xfile ]; then - readarray -t -O ${#exclude_tests[@]} exclude_tests < \ - <(sed "s/#.*$//" $xfile) - fi - ;; -s) RUN_SECTION="$RUN_SECTION $2"; shift ;; -S) EXCLUDE_SECTION="$EXCLUDE_SECTION $2"; shift ;; -l) diff="diff" ;; -udiff) diff="$diff -u" ;; -n) showme=true ;; - -r) - if $exact_order; then - echo "Cannot specify -r and --exact-order." - exit 1 - fi - randomize=true - ;; - --exact-order) - if $randomize; then - echo "Cannnot specify --exact-order and -r." - exit 1 - fi - exact_order=true - ;; -i) iterations=$2; shift ;; -I) iterations=$2; istop=true; shift ;; -T) timestamp=true ;;@@ -346,13 +172,13 @@ while [ $# -gt 0 ]; do -*) usage ;; *) # not an argument, we've got tests now. - have_test_arg=true ;; + _tl_setup_cli $* esac # if we've found a test specification, the break out of the processing # loop before we shift the arguments so that this is the first argument # that we process in the test arg loop below. - if $have_test_arg; then + if $_tl_have_test_args; then break; fi@@ -392,51 +218,6 @@ if [ -n "$FUZZ_REWRITE_DURATION" ]; then fi fi -if [ -n "$subdir_xfile" ]; then - for d in $SRC_GROUPS $FSTYP; do - [ -f $SRC_DIR/$d/$subdir_xfile ] || continue - for f in `sed "s/#.*$//" $SRC_DIR/$d/$subdir_xfile`; do - exclude_tests+=($d/$f) - done - done -fi - -# Process tests from command line now. -if $have_test_arg; then - while [ $# -gt 0 ]; do - case "$1" in - -*) echo "Arguments before tests, please!" - status=1 - exit $status - ;; - *) # Expand test pattern (e.g. xfs/???, *fs/001) - list=$(cd $SRC_DIR; echo $1) - for t in $list; do - t=${t#$SRC_DIR/} - test_dir=${t%%/*} - test_name=${t##*/} - group_file=$SRC_DIR/$test_dir/group.list - - if grep -Eq "^$test_name" $group_file; then - # in group file ... OK - echo $SRC_DIR/$test_dir/$test_name \ - >>$tmp.arglist - else - # oops - echo "$t - unknown test, ignored" - fi - done - ;; - esac - - shift - done -elif [ -z "$GROUP_LIST" ]; then - # default group list is the auto group. If any other group or test is - # specified, we use that instead. - GROUP_LIST="auto" -fi - if [ `id -u` -ne 0 ] then echo "check: QA must be run as root"@@ -597,21 +378,6 @@ _check_filesystems() return $ret } -_expunge_test() -{ - local TEST_ID="$1" - - for f in "${exclude_tests[@]}"; do - # $f may contain traling spaces and comments - local id_regex="^${TEST_ID}\b" - if [[ "$f" =~ ${id_regex} ]]; then - echo " [expunged]" - return 0 - fi - done - return 1 -} - # retain files which would be overwritten in subsequent reruns of the same test _stash_fail_loop_files() { local seq_prefix="${REPORT_DIR}/${1}"@@ -719,7 +485,7 @@ _run_seq() { } _detect_kmemleak -_prepare_test_list +_tl_prepare_test_list fstests_start_time="$(date +"%F %T")" if $OPTIONS_HAVE_SECTIONS; then@@ -793,7 +559,7 @@ function run_section() # common/rc again with correct FSTYP to get FSTYP specific configs, # e.g. common/xfs . common/rc - _prepare_test_list + _tl_prepare_test_list elif [ "$OLD_TEST_FS_MOUNT_OPTS" != "$TEST_FS_MOUNT_OPTS" ]; then # Unmount TEST_DEV to apply the updated mount options. # It will be mounted again by init_rc(), called shortly after.@@ -854,13 +620,13 @@ function run_section() loop_status=() # track rerun-on-failure state local tc_status ix - local -a _list=( $list ) + local -a _list=( $_tl_tests ) for ((ix = 0; ix < ${#_list[*]}; !${#loop_status[*]} && ix++)); do seq="${_list[$ix]}" # the filename for the test and the name output are different. # we don't include the tests/ directory in the name output. - export seqnum=${seq#$SRC_DIR/} + export seqnum=$(_tl_strip_src_dir $seq) group=${seqnum%%/*} if $OPTIONS_HAVE_SECTIONS; then REPORT_DIR="$RESULT_BASE/$section"@@ -882,7 +648,7 @@ function run_section() echo -n "$seqnum" if $showme; then - if _expunge_test $seqnum; then + if _tl_expunge_test $seqnum; then tc_status="expunge" else echo@@ -908,7 +674,7 @@ function run_section() rm -f $seqres.out.bad $seqres.hints # check if we really should run it - if _expunge_test $seqnum; then + if _tl_expunge_test $seqnum; then tc_status="expunge" _stash_test_status "$seqnum" "$tc_status" continuecommon/report +1
-1@@ -196,7 +196,7 @@ _xunit_make_testcase_report() echo -e "\t\t<skipped/>" >> $report ;; "fail") - local out_src="${SRC_DIR}/${test_name}.out" + local out_src="${_tl_src_dir}/${test_name}.out" local full_file="${REPORT_DIR}/${test_name}.full" local dmesg_file="${REPORT_DIR}/${test_name}.dmesg" local outbad_file="${REPORT_DIR}/${test_name}.out.bad"common/test_list new file +0
-0
common/test_list line 8
Quoted 8 lines
common/test_list new file +8
-0@@ -0,0 +1,295 @@ +##/bin/bash +# SPDX-License-Identifier: GPL-2.0+ +# Copyright (c) 2000-2002,2006 Silicon Graphics, Inc. All Rights Reserved. +# Copyright (c) 2024 Red Hat, Inc. All Rights Reserved. +# +# Test list parsing and building functions +# +# Note: this file must stand alone and not be dependent on any other includes,
We can include common/exit, right? That file doesn't file only has a couple of exit related functions (with no executable code) and no further dependencies, correct?
common/test_list line 15
Quoted 7 lines
common/test_list new file +7
-0+# most especially common/rc and common/config. This is because we have to +# include this file before option parsing, whilst the rc/config includes need to +# be included -after- option parsing. +# +# Any function or variable that is public should have a "_tl_" prefix. + +export _tl_src_dir="tests"
Minor: There was a change[1] which converted some of global variables in small case to upper case. Should we follow the same convention here, i.e, _TL_SRC_DIR or maybe _tl_SRC_DIR?
[1]
https://web.git.kernel.org/pub/scm/fs/xfs/xfstests-dev.git/commit/?h=for-next&id=ab459c67c5e0347ab4a5c28eadfe5ee4e3fd2f01
common/test_list line 25
Quoted 10 lines
common/test_list new file +10
-0+ +_SRC_GROUPS="generic" +_GROUP_LIST= +_XGROUP_LIST= +_tl_exact_order=false +_tl_randomise=false +_tl_have_test_args=false +_tl_file="$tmp.test_list" +_tl_exclude_tests=() +_tl_tests=
Similar comments for all the above lower cased global variables?
Quoted 100 lines
common/test_list new file +100
-0+ +_tl_strip_src_dir() +{ + local test="$1" + + echo ${test#$_tl_src_dir/} +} + +get_sub_group_list() +{ + local d=$1 + local grp=$2 + + test -s "$_tl_src_dir/$d/group.list" || return 1 + + local grpl=$(sed -n < $_tl_src_dir/$d/group.list \ + -e 's/#.*//' \ + -e 's/$/ /' \ + -e "s;^\($VALID_TEST_NAME\).* $grp .*;$_tl_src_dir/$d/\1;p") + echo $grpl +} + +get_group_list() +{ + local grp=$1 + local grpl="" + local sub=$(dirname $grp) + local fsgroup="$FSTYP" + + if [ -n "$sub" -a "$sub" != "." -a -d "$_tl_src_dir/$sub" ]; then + # group is given as <subdir>/<group> (e.g. xfs/quick) + grp=$(basename $grp) + get_sub_group_list $sub $grp + return + fi + + if [ "$FSTYP" = ext2 -o "$FSTYP" = ext3 ]; then + fsgroup=ext4 + fi + for d in $_SRC_GROUPS $fsgroup; do + if ! test -d "$_tl_src_dir/$d" ; then + continue + fi + grpl="$grpl $(get_sub_group_list $d $grp)" + done + echo $grpl +} + +# Find all tests, excluding files that are test metadata such as group files. +# It matches test names against $VALID_TEST_NAME defined in common/rc +get_all_tests() +{ + touch $tmp.list + for d in $_SRC_GROUPS $FSTYP; do + if ! test -d "$_tl_src_dir/$d" ; then + continue + fi + ls $_tl_src_dir/$d/* | \ + grep -v "\..*" | \ + grep "^$_tl_src_dir/$d/$VALID_TEST_NAME"| \ + grep -v "group\|Makefile" >> $tmp.list 2>/dev/null + done +} + +# takes the list of tests to run in $tmp.list, and removes the tests passed to +# the function from that list. +trim_test_list() +{ + local test_list="$*" + + rm -f $tmp.grep + local numsed=0 + for t in $test_list + do + if [ $numsed -gt 100 ]; then + grep -v -f $tmp.grep <$tmp.list >$tmp.tmp + mv $tmp.tmp $tmp.list + numsed=0 + rm -f $tmp.grep + fi + echo "^$t\$" >>$tmp.grep + numsed=`expr $numsed + 1` + done + grep -v -f $tmp.grep <$tmp.list >$tmp.tmp + mv $tmp.tmp $tmp.list + rm -f $tmp.grep +} + +_tl_prepare_test_list() +{ + unset _tl_tests + # Tests specified on the command line + if [ -s $_tl_file ]; then + cat $_tl_file > $tmp.list + else + touch $tmp.list + fi + + # Specified groups to include + # Note that the CLI processing adds a leading space to the first group
common/test_list line 135
Quoted 10 lines
common/test_list new file +10
-0+ # parameter, so we have to catch that here checking for "all" + if ! $_tl_have_test_args && [ "$_GROUP_LIST" == " all" ]; then + # no test numbers, do everything + get_all_tests + else + for group in $_GROUP_LIST; do + list=$(get_group_list $group) + if [ -z "$list" ]; then + echo "Group \"$group\" is empty or not defined?" + exit 1
_fatal "Group \"$group\" is empty or not defined?" ?
Quoted 88 lines
common/test_list new file +88
-0+ fi + + for t in $list; do + grep -s "^$t\$" $tmp.list >/dev/null || \ + echo "$t" >>$tmp.list + done + done + fi + + # Specified groups to exclude + for xgroup in $_XGROUP_LIST; do + list=$(get_group_list $xgroup) + if [ -z "$list" ]; then + echo "Group \"$xgroup\" is empty or not defined?" + continue + fi + + trim_test_list $list + done + + # sort the list of tests into numeric order unless we're running tests + # in the exact order specified + if ! $_tl_exact_order; then + if $_tl_randomise; then + if type shuf >& /dev/null; then + sorter="shuf" + else + sorter="awk -v seed=$RANDOM -f randomize.awk" + fi + else + sorter="cat" + fi + _tl_tests=`sort -n $tmp.list | uniq | $sorter` + else + _tl_tests=`cat $tmp.list` + fi + rm -f $tmp.list +} + +_tl_expunge_test() +{ + local TEST_ID="$1" + + for f in "${_tl_exclude_tests[@]}"; do + # $f may contain traling spaces and comments + local id_regex="^${TEST_ID}\b" + if [[ "$f" =~ ${id_regex} ]]; then + echo " [expunged]" + return 0 + fi + done + return 1 +} + +_tl_setup_exclude_tests() +{ + local list="$1" + + readarray -t -O "${#_tl_exclude_tests[@]}" _tl_exclude_tests < \ + <(echo "$list" | tr ', ' '\n\n') +} + +_tl_setup_exclude_file() +{ + local xfile="$1" + + if [ -f $xfile ]; then + readarray -t -O ${#_tl_exclude_tests[@]} _tl_exclude_tests < \ + <(sed "s/#.*$//" $xfile) + fi +} + +_tl_setup_exclude_subdir() +{ + local xfile="$1" + local d + local f + + [ -z "$xfile" ] && return + + for d in $_SRC_GROUPS $FSTYP; do + [ -f $_tl_src_dir/$d/$xfile ] || continue + for f in `sed "s/#.*$//" $_tl_src_dir/$d/$xfile`; do + _tl_exclude_tests+=($d/$f) + done + done +} +
common/test_list line 233
Quoted 10 lines
common/test_list new file +10
-0+_tl_setup_exclude_group() +{ + local xgroup="$1" + + _XGROUP_LIST="$_XGROUP_LIST ${xgroup//,/ }" +} + +_tl_setup_group() +{ + local group="$1"
Minor: Since this function is public , do you think it is helpful to have a quick sanitization that makes sure that $1 is not empty and prints message otherwise? Similar comment for _tl_setup_exclude_group().
common/test_list line 242
Quoted 9 lines
common/test_list new file +9
-0+ + _GROUP_LIST="$_GROUP_LIST ${group//,/ }" +} + +_tl_setup_randomise() +{ + if $_tl_exact_order; then + echo "Cannot specify -r and --exact-order." + exit 1
Use _fatal?
common/test_list line 250
Quoted 8 lines
common/test_list new file +8
-0+ fi + _tl_randomise=true +} + +_tl_setup_ordered() +{ + if $_tl_randomise; then + echo "Cannnot specify --exact-order and -r."
Minor/typo: "Cannnot" -> "Cannot"
common/test_list line 251
Quoted 1 line
common/test_list new file +1
-0+ exit 1
use _fatal?
common/test_list line 256
Quoted 5 lines
common/test_list new file +5
-0+ fi + _tl_exact_order=true +} + +_tl_setup_cli()
Nit: This function mostly sets up test list related variables - so maybe a name that suggests that. Maybe something like _tl_setup_test_list_vars or _tl_setup_test_vars - just a suggestion, no hard preferences here.
common/test_list line 262
Quoted 6 lines
common/test_list new file +6
-0+{ + while [ $# -gt 0 ]; do + case "$1" in + -*) echo "Arguments before tests, please!" + status=1 + exit $status
use _fatal?
Quoted 9 lines
common/test_list new file +9
-0+ ;; + *) # Expand test pattern (e.g. xfs/???, *fs/001) + local list=$(cd $_tl_src_dir; echo $1) + local t + + for t in $list; do + t=${t#$_tl_src_dir/} + local test_dir=${t%%/*} + local test_name=${t##*/}
common/test_list line 281
Quoted 10 lines
common/test_list new file +10
-0+ local group_file=$_tl_src_dir/$test_dir/group.list + + if grep -Eq "^$test_name" $group_file; then + # in group file ... OK + echo $_tl_src_dir/$test_dir/$test_name \ + >> $_tl_file + _tl_have_test_args=true + else + # oops + echo "$t - unknown test, ignored"
Minor: Sometimes it has happened with me that I have added a new test (with ./new ) but forgot to run make and hence the test wasn't getting run and was reported to be an invalid test. Should we add a suggestion to run make, in case we encounter an unidentified test name? --NR
Quoted 14 lines
common/test_list new file +14
-0+ fi + done + ;; + esac + + shift + done + + if ! $_tl_have_test_args && [ -z "$_GROUP_LIST" ]; then + # default group list is the auto group. If any other group or + # test is specified, we use that instead. + _GROUP_LIST="auto" + fi +}
Re: [PATCH 05/28] check: factor out test list building code
Quoted 7 lines · Nirjhar Roy (IBM)
Quoted 6 lines · Dave Chinner
From: Dave Chinner <[email protected]>
Factor out all the test list parsing and building code to common/test_list so that it can be used by both check and check-parallel.
Quoted 8 lines
Quoted 5 lines
This also namespaces all the test list code to use _tl_ prefixes, and adds wrappers to set up test list parsing parameters.
Note: there is still some future work to convert externally visible parameters like
SRC_DIRto the new namespace.I have gone through this patch. I did some basic testing too with ./check and I havent' found any obvious issues. There are some minor feedback that I have given below:
Can you trim the stuff you aren't commenting on out? It makes it much easier to find what you are commenting on....
....
Quoted 9 lines
Quoted 9 lines
common/test_list new file +3
-0@@ -0,0 +1,295 @@ +##/bin/bash +# SPDX-License-Identifier: GPL-2.0+ +# Copyright (c) 2000-2002,2006 Silicon Graphics, Inc. All Rights Reserved.
Quoted 8 lines
common/test_list line 8
Quoted 5 lines
common/test_list new file +5
-0+# Copyright (c) 2024 Red Hat, Inc. All Rights Reserved. +# +# Test list parsing and building functions +# +# Note: this file must stand alone and not be dependent on any other includes,We can include common/exit, right? That file doesn't file only has a couple of exit related functions (with no executable code) and no further dependencies, correct?
I wrote that before common/exit was really a thing.
As it is, I don't think we should be sourcing files from sourced files. it hides all the dependencies, and results in the same file being sourced multiple times in the same process context.
i.e. if the top level is doing:
. common/exit
.....
. common/test_list
.....
Then we don't need to source common/exit from common/test_list because it already has been sourced.
That's what the comment is trying to say, and when I wrote it I was specifically thinking about the mess that common/rc and common/config was causing.
Quoted 10 lines
common/test_list line 15
Quoted 7 lines
common/test_list new file +7
-0+# most especially common/rc and common/config. This is because we have to +# include this file before option parsing, whilst the rc/config includes need to +# be included -after- option parsing. +# +# Any function or variable that is public should have a "_tl_" prefix. + +export _tl_src_dir="tests"Minor: There was a change[1] which converted some of global variables in small case to upper case. Should we follow the same convention here, i.e,
_TL_SRC_DIRor maybe_tl_SRC_DIR?
I don't think so, I very much dislike the semi-random capitalisation of namespace-less variables across fstests. I find coe full of upper case variables much more difficult to read and follow than properly namespaced lower case variables.
That's the other thing that is important - lots of the global scope variables sourced from common files are namespaceless and it makes it easy easy for tests to accidentally step on such variables. Once the varaibles are namespaced, there is no need for capitalisation to indicate that it might be a global variable - the namespace prefix tells you that as well as where it comes from.
Quoted 2 lines
Yeah, that was a case of choosing your battles.
It wasn't a bug, but it was buried in a long series of bug fixes and at the time people were shouting and being really obnoxious about the check-parallel changes. There was no point in dying on an unimportant hill by objecting to that change.
I still do think it was the wrong thing to do because it makes that code be inconsistent with every other local file loop device variable. Those devices are not used outside of common/metadump - it is internal implemenation variable that the code was written specifically to be hidden from callers wanting to manipulate and test metadumps...
Quoted 10 lines
common/test_list line 25
Quoted 9 lines
common/test_list new file +9
-0+_SRC_GROUPS="generic" +_GROUP_LIST= +_XGROUP_LIST= +_tl_exact_order=false +_tl_randomise=false +_tl_have_test_args=false +_tl_file="$tmp.test_list" +_tl_exclude_tests=() +_tl_tests=Similar comments for all the above lower cased global variables?
I didn't changed the GROUP names because they are internal to the test list implementation. I was largely moving the code and didn't want to change anything that was internal that I didn't need to touch. I'll have a look at cleaning that up.
.....
Quoted 5 lines
Quoted 5 lines
common/test_list new file +5
-0+ # Specified groups to include + # Note that the CLI processing adds a leading space to the first group + # parameter, so we have to catch that here checking for "all" + if ! $_tl_have_test_args && [ "$_GROUP_LIST" == " all" ]; then + # no test numbers, do everything
Quoted 8 lines
common/test_list line 135
Quoted 7 lines
common/test_list new file +7
-0+ get_all_tests + else + for group in $_GROUP_LIST; do + list=$(get_group_list $group) + if [ -z "$list" ]; then + echo "Group \"$group\" is empty or not defined?" + exit 1
_fatal"Group \"$group\" is empty or not defined?" ?
Remember that this was posted before those changes were made. This is the sort of thing that gets fixed on rebase...
Quoted 15 lines
common/test_list line 233
Quoted 10 lines
common/test_list new file +10
-0+_tl_setup_exclude_group() +{ + local xgroup="$1" + + _XGROUP_LIST="$_XGROUP_LIST ${xgroup//,/ }" +} + +_tl_setup_group() +{ + local group="$1"Minor: Since this function is public , do you think it is helpful to have a quick sanitization that makes sure that $1 is not empty and prints message otherwise? Similar comment for
_tl_setup_exclude_group().
Quoted 3 lines
common/test_list line 236
Quoted 3 lines
common/test_list new file +3
-0+ + _GROUP_LIST="$_GROUP_LIST ${group//,/ }" +}
I don't think it needs it. It should be checked by the caller if necessary because it is the one doing option parsing to set up the group.
Quoted 6 lines
common/test_list line 51
Quoted 2 lines
common/test_list new file +2
-0+ +_tl_setup_cli()Nit: This function mostly sets up test list related variables - so maybe a name that suggests that. Maybe something like
_tl_setup_test_list_varsor_tl_setup_test_vars- just a suggestion, no hard preferences here.
Ok, I'll come up with a new name for it, but keep in mind that this function is actually parsing arguments that have come directly from the CLI.
Quoted 22 lines
common/test_list line 262
Quoted 6 lines
common/test_list new file +6
-0+{ + while [ $# -gt 0 ]; do + case "$1" in + -*) echo "Arguments before tests, please!" + status=1 + exit $statususe
_fatal?Quoted 15 lines
common/test_list new file +15
-0+ ;; + *) # Expand test pattern (e.g. xfs/???, *fs/001) + local list=$(cd $_tl_src_dir; echo $1) + local t + + for t in $list; do + t=${t#$_tl_src_dir/} + local test_dir=${t%%/*} + local test_name=${t##*/} + local group_file=$_tl_src_dir/$test_dir/group.list + + if grep -Eq "^$test_name" $group_file; then + # in group file ... OK + echo $_tl_src_dir/$test_dir/$test_name \ + >> $_tl_file
Quoted 8 lines
common/test_list line 281
Quoted 4 lines
common/test_list new file +4
-0+ _tl_have_test_args=true + else + # oops + echo "$t - unknown test, ignored"Minor: Sometimes it has happened with me that I have added a new test (with ./new ) but forgot to run make and hence the test wasn't getting run and was reported to be an invalid test. Should we add a suggestion to run make, in case we encounter an unidentified test name?
Not in this patch set. If you want these sorts of process reminders, please send them as separate feature additions.
-Dave.
Signature · 2 lines
Dave Chinner [email protected]
Re: [PATCH 05/28] check: factor out test list building code
Quoted 11 lines · Dave Chinner
Quoted 10 lines · Nirjhar Roy (IBM)
Quoted 9 lines · Dave Chinner
From: Dave Chinner <[email protected]>
Factor out all the test list parsing and building code to common/test_list so that it can be used by both check and check-parallel.
This also namespaces all the test list code to use _tl_ prefixes, and adds wrappers to set up test list parsing parameters.
Quoted 7 lines
Quoted 5 lines
Quoted 2 lines
Note: there is still some future work to convert externally visible parameters like
SRC_DIRto the new namespace.I have gone through this patch. I did some basic testing too with ./check and I havent' found any obvious issues. There are some minor feedback that I have given below:
Can you trim the stuff you aren't commenting on out? It makes it much easier to find what you are commenting on....
Yeah, sorry. I think there was some issue with the editor of my email client.
Quoted 19 lines
....
Quoted 9 lines
Quoted 9 lines
common/test_list new file +3
-0@@ -0,0 +1,295 @@ +##/bin/bash +# SPDX-License-Identifier: GPL-2.0+ +# Copyright (c) 2000-2002,2006 Silicon Graphics, Inc. All Rights Reserved.Quoted 8 lines
common/test_list line 8
Quoted 5 lines
common/test_list new file +5
-0+# Copyright (c) 2024 Red Hat, Inc. All Rights Reserved. +# +# Test list parsing and building functions +# +# Note: this file must stand alone and not be dependent on any other includes,We can include common/exit, right? That file doesn't file only has a couple of exit related functions (with no executable code) and no further dependencies, correct?
Quoted 5 lines
I wrote that before common/exit was really a thing.
As it is, I don't think we should be sourcing files from sourced files. it hides all the dependencies, and results in the same file being sourced multiple times in the same process context.
Yeah, right.
Quoted 8 lines
i.e. if the top level is doing:
. common/exit
.....
. common/test_list
.....
Quoted 6 lines
Then we don't need to source common/exit from common/test_list because it already has been sourced.
That's what the comment is trying to say, and when I wrote it I was specifically thinking about the mess that common/rc and common/config was causing.
Okay.
Quoted 12 lines
Quoted 12 lines
common/test_list line 15
Quoted 9 lines
common/test_list new file +7
-0+# most especially common/rc and common/config. This is because we have to +# include this file before option parsing, whilst the rc/config includes need to +# be included -after- option parsing. +# +# Any function or variable that is public should have a "_tl_" prefix. + +export _tl_src_dir="tests"Minor: There was a change[1] which converted some of global variables in small case to upper case. Should we follow the same convention here, i.e,
_TL_SRC_DIRor maybe_tl_SRC_DIR?
Quoted 8 lines
I don't think so, I very much dislike the semi-random capitalisation of namespace-less variables across fstests. I find coe full of upper case variables much more difficult to read and follow than properly namespaced lower case variables.
That's the other thing that is important - lots of the global scope variables sourced from common files are namespaceless and it makes it easy easy for tests to accidentally step on such variables. Once
Yeah, accidental re-use and modification of global variable can happen. I agree.
Quoted 13 lines
the varaibles are namespaced, there is no need for capitalisation to indicate that it might be a global variable - the namespace prefix tells you that as well as where it comes from.
Quoted 2 lines
Yeah, that was a case of choosing your battles.
It wasn't a bug, but it was buried in a long series of bug fixes and at the time people were shouting and being really obnoxious about the check-parallel changes. There was no point in dying on an unimportant hill by objecting to that change.
Quoted 6 lines
I still do think it was the wrong thing to do because it makes that code be inconsistent with every other local file loop device variable. Those devices are not used outside of common/metadump - it is internal implemenation variable that the code was written specifically to be hidden from callers wanting to manipulate and test metadumps...
Okay, got it.
Quoted 11 lines
Quoted 11 lines
common/test_list line 25
Quoted 10 lines
common/test_list new file +9
-0+_SRC_GROUPS="generic" +_GROUP_LIST= +_XGROUP_LIST= +_tl_exact_order=false +_tl_randomise=false +_tl_have_test_args=false +_tl_file="$tmp.test_list" +_tl_exclude_tests=() +_tl_tests=Similar comments for all the above lower cased global variables?
Quoted 4 lines
I didn't changed the GROUP names because they are internal to the test list implementation. I was largely moving the code and didn't want to change anything that was internal that I didn't need to touch. I'll have a look at cleaning that up.
Okay.
Quoted 15 lines
.....
Quoted 5 lines
Quoted 5 lines
common/test_list new file +5
-0+ # Specified groups to include + # Note that the CLI processing adds a leading space to the first group + # parameter, so we have to catch that here checking for "all" + if ! $_tl_have_test_args && [ "$_GROUP_LIST" == " all" ]; then + # no test numbers, do everythingQuoted 8 lines
common/test_list line 135
Quoted 7 lines
common/test_list new file +7
-0+ get_all_tests + else + for group in $_GROUP_LIST; do + list=$(get_group_list $group) + if [ -z "$list" ]; then + echo "Group \"$group\" is empty or not defined?" + exit 1
_fatal"Group \"$group\" is empty or not defined?" ?
Quoted 2 lines
Remember that this was posted before those changes were made. This is the sort of thing that gets fixed on rebase...
Yeah, right.
Quoted 16 lines
Quoted 16 lines
common/test_list line 233
Quoted 11 lines
common/test_list new file +10
-0+_tl_setup_exclude_group() +{ + local xgroup="$1" + + _XGROUP_LIST="$_XGROUP_LIST ${xgroup//,/ }" +} + +_tl_setup_group() +{ + local group="$1"Minor: Since this function is public , do you think it is helpful to have a quick sanitization that makes sure that $1 is not empty and prints message otherwise? Similar comment for
_tl_setup_exclude_group().
Quoted 6 lines
Quoted 3 lines
common/test_list line 236
Quoted 3 lines
common/test_list new file +3
-0+ + _GROUP_LIST="$_GROUP_LIST ${group//,/ }" +}I don't think it needs it. It should be checked by the caller if necessary because it is the one doing option parsing to set up the group.
Okay.
Quoted 10 lines
Quoted 7 lines
common/test_list
Quoted 3 lines
common/test_list new file +2
-0+ +_tl_setup_cli()Nit: This function mostly sets up test list related variables - so maybe a name that suggests that. Maybe something like
_tl_setup_test_list_varsor_tl_setup_test_vars- just a suggestion, no hard preferences here.Ok, I'll come up with a new name for it, but keep in mind that this function is actually parsing arguments that have come directly from the CLI.
Yes, I have got that.
Quoted 30 lines
Quoted 22 lines
common/test_list line 262
Quoted 6 lines
common/test_list new file +6
-0+{ + while [ $# -gt 0 ]; do + case "$1" in + -*) echo "Arguments before tests, please!" + status=1 + exit $statususe
_fatal?Quoted 15 lines
common/test_list new file +15
-0+ ;; + *) # Expand test pattern (e.g. xfs/???, *fs/001) + local list=$(cd $_tl_src_dir; echo $1) + local t + + for t in $list; do + t=${t#$_tl_src_dir/} + local test_dir=${t%%/*} + local test_name=${t##*/} + local group_file=$_tl_src_dir/$test_dir/group.list + + if grep -Eq "^$test_name" $group_file; then + # in group file ... OK + echo $_tl_src_dir/$test_dir/$test_name \ + >> $_tl_fileQuoted 8 lines
common/test_list line 281
Quoted 4 lines
common/test_list new file +4
-0+ _tl_have_test_args=true + else + # oops + echo "$t - unknown test, ignored"Minor: Sometimes it has happened with me that I have added a new test (with ./new ) but forgot to run make and hence the test wasn't getting run and was reported to be an invalid test. Should we add a suggestion to run make, in case we encounter an unidentified test name?
Quoted 2 lines
Not in this patch set. If you want these sorts of process reminders, please send them as separate feature additions.
Yeah.
--NR
Quoted 2 lines
-Dave.
Signature · 3 lines
Nirjhar Roy Linux Kernel Developer IBM, Bangalore
PATCH 6/28 check-parallel: use common group list parsing code
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
Build the test list directly from command line prompts, rather than hard coding the tests and using the check infrastructure to filter that list.
We still pass exact test lists to check to execute the tests that each runner needs to execute, but all other test list commands are no longer passed to check.
As a result of this change, check-parallel no longer passes unknown CLI parameters through to the internal check invocations. At this point, the only non test-list related option is config file section selection; more of the check options will be brought across as needed in future patches.
--- check | 6 +- check-parallel | 156 ++++++++++++++++++++++++++++++++++++++++------- common/test_list | 7 +++ 3 files changed, 143 insertions(+), 26 deletions(-)
check +2 -4
@@ -43,11 +43,9 @@ timestamp=${TIMESTAMP:=false} rm -f $tmp.list $tmp.tmp $tmp.grep $here/$iam.out $tmp.report.* $tmp.arglist -# We need to include the test list processing first as argument parsing -# requires test list parsing and setup. -. ./common/test_list - . ./common/exit +. ./common/test_names +. ./common/test_list usage() {
check-parallel +134 -22
236 lines not on this page · raw message
common/test_list +7 -0
@@ -24,6 +24,7 @@ _tl_file="$tmp.test_list" _tl_exclude_tests=() _tl_tests= +# strip 'tests\' prefix from the provided test name _tl_strip_src_dir() { local test="$1"
@@ -31,6 +32,12 @@ _tl_strip_src_dir() echo ${test#$_tl_src_dir/} } +# strip 'tests\' prefix from all the tests in the test list +_tl_strip_test_list() +{ + _tl_tests=$(echo $_tl_tests | sed -e "s/$_tl_src_dir\///g") +} + get_sub_group_list() { local d=$1
Signature · 1 line
2.45.2
Re: [PATCH 06/28] check-parallel: use common group list parsing code
Quoted 110 lines · Dave Chinner
From: Dave Chinner <[email protected]>
Build the test list directly from command line prompts, rather than hard coding the tests and using the check infrastructure to filter that list.
We still pass exact test lists to check to execute the tests that each runner needs to execute, but all other test list commands are no longer passed to check.
As a result of this change, check-parallel no longer passes unknown CLI parameters through to the internal check invocations. At this point, the only non test-list related option is config file section selection; more of the check options will be brought across as needed in future patches.
Signed-off-by Dave Chinner <[email protected]>--- check | 6 +- check-parallel | 156 ++++++++++++++++++++++++++++++++++++++++------- common/test_list | 7 +++ 3 files changed, 143 insertions(+), 26 deletions(-)check +2
-4@@ -43,11 +43,9 @@ timestamp=${TIMESTAMP:=false} rm -f $tmp.list $tmp.tmp $tmp.grep $here/$iam.out $tmp.report.* $tmp.arglist -# We need to include the test list processing first as argument parsing -# requires test list parsing and setup. -. ./common/test_list - . ./common/exit +. ./common/test_names +. ./common/test_list usage() {check-parallel +49
-7@@ -9,18 +9,114 @@ # for them and runs the test in the background. When it completes, it tears down # the loop devices. -export SRC_DIR="tests" -basedir=$1 -shift -check_args="$*" +basedir="" runners=64 runner_list=() runtimes=() +show_test_list= +run_section="" +tmp=/tmp/check-parallel.$$ -# tests in auto group -test_list=$(awk '/^[0-9].*auto/ { print "generic/" $1 }' tests/generic/group.list) -test_list+=$(awk '/^[0-9].*auto/ { print "xfs/" $1 }' tests/xfs/group.list) +export FSTYP=xfs + +. ./common/exit +. ./common/test_names +. ./common/test_list + +usage() +{ + echo "Usage: $0 [options] [testlist]"' + +check options + -D <dir> Directory to run in + -n Output test list, do not run tests + -r randomize test order + --exact-order run tests in the exact order specified + -s section run only specified section from config file + +testlist options + -g group[,group...] include tests from these groups + -x group[,group...] exclude tests from these groups + -X exclude_file exclude individual tests + -e testlist exclude a specific list of tests + -E external_file exclude individual tests + [testlist] include tests matching names in testlist + +testlist argument is a list of tests in the form of <test dir>/<test name>. + +<test dir> is a directory under tests that contains a group file, +with a list of the names of the tests in that directory. + +<test name> may be either a specific test file name (e.g. xfs/001) or +a test file name match pattern (e.g. xfs/*). + +group argument is either a name of a tests group to collect from all +the test dirs (e.g. quick) or a name of a tests group to collect from +a specific tests dir in the form of <test dir>/<group name> (e.g. xfs/quick). +If you want to run all the tests in the test suite, use "-g all" to specify all +groups. + +exclude_file argument refers to a name of a file inside each test directory. +for every test dir where this file is found, the listed test names are +excluded from the list of tests to run from that test dir. + +external_file argument is a path to a single file containing a list of tests +to exclude in the form of <test dir>/<test name>.
check-parallel line 75
Quoted 10 lines
check-parallel +10
-0+ +examples: + check-parallel -D /mnt xfs/001 + check-parallel -D /mnt -g quick + check-parallel -D /mnt -g xfs/quick + check-parallel -D /mnt -x stress xfs/* + check-parallel -D /mnt -X .exclude -g auto + check-parallel -D /mnt -E ~/.xfstests.exclude +' + exit 1
_fatal ?
Quoted 160 lines
check-parallel +74
-14+} + +# Process command arguments first. +while [ $# -gt 0 ]; do + case "$1" in + -\? | -h | --help) usage ;; + + -D) basedir=$2; shift ;; + -g) _tl_setup_group $2 ; shift ;; + -e) _tl_setup_exclude_tests $2 ; shift ;; + -E) _tl_setup_exclude_file $2 ; shift ;; + -x) _tl_setup_exclude_group $2; shift ;; + -X) _tl_setup_exclude_subdir $2; shift ;; + -r) _tl_setup_randomise ;; + --exact-order) _tl_setup_ordered ;; + -n) show_test_list="yes" ;; + + -s) run_section="$run_section -s $2"; shift ;; + + -*) usage ;; + *) # not an argument, we've got tests now. + _tl_setup_cli $* + esac + + # if we've found a test specification, the break out of the processing + # loop before we shift the arguments so that this is the first argument + # that we process in the test arg loop below. + if $_tl_have_test_args; then + break; + fi + + shift +done + +if [ ! -d "$basedir" ]; then + echo "Invalid basedir specification" + usage +fi +if [ -d "$basedir/runner-0/" ]; then + prev_results=`ls -tr $basedir/runner-0/ | grep results | tail -1` +fi + +_tl_prepare_test_list +_tl_strip_test_list # grab all previously run tests and order them from highest runtime to lowest # We are going to try to run the longer tests first, hopefully so we can avoid@@ -30,25 +126,23 @@ test_list+=$(awk '/^[0-9].*auto/ { print "xfs/" $1 }' tests/xfs/group.list) # # If we have tests in the test list that don't have runtimes recorded, then # append them to be run last. - -build_runner_list() +time_order_test_list() { local runtimes local run_list=() - local prev_results=`ls -tr $basedir/runner-0/ | grep results | tail -1` runtimes=$(cat $basedir/*/$prev_results/check.time | sort -k 2 -nr | cut -d " " -f 1) # Iterate the timed list first. For every timed list entry that # is found in the test_list, add it to the local runner list. local -a _list=( $runtimes ) - local -a _tlist=( $test_list ) + local -a _tlist=( $_tl_tests ) local rx=0 local ix local jx #set -x for ((ix = 0; ix < ${#_list[*]}; ix++)); do - echo $test_list | grep -q ${_list[$ix]} + echo $_tl_tests | grep -q ${_list[$ix]} if [ $? == 0 ]; then # add the test to the new run list and remove # it from the remaining test list.@@ -60,24 +154,27 @@ build_runner_list() # The final test list is all the time ordered tests followed by # all the tests we didn't find time records for. - test_list="${run_list[*]} ${_tlist[*]}" + _tl_tests="${run_list[*]} ${_tlist[*]}" } -if [ -f $basedir/runner-0/results/check.time ]; then - build_runner_list +if ! $_tl_randomise -a ! $_tl_exact_order; then + if [ -f $basedir/runner-0/$prev_results/check.time ]; then + time_order_test_list + fi fi # split the list amongst N runners - split_runner_list() { local ix local rx - local -a _list=( $test_list ) + local -a _list=( $_tl_tests ) for ((ix = 0; ix < ${#_list[*]}; ix++)); do seq="${_list[$ix]}" rx=$((ix % $runners)) - runner_list[$rx]+="${_list[$ix]} " + if ! _tl_expunge_test $seq; then + runner_list[$rx]+="${_list[$ix]} " + fi #echo $seq done }@@ -137,7 +234,7 @@ runner_go() # Run the tests in it's own mount namespace, as per the comment below # that precedes making the basedir a private mount. - ./src/nsexec -m ./check $check_args -x unreliable_in_parallel --exact-order ${runner_list[$id]} > $me/log 2>&1 + ./src/nsexec -m ./check $run_section -x unreliable_in_parallel --exact-order ${runner_list[$id]} > $me/log 2>&1 wait sleep 1@@ -165,6 +262,13 @@ cleanup() trap "cleanup; exit" HUP INT QUIT TERM +split_runner_list +if [ -n "$show_test_list" ]; then + echo Time ordered test list: + echo $_tl_tests + echo +fi + # Each parallel test runner needs to only see it's own mount points. If we # leave the basedir as shared, then all tests see all mounts and then we get@@ -178,15 +282,23 @@ trap "cleanup; exit" HUP INT QUIT TERM # in it's own mount namespace so that they cannot see mounts that other tests # are performing. mount --make-private $basedir -split_runner_list + now=`date +%Y-%m-%d-%H:%M:%S` for ((i = 0; i < $runners; i++)); do - runner_go $i $now & + if [ -n "$show_test_list" ]; then + echo "Runner $i: ${runner_list[$i]}" + else + runner_go $i $now & + fi done; wait +if [ -n "$show_test_list" ]; then + exit 0 +fi + echo -n "Tests run: " grep Ran $basedir/*/log | sed -e 's,^.*:,,' -e 's, ,\n,g' | sort | uniq | wc -l
check-parallel grep Failures: $basedir/*/log | uniq | sed -e "s/^.*Failures://" -e "s,\([0-9]\) · line 313
Quoted 5 lines
check-parallel +1
-1@@ -198,7 +310,7 @@ grep Failures: $basedir/*/log | uniq | sed -e "s/^.*Failures://" -e "s,\([0-9]\) echo echo Ten slowest tests - runtime in seconds: -cat $basedir/*/results/check.time | sort -k 2 -nr | head -10 +cat $basedir/*/results-$now/check.time | sort -k 2 -nr | head -10
Maybe have this default to top 10 list but make it parameterized, just like we do it for DIFF_LENGTH (# number of diff lines from a failed test, 0 for whole output)? Do you think that would be useful?
Quoted 28 lines
check-parallel +0
-0echo echo Cleanup on Aisle 5?common/test_list +7
-0@@ -24,6 +24,7 @@ _tl_file="$tmp.test_list" _tl_exclude_tests=() _tl_tests= +# strip 'tests\' prefix from the provided test name _tl_strip_src_dir() { local test="$1"@@ -31,6 +32,12 @@ _tl_strip_src_dir() echo ${test#$_tl_src_dir/} } +# strip 'tests\' prefix from all the tests in the test list +_tl_strip_test_list() +{ + _tl_tests=$(echo $_tl_tests | sed -e "s/$_tl_src_dir\///g") +} + get_sub_group_list() { local d=$1
Re: [PATCH 06/28] check-parallel: use common group list parsing code
Quoted 18 lines · Nirjhar Roy (IBM)
Quoted 17 lines · Dave Chinner
From: Dave Chinner <[email protected]>
Build the test list directly from command line prompts, rather than hard coding the tests and using the check infrastructure to filter that list.
We still pass exact test lists to check to execute the tests that each runner needs to execute, but all other test list commands are no longer passed to check.
As a result of this change, check-parallel no longer passes unknown CLI parameters through to the internal check invocations. At this point, the only non test-list related option is config file section selection; more of the check options will be brought across as needed in future patches.
Signed-off-by Dave Chinner <[email protected]>
.....
Quoted 42 lines
Quoted 42 lines
check-parallel +42
-0+usage() +{ + echo "Usage: $0 [options] [testlist]"' + +check options + -D <dir> Directory to run in + -n Output test list, do not run tests + -r randomize test order + --exact-order run tests in the exact order specified + -s section run only specified section from config file + +testlist options + -g group[,group...] include tests from these groups + -x group[,group...] exclude tests from these groups + -X exclude_file exclude individual tests + -e testlist exclude a specific list of tests + -E external_file exclude individual tests + [testlist] include tests matching names in testlist + +testlist argument is a list of tests in the form of <test dir>/<test name>. + +<test dir> is a directory under tests that contains a group file, +with a list of the names of the tests in that directory. + +<test name> may be either a specific test file name (e.g. xfs/001) or +a test file name match pattern (e.g. xfs/*). + +group argument is either a name of a tests group to collect from all +the test dirs (e.g. quick) or a name of a tests group to collect from +a specific tests dir in the form of <test dir>/<group name> (e.g. xfs/quick). +If you want to run all the tests in the test suite, use "-g all" to specify all +groups. + +exclude_file argument refers to a name of a file inside each test directory. +for every test dir where this file is found, the listed test names are +excluded from the list of tests to run from that test dir. + +external_file argument is a path to a single file containing a list of tests +to exclude in the form of <test dir>/<test name>. + +examples: + check-parallel -D /mnt xfs/001
Quoted 8 lines
check-parallel line 75
Quoted 7 lines
+ check-parallel -D /mnt -g quick
+ check-parallel -D /mnt -g xfs/quick
+ check-parallel -D /mnt -x stress xfs/*
+ check-parallel -D /mnt -X .exclude -g auto
+ check-parallel -D /mnt -E ~/.xfstests.exclude
+'check-parallel +1
-0+ exit 1
_fatal?
This is a common usage() function pattern in shell scripts. It's clear that it exits with a non-zero error status when someone gets a command line parameter wrong.
Also, check-parallel does not have a status variable that needs to be set to propagate exit values, nor does it have any traps set up at this point to need exit value propagation.
Hence calling exit directly is really the right thing to do here.
Quoted 9 lines
check-parallel grep Failures: $basedir/*/log | uniq | sed -e "s/^.*Failures://" -e "s,\([0-9]\) · line 313
Quoted 6 lines
check-parallel +1
-1@@ -198,7 +310,7 @@ grep Failures: $basedir/*/log | uniq | sed -e "s/^.*Failures://" -e "s,\([0-9]\) echo echo Ten slowest tests - runtime in seconds: -cat $basedir/*/results/check.time | sort -k 2 -nr | head -10 +cat $basedir/*/results-$now/check.time | sort -k 2 -nr | head -10Maybe have this default to top 10 list but make it parameterized, just like we do it for
DIFF_LENGTH(# number of diff lines from a failed test, 0 for whole output)? Do you think that would be useful?
Not at this point in time. This is largely initial debug information; it tells me which test(s) are consuming all the time, and how close the overall runtime is to being bound by single test runtime. I only put this here because I was running that command on the CLI after a test run frequently, and generally I was only looking at the top 2 or 3 tests in the list....
e.g. if the whole auto group test run takes 9 minutes, and the longest test takes 8m30s, then it is clear that total runtime is bound by the slowest test. Make that test run faster, and overall auto group test time will run faster, too.
This tells me what tests I need to focus on for runtime optimisation (e.g. sync -> syncfs conversions). Once all the low hanging fruit is gone, it won't tell us anything useful and I'll remove it...
-Dave.
Signature · 2 lines
Dave Chinner [email protected]
Re: [PATCH 06/28] check-parallel: use common group list parsing code
Quoted 70 lines · Dave Chinner
Quoted 18 lines · Nirjhar Roy (IBM)
Quoted 17 lines · Dave Chinner
From: Dave Chinner <[email protected]>
Build the test list directly from command line prompts, rather than hard coding the tests and using the check infrastructure to filter that list.
We still pass exact test lists to check to execute the tests that each runner needs to execute, but all other test list commands are no longer passed to check.
As a result of this change, check-parallel no longer passes unknown CLI parameters through to the internal check invocations. At this point, the only non test-list related option is config file section selection; more of the check options will be brought across as needed in future patches.
Signed-off-by Dave Chinner <[email protected]>.....
Quoted 42 lines
Quoted 42 lines
check-parallel +42
-0+usage() +{ + echo "Usage: $0 [options] [testlist]"' + +check options + -D <dir> Directory to run in + -n Output test list, do not run tests + -r randomize test order + --exact-order run tests in the exact order specified + -s section run only specified section from config file + +testlist options + -g group[,group...] include tests from these groups + -x group[,group...] exclude tests from these groups + -X exclude_file exclude individual tests + -e testlist exclude a specific list of tests + -E external_file exclude individual tests + [testlist] include tests matching names in testlist + +testlist argument is a list of tests in the form of <test dir>/<test name>. + +<test dir> is a directory under tests that contains a group file, +with a list of the names of the tests in that directory. + +<test name> may be either a specific test file name (e.g. xfs/001) or +a test file name match pattern (e.g. xfs/*). + +group argument is either a name of a tests group to collect from all +the test dirs (e.g. quick) or a name of a tests group to collect from +a specific tests dir in the form of <test dir>/<group name> (e.g. xfs/quick). +If you want to run all the tests in the test suite, use "-g all" to specify all +groups. + +exclude_file argument refers to a name of a file inside each test directory. +for every test dir where this file is found, the listed test names are +excluded from the list of tests to run from that test dir. + +external_file argument is a path to a single file containing a list of tests +to exclude in the form of <test dir>/<test name>. + +examples: + check-parallel -D /mnt xfs/001Quoted 8 lines
check-parallel line 75
Quoted 7 lines
+ check-parallel -D /mnt -g quick
+ check-parallel -D /mnt -g xfs/quick
+ check-parallel -D /mnt -x stress xfs/*
+ check-parallel -D /mnt -X .exclude -g auto
+ check-parallel -D /mnt -E ~/.xfstests.exclude
+'check-parallel +1
-0+ exit 1
_fatal?
Quoted 7 lines
This is a common
usage()function pattern in shell scripts. It's clear that it exits with a non-zero error status when someone gets a command line parameter wrong.Also, check-parallel does not have a status variable that needs to be set to propagate exit values, nor does it have any traps set up at this point to need exit value propagation.
Oh, right. Yeah, check-parallel directly will exit from here, so yes exit 1 should be fine. Thanks.
Quoted 19 lines
Hence calling exit directly is really the right thing to do here.
Quoted 9 lines
check-parallel grep Failures: $basedir/*/log | uniq | sed -e "s/^.*Failures://" -e "s,\([0-9]\) · line 313
Quoted 6 lines
check-parallel +1
-1@@ -198,7 +310,7 @@ grep Failures: $basedir/*/log | uniq | sed -e "s/^.*Failures://" -e "s,\([0-9]\) echo echo Ten slowest tests - runtime in seconds: -cat $basedir/*/results/check.time | sort -k 2 -nr | head -10 +cat $basedir/*/results-$now/check.time | sort -k 2 -nr | head -10Maybe have this default to top 10 list but make it parameterized, just like we do it for
DIFF_LENGTH(# number of diff lines from a failed test, 0 for whole output)? Do you think that would be useful?Not at this point in time. This is largely initial debug information; it tells me which test(s) are consuming all the time, and how close the overall runtime is to being bound by single test runtime. I only put this here because I was running that command on the CLI after a test run frequently, and generally I was only looking at the top 2 or 3 tests in the list....
Quoted 8 lines
e.g. if the whole auto group test run takes 9 minutes, and the longest test takes 8m30s, then it is clear that total runtime is bound by the slowest test. Make that test run faster, and overall auto group test time will run faster, too.
This tells me what tests I need to focus on for runtime optimisation (e.g. sync -> syncfs conversions). Once all the low hanging fruit is gone, it won't tell us anything useful and I'll remove it...
Ah, okay. Now I get the reason for introducing the information of the top slow running tests. However, we can still keep this feature - in case someone writes a new test and uses some functionality or command (like syncfs instead of sync) and slows it down, although that can be caught in code reviews too. Anyway, no hard preferences here. But thanks for pointing out the reason for introducing the top slow running tests information.
--NR
Quoted 2 lines
-Dave.
Signature · 3 lines
Nirjhar Roy Linux Kernel Developer IBM, Bangalore
PATCH 7/28 check-parallel: adjust concurrency according to CPU count
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
Concurrency is currently hard coded at 64 worker threads. This is too many for small CPU count machines; the idea is to create a sustained load of roughly one test per CPU as they are mostly single threaded/single process tests. The number "64" was chosen because I've been developing this functionality on a 64p VM.
Rather than hard coding the concurrency, probe the number of CPUs available and create that many running contexts as the default concurrency to use.
Further, add a CLI option to specify the number of threads to run so that we can over- or under-commit the CPU resources to enable direct benchmarking of performance with different levels of concurrency.
Let's use that capability to show how much check-parallel can benefit small systems. Using a single check execution thread for all tests inside a 4p control group to limit maximum CPU usage to the equivalent of a small 4p machine:
$ time sudo numactl -C 4-7 ./check-parallel -D /mnt/xfs -t 1 -g quick -s xfs -x dump -X generic/531 Runner 0 Failures: generic/504 Tests run: 921 Tests _notrun: 272 Failure count: 2 .....
real 61m31.362s user 0m0.029s sys 0m0.059s
the quick group on XFS takes *over an hour* to run.
If we use the same 4p control group setup and run with 8 test execution threads to ensure the 4 CPUs are fully utilised for most of the test run:
$ time sudo numactl -C 4-7 ./check-parallel -D /mnt/xfs -t 8 -g quick -s xfs -x dump -X generic/531 Runner 7 Failures: generic/504 Tests run: 921 Tests _notrun: 145 Failure count: 1 .....
real 17m33.124s user 0m0.009s sys 0m0.017s
The same test run takes only 17m33s. The same number of tests were run, the same failures occurred. [ Ignore the differences in notrun/failure count - the multi-file aggregation currently doesn't work correctly for the single log file case. ]
That's a reduction in test runtime of ~72% for a 4 CPU system. Or, if we want to measure it the other way, we get a ~3.5x improvement in runtime scalability. i.e. going from 1 -> 4 CPUs being used for test execution (4x increase) we get a 3.5x improvement in scalability when we go from check to check-parallel.
--- check-parallel | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)
check-parallel +8 -1
@@ -10,7 +10,7 @@ # the loop devices. basedir="" -runners=64 +runners=$(getconf _NPROCESSORS_CONF) runner_list=() runtimes=() show_test_list=
@@ -30,6 +30,7 @@ usage() check options -D <dir> Directory to run in + -t <n> Number of concurrent tests to run -n Output test list, do not run tests -r randomize test order --exact-order run tests in the exact order specified
@@ -81,6 +82,7 @@ while [ $# -gt 0 ]; do -\? | -h | --help) usage ;; -D) basedir=$2; shift ;; + -t) runners=$2; shift ;; -g) _tl_setup_group $2 ; shift ;; -e) _tl_setup_exclude_tests $2 ; shift ;; -E) _tl_setup_exclude_file $2 ; shift ;;
@@ -111,6 +113,11 @@ if [ ! -d "$basedir" ]; then echo "Invalid basedir specification" usage fi +if [[ $runners -le 0 || $runners -gt 1024 ]]; then + echo "Invalid thread specificaton: $runners" + usage +fi + if [ -d "$basedir/runner-0/" ]; then prev_results=`ls -tr $basedir/runner-0/ | grep results | tail -1` fi
Signature · 1 line
2.45.2
Re: [PATCH 07/28] check-parallel: adjust concurrency according to CPU count
Quoted 54 lines · Dave Chinner
From: Dave Chinner <[email protected]>
Concurrency is currently hard coded at 64 worker threads. This is too many for small CPU count machines; the idea is to create a sustained load of roughly one test per CPU as they are mostly single threaded/single process tests. The number "64" was chosen because I've been developing this functionality on a 64p VM.
Rather than hard coding the concurrency, probe the number of CPUs available and create that many running contexts as the default concurrency to use.
Further, add a CLI option to specify the number of threads to run so that we can over- or under-commit the CPU resources to enable direct benchmarking of performance with different levels of concurrency.
Let's use that capability to show how much check-parallel can benefit small systems. Using a single check execution thread for all tests inside a 4p control group to limit maximum CPU usage to the equivalent of a small 4p machine:
$ time sudo numactl -C 4-7 ./check-parallel -D /mnt/xfs -t 1 -g quick -s xfs -x dump -X generic/531 Runner 0 Failures: generic/504 Tests run: 921 Tests _notrun: 272 Failure count: 2 .....real 61m31.362s user 0m0.029s sys 0m0.059sthe quick group on XFS takes *over an hour* to run.
If we use the same 4p control group setup and run with 8 test execution threads to ensure the 4 CPUs are fully utilised for most of the test run:
$ time sudo numactl -C 4-7 ./check-parallel -D /mnt/xfs -t 8 -g quick -s xfs -x dump -X generic/531 Runner 7 Failures: generic/504 Tests run: 921 Tests _notrun: 145 Failure count: 1 .....real 17m33.124s user 0m0.009s sys 0m0.017sThe same test run takes only 17m33s. The same number of tests were run, the same failures occurred. [ Ignore the differences in notrun/failure count - the multi-file aggregation currently doesn't work correctly for the single log file case. ]
Quoted 5 lines
That's a reduction in test runtime of ~72% for a 4 CPU system. Or, if we want to measure it the other way, we get a ~3.5x improvement in runtime scalability. i.e. going from 1 -> 4 CPUs being used for test execution (4x increase) we get a 3.5x improvement in scalability when we go from check to check-parallel.
The functionality looks useful to me and the implementation is also fine as far as I understand. I have some minor/nit comments below. Other than that:
Quoted 11 lines
Signed-off-by Dave Chinner <[email protected]>--- check-parallel | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)check-parallel +0
-0
check-parallel line 13
Quoted 5 lines
check-parallel +1
-1@@ -10,7 +10,7 @@ # the loop devices. basedir="" -runners=64 +runners=$(getconf _NPROCESSORS_CONF)
Minor: Not related to this change. Maybe we should have a _get_nproc() just like_get_page_size()
{
echo $(getconf PAGE_SIZE)
}
_get_nproc()
{
echo $(getconf _NPROCESSORS_CONF)
}and replace all the $(getconf _NPROCESSORS_CONF) with $(_get_nproc)
check-parallel usage() · line 33
Quoted 8 lines
check-parallel +1
-0runner_list=() runtimes=() show_test_list=@@ -30,6 +30,7 @@ usage() check options -D <dir> Directory to run in + -t <n> Number of concurrent tests to run
Minor: Maybe we should mention the valid range of <n> i.e, 0 to 1024?
check-parallel usage() · line 34
Quoted 1 line
check-parallel +0
-0-n Output test list, do not run tests
Nit: Maybe there is some spacing issue here? "Number of concurrent ..." and "Output test..." don't begin together.
--NR
Quoted 11 lines
check-parallel +1
-0-r randomize test order --exact-order run tests in the exact order specified@@ -81,6 +82,7 @@ while [ $# -gt 0 ]; do -\? | -h | --help) usage ;; -D) basedir=$2; shift ;; + -t) runners=$2; shift ;; -g) _tl_setup_group $2 ; shift ;; -e) _tl_setup_exclude_tests $2 ; shift ;; -E) _tl_setup_exclude_file $2 ; shift ;;
check-parallel if [ ! -d "$basedir" ]; then · line 117
Quoted 5 lines
check-parallel +2
-0@@ -111,6 +113,11 @@ if [ ! -d "$basedir" ]; then echo "Invalid basedir specification" usage fi +if [[ $runners -le 0 || $runners -gt 1024 ]]; then + echo "Invalid thread specificaton: $runners"
Minor: Maybe we should mention the valid range of "runners" in this error message (0 to 1024)?
Quoted 6 lines
check-parallel +3
-0+ usage +fi + if [ -d "$basedir/runner-0/" ]; then prev_results=`ls -tr $basedir/runner-0/ | grep results | tail -1` fi
Re: [PATCH 07/28] check-parallel: adjust concurrency according to CPU count
Quoted 8 lines · Nirjhar Roy (IBM)
Quoted 7 lines · Dave Chinner
From: Dave Chinner <[email protected]>
Concurrency is currently hard coded at 64 worker threads. This is too many for small CPU count machines; the idea is to create a sustained load of roughly one test per CPU as they are mostly single threaded/single process tests. The number "64" was chosen because I've been developing this functionality on a 64p VM.
.....
Quoted 13 lines
check-parallel line 13
Quoted 10 lines
check-parallel +1
-1@@ -10,7 +10,7 @@ # the loop devices. basedir="" -runners=64 +runners=$(getconf _NPROCESSORS_CONF)Minor: Not related to this change. Maybe we should have a
_get_nproc()just like_get_page_size()
Quoted 8 lines
{ echo $(getconf PAGE_SIZE) } _get_nproc() { echo $(getconf _NPROCESSORS_CONF) }and replace all the $(getconf
_NPROCESSORS_CONF) with $(_get_nproc)
I have thoughts on that.
I think determining test scaling based on the number of CPUs in the machine is wrong. If I run in a cgroup that limits tests to 4p on a 64p machine, then fstests will think it is running on a 64p machine rahter than on 4p. That's ... not ideal.
I think there should be a global variable that defines the concurrency that tests should use to scale load/processes rather than use the CPU count. Then check/check-parallel can set the concurrency as they desire and we no longer have to worry about tests creating excesively huge loads on high CPU count machines because they were sized to load a small, low concurrency test system.
i.e. I intend to separate the concurrency with which check-parallel runs from the concurrency that individual tests use to scale. For check-parallel, I am thinking of fixing test concurrency scaling at something like min(nr_cpus, 8), whilst the check-parallel harness itself uses nr_cpus to determine how many concurrent runners to instantiate.
Quoted 4 lines
Quoted 4 lines
check-parallel +0
-0runtimes=() show_test_list=@@ -30,6 +30,7 @@ usage()
Quoted 8 lines
check-parallel usage() · line 33
Quoted 3 lines
check-parallel +1
-0check options -D <dir> Directory to run in + -t <n> Number of concurrent tests to runMinor: Maybe we should mention the valid range of <n> i.e, 0 to 1024?
check-parallel usage() · line 34
Quoted 1 line
check-parallel +0
-0-n Output test list, do not run testsNit: Maybe there is some spacing issue here? "Number of concurrent ..." and "Output test..." don't begin together.
Not that I can see, I think this is just email quoting and tabs doing funky stuff.
Quoted 5 lines
Quoted 5 lines
check-parallel +0
-0-r randomize test order --exact-order run tests in the exact order specified@@ -81,6 +82,7 @@ while [ $# -gt 0 ]; do -\? | -h | --help) usage ;;
Quoted 13 lines
check-parallel if [ ! -d "$basedir" ]; then · line 117
Quoted 11 lines
check-parallel +3
-0-D) basedir=$2; shift ;; + -t) runners=$2; shift ;; -g) _tl_setup_group $2 ; shift ;; -e) _tl_setup_exclude_tests $2 ; shift ;; -E) _tl_setup_exclude_file $2 ; shift ;;@@ -111,6 +113,11 @@ if [ ! -d "$basedir" ]; then echo "Invalid basedir specification" usage fi +if [[ $runners -le 0 || $runners -gt 1024 ]]; then + echo "Invalid thread specificaton: $runners"Minor: Maybe we should mention the valid range of "runners" in this error message (0 to 1024)?
/me shrugs.
It's an arbitrary "check-parallel won't ever scale past this" limit because the diminishing returns from added concurrency that Amdahl's Law define is already kicking in at 64 threads. If you are running check-parallel on a >1024p machine, you already know enough to be able to look at the source code to determine why the error is being emitted. :)
-Dave.
Signature · 2 lines
Dave Chinner [email protected]
Re: [PATCH 07/28] check-parallel: adjust concurrency according to CPU count
Quoted 46 lines · Dave Chinner
Quoted 8 lines · Nirjhar Roy (IBM)
Quoted 7 lines · Dave Chinner
From: Dave Chinner <[email protected]>
Concurrency is currently hard coded at 64 worker threads. This is too many for small CPU count machines; the idea is to create a sustained load of roughly one test per CPU as they are mostly single threaded/single process tests. The number "64" was chosen because I've been developing this functionality on a 64p VM.
.....
Quoted 13 lines
check-parallel line 13
Quoted 10 lines
check-parallel +1
-1@@ -10,7 +10,7 @@ # the loop devices. basedir="" -runners=64 +runners=$(getconf _NPROCESSORS_CONF)Minor: Not related to this change. Maybe we should have a
_get_nproc()just like_get_page_size()Quoted 8 lines
{ echo $(getconf PAGE_SIZE) } _get_nproc() { echo $(getconf _NPROCESSORS_CONF) }and replace all the $(getconf
_NPROCESSORS_CONF) with $(_get_nproc)I have thoughts on that.
I think determining test scaling based on the number of CPUs in the machine is wrong. If I run in a cgroup that limits tests to 4p on a 64p machine, then fstests will think it is running on a 64p machine rahter than on 4p. That's ... not ideal.
I think there should be a global variable that defines the concurrency that tests should use to scale load/processes rather than use the CPU count. Then check/check-parallel can set the concurrency as they desire and we no longer have to worry about tests creating excesively huge loads on high CPU count machines because they were sized to load a small, low concurrency test system.
Quoted 6 lines
i.e. I intend to separate the concurrency with which check-parallel runs from the concurrency that individual tests use to scale. For check-parallel, I am thinking of fixing test concurrency scaling at something like
min(nr_cpus, 8), whilst the check-parallel harness itself usesnr_cpusto determine how many concurrent runners to instantiate.
Okay yes. That makes sense.
Quoted 9 lines
Quoted 9 lines
check-parallel usage() · line 33
Quoted 8 lines
check-parallel +1
-0runtimes=() show_test_list=@@ -30,6 +30,7 @@ usage() check options -D <dir> Directory to run in + -t <n> Number of concurrent tests to runMinor: Maybe we should mention the valid range of <n> i.e, 0 to 1024?
Quoted 5 lines
Quoted 3 lines
check-parallel usage() · line 34
Quoted 1 line
check-parallel +0
-0-n Output test list, do not run testsNit: Maybe there is some spacing issue here? "Number of concurrent ..." and "Output test..." don't begin together.
Not that I can see, I think this is just email quoting and tabs doing funky stuff.
Okay.
Quoted 18 lines
Quoted 5 lines
Quoted 5 lines
check-parallel +0
-0-r randomize test order --exact-order run tests in the exact order specified@@ -81,6 +82,7 @@ while [ $# -gt 0 ]; do -\? | -h | --help) usage ;;Quoted 13 lines
check-parallel if [ ! -d "$basedir" ]; then · line 117
Quoted 11 lines
check-parallel +3
-0-D) basedir=$2; shift ;; + -t) runners=$2; shift ;; -g) _tl_setup_group $2 ; shift ;; -e) _tl_setup_exclude_tests $2 ; shift ;; -E) _tl_setup_exclude_file $2 ; shift ;;@@ -111,6 +113,11 @@ if [ ! -d "$basedir" ]; then echo "Invalid basedir specification" usage fi +if [[ $runners -le 0 || $runners -gt 1024 ]]; then + echo "Invalid thread specificaton: $runners"Minor: Maybe we should mention the valid range of "runners" in this error message (0 to 1024)?
Quoted 8 lines
/me shrugs.
It's an arbitrary "check-parallel won't ever scale past this" limit because the diminishing returns from added concurrency that Amdahl's Law define is already kicking in at 64 threads. If you are running check-parallel on a >1024p machine, you already know enough to be able to look at the source code to determine why the error is being emitted. :)
Yeah. Got it.
This patch looks okay to me. Thanks for the clarifications.
--NR
Quoted 2 lines
-Dave.
Signature · 3 lines
Nirjhar Roy Linux Kernel Developer IBM, Bangalore
PATCH 8/28 check-parallel: add logwrite device support
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
Every logwrite test will use the same /dev/mapper/<dev> name for the logwrites device, so we also need to convert common/dmlogwrite to use per-test device names as we have done for other dm devices.
Then add a per-test-runner LOGWRITES_DEV so that all tests using dm-logwrites now get run by check-parallel.
--- check-parallel | 4 ++++ common/dmlogwrites | 5 +++-- 2 files changed, 7 insertions(+), 2 deletions(-)
check-parallel +4 -0
@@ -216,12 +216,14 @@ runner_go() local me=$basedir/runner-$id local _test=$me/test.img local _scratch=$me/scratch.img + local _logwrites=$me/logwrites.img local _results=$me/results-$2 mkdir -p $me xfs_io -f -c 'truncate 2g' $_test xfs_io -f -c 'truncate 8g' $_scratch + xfs_io -f -c 'truncate 1g' $_logwrites mkfs.xfs -f $_test > /dev/null 2>&1
@@ -229,6 +231,7 @@ runner_go() export TEST_DIR=$me/test export SCRATCH_DEV=$(_create_loop_device $_scratch) export SCRATCH_MNT=$me/scratch + export LOGWRITES_DEV=$(_create_loop_device $_logwrites) export FSTYP=xfs export RESULT_BASE=$_results
@@ -249,6 +252,7 @@ runner_go() umount -R $SCRATCH_MNT 2> /dev/null _destroy_loop_device $TEST_DEV _destroy_loop_device $SCRATCH_DEV + _destroy_loop_device $LOGWRITES_DEV grep -q Failures: $me/log if [ $? -eq 0 ]; then
common/dmlogwrites +3 -2
@@ -4,6 +4,9 @@ # # common functions for setting up and tearing down a dm log-writes device +LOGWRITES_NAME=logwrites-$seq +LOGWRITES_DMDEV=/dev/mapper/$LOGWRITES_NAME + _require_log_writes() { [ -z "$LOGWRITES_DEV" -o ! -b "$LOGWRITES_DEV" ] && \
@@ -81,8 +84,6 @@ _log_writes_init() BLK_DEV_SIZE=$((length / blksz)) fi - LOGWRITES_NAME=logwrites-test - LOGWRITES_DMDEV=/dev/mapper/$LOGWRITES_NAME LOGWRITES_TABLE="0 $BLK_DEV_SIZE log-writes $blkdev $LOGWRITES_DEV" _dmsetup_create $LOGWRITES_NAME --table "$LOGWRITES_TABLE" || \ _fail "failed to create log-writes device"
Signature · 1 line
2.45.2
Re: [PATCH 08/28] check-parallel: add logwrite device support
Quoted 56 lines · Dave Chinner
From: Dave Chinner <[email protected]>
Every logwrite test will use the same /dev/mapper/<dev> name for the logwrites device, so we also need to convert common/dmlogwrite to use per-test device names as we have done for other dm devices.
Then add a per-test-runner
LOGWRITES_DEVso that all tests using dm-logwrites now get run by check-parallel.Signed-off-by Dave Chinner <[email protected]>--- check-parallel | 4 ++++ common/dmlogwrites | 5 +++-- 2 files changed, 7 insertions(+), 2 deletions(-)check-parallel +4
-0@@ -216,12 +216,14 @@ runner_go() local me=$basedir/runner-$id local _test=$me/test.img local _scratch=$me/scratch.img + local _logwrites=$me/logwrites.img local _results=$me/results-$2 mkdir -p $me xfs_io -f -c 'truncate 2g' $_test xfs_io -f -c 'truncate 8g' $_scratch + xfs_io -f -c 'truncate 1g' $_logwrites mkfs.xfs -f $_test > /dev/null 2>&1@@ -229,6 +231,7 @@ runner_go() export TEST_DIR=$me/test export SCRATCH_DEV=$(_create_loop_device $_scratch) export SCRATCH_MNT=$me/scratch + export LOGWRITES_DEV=$(_create_loop_device $_logwrites) export FSTYP=xfs export RESULT_BASE=$_results@@ -249,6 +252,7 @@ runner_go() umount -R $SCRATCH_MNT 2> /dev/null _destroy_loop_device $TEST_DEV _destroy_loop_device $SCRATCH_DEV + _destroy_loop_device $LOGWRITES_DEV grep -q Failures: $me/log if [ $? -eq 0 ]; thencommon/dmlogwrites +0
-0
common/dmlogwrites line 7
Quoted 4 lines
common/dmlogwrites +1
-0@@ -4,6 +4,9 @@ # # common functions for setting up and tearing down a dm log-writes device +LOGWRITES_NAME=logwrites-$seq
if 2 different runners are running tests with the same $seq, won't there be a conflict? For example runner-0 is running xfs/xyz and runner-1 is running generic/xyz ?
The rest looks fine to me.
--NR
Quoted 14 lines
common/dmlogwrites +2
-2+LOGWRITES_DMDEV=/dev/mapper/$LOGWRITES_NAME + _require_log_writes() { [ -z "$LOGWRITES_DEV" -o ! -b "$LOGWRITES_DEV" ] && \@@ -81,8 +84,6 @@ _log_writes_init() BLK_DEV_SIZE=$((length / blksz)) fi - LOGWRITES_NAME=logwrites-test - LOGWRITES_DMDEV=/dev/mapper/$LOGWRITES_NAME LOGWRITES_TABLE="0 $BLK_DEV_SIZE log-writes $blkdev $LOGWRITES_DEV" _dmsetup_create $LOGWRITES_NAME --table "$LOGWRITES_TABLE" || \ _fail "failed to create log-writes device"
Re: [PATCH 08/28] check-parallel: add logwrite device support
Quoted 12 lines · Nirjhar Roy (IBM)
Quoted 11 lines · Dave Chinner
From: Dave Chinner <[email protected]>
Every logwrite test will use the same /dev/mapper/<dev> name for the logwrites device, so we also need to convert common/dmlogwrite to use per-test device names as we have done for other dm devices.
Then add a per-test-runner
LOGWRITES_DEVso that all tests using dm-logwrites now get run by check-parallel.Signed-off-by Dave Chinner <[email protected]>
.....
Quoted 8 lines
Quoted 8 lines
common/dmlogwrites +0
-0@@ -4,6 +4,9 @@ # # common functions for setting up and tearing down a dm log-writes device
Quoted 4 lines
common/dmlogwrites line 7
Quoted 1 line
common/dmlogwrites +1
-0+LOGWRITES_NAME=logwrites-$seqif 2 different runners are running tests with the same $seq, won't there be a conflict? For example runner-0 is running xfs/xyz and runner-1 is running generic/xyz ?
Yes, it could happen, but it won't happen right now as there is no overlap in test numbers using dmlogwrites between different test directories.
I do need to solve this generically (i.e. for all the dm device types) in the near future because I want to be able to run the same test N times in parallel. e.g do flakey test profiling really quickly over thousands of iterations by running 64 instances of the test at the same time.
For this use case, the dm device name cannot rely on the test name/sequence number at all, and so I'm going to have to come up with a unique ID of some kind for this purpose. It may be as simple as encoding the unique runner ID along with the test sequence number into the device name - it's not a difficult issue to solve.
Hence I haven't spent much more thought on it than that, because I've answered the two main questions that were relevant: "can I solve it?" (yes) and "do I need to solve it now?" (no).
-Dave.
Signature · 2 lines
Dave Chinner [email protected]
Re: [PATCH 08/28] check-parallel: add logwrite device support
Quoted 42 lines · Dave Chinner
Quoted 12 lines · Nirjhar Roy (IBM)
Quoted 11 lines · Dave Chinner
From: Dave Chinner <[email protected]>
Every logwrite test will use the same /dev/mapper/<dev> name for the logwrites device, so we also need to convert common/dmlogwrite to use per-test device names as we have done for other dm devices.
Then add a per-test-runner
LOGWRITES_DEVso that all tests using dm-logwrites now get run by check-parallel.Signed-off-by Dave Chinner <[email protected]>.....
Quoted 8 lines
Quoted 8 lines
common/dmlogwrites +0
-0@@ -4,6 +4,9 @@ # # common functions for setting up and tearing down a dm log-writes deviceQuoted 4 lines
common/dmlogwrites line 7
Quoted 1 line
common/dmlogwrites +1
-0+LOGWRITES_NAME=logwrites-$seqif 2 different runners are running tests with the same $seq, won't there be a conflict? For example runner-0 is running xfs/xyz and runner-1 is running generic/xyz ?
Yes, it could happen, but it won't happen right now as there is no overlap in test numbers using dmlogwrites between different test directories.
I do need to solve this generically (i.e. for all the dm device types) in the near future because I want to be able to run the same test N times in parallel. e.g do flakey test profiling really quickly over thousands of iterations by running 64 instances of the test at the same time.
For this use case, the dm device name cannot rely on the test name/sequence number at all, and so I'm going to have to come up with a unique ID of some kind for this purpose. It may be as simple as encoding the unique runner ID along with the test sequence number into the device name - it's not a difficult issue to solve.
Quoted 3 lines
Hence I haven't spent much more thought on it than that, because I've answered the two main questions that were relevant: "can I solve it?" (yes) and "do I need to solve it now?" (no).
Okay. Yes, I just wanted to point this out and confirm if my understanding is correct.
--NR
Quoted 2 lines
-Dave.
Signature · 3 lines
Nirjhar Roy Linux Kernel Developer IBM, Bangalore
PATCH 9/28 check-parallel: allow FSTYP selection from the CLI
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
Add a CLI option to specify the initial FSTYP to test. If this is not specified the the default of "xfs" will be used. This option is different to the way check has FSTYP specified as check-parallel has no infrastructure to support non block device based filesystems and hence we have to reject virtual or network based filesysetms are this point in time.
Note: This patch only implements default mkfs parameter support for the test device. Config sections can be used to override this as check will then format the test device when the section that defines non-default test device mkfs options is selected.
--- check-parallel | 44 +++++++++++++++++++++++++++++++++++++------- 1 file changed, 37 insertions(+), 7 deletions(-)
check-parallel +37 -7
@@ -18,7 +18,7 @@ run_section="" tmp=/tmp/check-parallel.$$ -export FSTYP=xfs +FSTYP= . ./common/exit . ./common/test_names
@@ -35,6 +35,7 @@ check options -r randomize test order --exact-order run tests in the exact order specified -s section run only specified section from config file + -f <FSTYPE> specify the filesystem type to test testlist options -g group[,group...] include tests from these groups
@@ -66,16 +67,39 @@ external_file argument is a path to a single file containing a list of tests to exclude in the form of <test dir>/<test name>. examples: - check-parallel -D /mnt xfs/001 - check-parallel -D /mnt -g quick + check-parallel -f xfs -D /mnt xfs/001 + check-parallel -f ext4 -D /mnt -g quick check-parallel -D /mnt -g xfs/quick check-parallel -D /mnt -x stress xfs/* - check-parallel -D /mnt -X .exclude -g auto - check-parallel -D /mnt -E ~/.xfstests.exclude + check-parallel -f btrfs -D /mnt -X .exclude -g auto + check-parallel -f udf -D /mnt -E ~/.xfstests.exclude ' exit 1 } +# Only support block device based filesystems with generic mkfs support +# at the moment. +is_supported_fstype() +{ + local fstype=$1 + + case $fstype in + xfs) ;; + ext2|ext3|ext4) ;; + udf) ;; + jfs) ;; + f2fs) ;; + btrfs) ;; + bcachefs) ;; + gfs2) ;; + ocfs2) ;; + *) + echo "unsupported FSTYPE: $fstype" + usage + ;; + esac +} + # Process command arguments first. while [ $# -gt 0 ]; do case "$1" in
@@ -92,6 +116,8 @@ while [ $# -gt 0 ]; do --exact-order) _tl_setup_ordered ;; -n) show_test_list="yes" ;; + -f) is_supported_fstype $2 ; FSTYP=$2; shift ;; + -s) run_section="$run_section -s $2"; shift ;; -*) usage ;;
@@ -109,6 +135,8 @@ while [ $# -gt 0 ]; do shift done +export FSTYP=${FSTYP:=xfs} + if [ ! -d "$basedir" ]; then echo "Invalid basedir specification" usage
@@ -225,8 +253,6 @@ runner_go() xfs_io -f -c 'truncate 8g' $_scratch xfs_io -f -c 'truncate 1g' $_logwrites - mkfs.xfs -f $_test > /dev/null 2>&1 - export TEST_DEV=$(_create_loop_device $_test) export TEST_DIR=$me/test export SCRATCH_DEV=$(_create_loop_device $_scratch)
@@ -240,6 +266,10 @@ runner_go() mkdir -p $RESULT_BASE rm -f $RESULT_BASE/check.* + # Only supports default mkfs parameters right now + wipefs -a $TEST_DEV > /dev/null 2>&1 + yes | mkfs -t $FSTYP $TEST_MKFS_OPTS $TEST_DEV > /dev/null 2>&1 + # export DUMP_CORRUPT_FS=1 # Run the tests in it's own mount namespace, as per the comment below
Signature · 1 line
2.45.2
Re: [PATCH 09/28] check-parallel: allow FSTYP selection from the CLI
Quoted 64 lines · Dave Chinner
From: Dave Chinner <[email protected]>
Add a CLI option to specify the initial FSTYP to test. If this is not specified the the default of "xfs" will be used. This option is different to the way check has FSTYP specified as check-parallel has no infrastructure to support non block device based filesystems and hence we have to reject virtual or network based filesysetms are this point in time.
Note: This patch only implements default mkfs parameter support for the test device. Config sections can be used to override this as check will then format the test device when the section that defines non-default test device mkfs options is selected.
Signed-off-by Dave Chinner <[email protected]>--- check-parallel | 44 +++++++++++++++++++++++++++++++++++++------- 1 file changed, 37 insertions(+), 7 deletions(-)check-parallel +12
-5@@ -18,7 +18,7 @@ run_section="" tmp=/tmp/check-parallel.$$ -export FSTYP=xfs +FSTYP= . ./common/exit . ./common/test_names@@ -35,6 +35,7 @@ check options -r randomize test order --exact-order run tests in the exact order specified -s section run only specified section from config file + -f <FSTYPE> specify the filesystem type to test testlist options -g group[,group...] include tests from these groups@@ -66,16 +67,39 @@ external_file argument is a path to a single file containing a list of tests to exclude in the form of <test dir>/<test name>. examples: - check-parallel -D /mnt xfs/001 - check-parallel -D /mnt -g quick + check-parallel -f xfs -D /mnt xfs/001 + check-parallel -f ext4 -D /mnt -g quick check-parallel -D /mnt -g xfs/quick check-parallel -D /mnt -x stress xfs/* - check-parallel -D /mnt -X .exclude -g auto - check-parallel -D /mnt -E ~/.xfstests.exclude + check-parallel -f btrfs -D /mnt -X .exclude -g auto + check-parallel -f udf -D /mnt -E ~/.xfstests.exclude ' exit 1 } +# Only support block device based filesystems with generic mkfs support +# at the moment. +is_supported_fstype() +{ + local fstype=$1 +
check-parallel external_file argument is a path to a single file containing a list of tests · line 95
Quoted 10 lines
check-parallel +10
-0+ case $fstype in + xfs) ;; + ext2|ext3|ext4) ;; + udf) ;; + jfs) ;; + f2fs) ;; + btrfs) ;; + bcachefs) ;; + gfs2) ;; + ocfs2) ;;
Extremely minor: Maybe list the test in alphabetical order? The rest looks fine to me.
Quoted 48 lines
check-parallel +15
-2+ *) + echo "unsupported FSTYPE: $fstype" + usage + ;; + esac +} + # Process command arguments first. while [ $# -gt 0 ]; do case "$1" in@@ -92,6 +116,8 @@ while [ $# -gt 0 ]; do --exact-order) _tl_setup_ordered ;; -n) show_test_list="yes" ;; + -f) is_supported_fstype $2 ; FSTYP=$2; shift ;; + -s) run_section="$run_section -s $2"; shift ;; -*) usage ;;@@ -109,6 +135,8 @@ while [ $# -gt 0 ]; do shift done +export FSTYP=${FSTYP:=xfs} + if [ ! -d "$basedir" ]; then echo "Invalid basedir specification" usage@@ -225,8 +253,6 @@ runner_go() xfs_io -f -c 'truncate 8g' $_scratch xfs_io -f -c 'truncate 1g' $_logwrites - mkfs.xfs -f $_test > /dev/null 2>&1 - export TEST_DEV=$(_create_loop_device $_test) export TEST_DIR=$me/test export SCRATCH_DEV=$(_create_loop_device $_scratch)@@ -240,6 +266,10 @@ runner_go() mkdir -p $RESULT_BASE rm -f $RESULT_BASE/check.* + # Only supports default mkfs parameters right now + wipefs -a $TEST_DEV > /dev/null 2>&1 + yes | mkfs -t $FSTYP $TEST_MKFS_OPTS $TEST_DEV > /dev/null 2>&1 + # export DUMP_CORRUPT_FS=1 # Run the tests in it's own mount namespace, as per the comment below
Re: [PATCH 09/28] check-parallel: allow FSTYP selection from the CLI
Quoted 68 lines · Nirjhar Roy (IBM)
Quoted 67 lines · Dave Chinner
From: Dave Chinner <[email protected]>
Add a CLI option to specify the initial FSTYP to test. If this is not specified the the default of "xfs" will be used. This option is different to the way check has FSTYP specified as check-parallel has no infrastructure to support non block device based filesystems and hence we have to reject virtual or network based filesysetms are this point in time.
Note: This patch only implements default mkfs parameter support for the test device. Config sections can be used to override this as check will then format the test device when the section that defines non-default test device mkfs options is selected.
Signed-off-by Dave Chinner <[email protected]>--- check-parallel | 44 +++++++++++++++++++++++++++++++++++++------- 1 file changed, 37 insertions(+), 7 deletions(-)check-parallel +15
-5@@ -18,7 +18,7 @@ run_section="" tmp=/tmp/check-parallel.$$ -export FSTYP=xfs +FSTYP= . ./common/exit . ./common/test_names@@ -35,6 +35,7 @@ check options -r randomize test order --exact-order run tests in the exact order specified -s section run only specified section from config file + -f <FSTYPE> specify the filesystem type to test testlist options -g group[,group...] include tests from these groups@@ -66,16 +67,39 @@ external_file argument is a path to a single file containing a list of tests to exclude in the form of <test dir>/<test name>. examples: - check-parallel -D /mnt xfs/001 - check-parallel -D /mnt -g quick + check-parallel -f xfs -D /mnt xfs/001 + check-parallel -f ext4 -D /mnt -g quick check-parallel -D /mnt -g xfs/quick check-parallel -D /mnt -x stress xfs/* - check-parallel -D /mnt -X .exclude -g auto - check-parallel -D /mnt -E ~/.xfstests.exclude + check-parallel -f btrfs -D /mnt -X .exclude -g auto + check-parallel -f udf -D /mnt -E ~/.xfstests.exclude ' exit 1 } +# Only support block device based filesystems with generic mkfs support +# at the moment. +is_supported_fstype() +{ + local fstype=$1 + + case $fstype in + xfs) ;; + ext2|ext3|ext4) ;;
Quoted 8 lines
check-parallel external_file argument is a path to a single file containing a list of tests · line 95
Quoted 7 lines
check-parallel +7
-0+ udf) ;; + jfs) ;; + f2fs) ;; + btrfs) ;; + bcachefs) ;; + gfs2) ;; + ocfs2) ;;Extremely minor: Maybe list the test in alphabetical order?
I copied the list from a case statement somewhere in common/rc that was already in that order. I don't it's worth the time and brain cells to order it, let alone require anyone to maintain that order over the long term. It's just a simple string check....
-Dave.
Signature · 2 lines
Dave Chinner [email protected]
Re: [PATCH 09/28] check-parallel: allow FSTYP selection from the CLI
Quoted 77 lines · Dave Chinner
Quoted 68 lines · Nirjhar Roy (IBM)
Quoted 67 lines · Dave Chinner
From: Dave Chinner <[email protected]>
Add a CLI option to specify the initial FSTYP to test. If this is not specified the the default of "xfs" will be used. This option is different to the way check has FSTYP specified as check-parallel has no infrastructure to support non block device based filesystems and hence we have to reject virtual or network based filesysetms are this point in time.
Note: This patch only implements default mkfs parameter support for the test device. Config sections can be used to override this as check will then format the test device when the section that defines non-default test device mkfs options is selected.
Signed-off-by Dave Chinner <[email protected]>--- check-parallel | 44 +++++++++++++++++++++++++++++++++++++------- 1 file changed, 37 insertions(+), 7 deletions(-)check-parallel +15
-5@@ -18,7 +18,7 @@ run_section="" tmp=/tmp/check-parallel.$$ -export FSTYP=xfs +FSTYP= . ./common/exit . ./common/test_names@@ -35,6 +35,7 @@ check options -r randomize test order --exact-order run tests in the exact order specified -s section run only specified section from config file + -f <FSTYPE> specify the filesystem type to test testlist options -g group[,group...] include tests from these groups@@ -66,16 +67,39 @@ external_file argument is a path to a single file containing a list of tests to exclude in the form of <test dir>/<test name>. examples: - check-parallel -D /mnt xfs/001 - check-parallel -D /mnt -g quick + check-parallel -f xfs -D /mnt xfs/001 + check-parallel -f ext4 -D /mnt -g quick check-parallel -D /mnt -g xfs/quick check-parallel -D /mnt -x stress xfs/* - check-parallel -D /mnt -X .exclude -g auto - check-parallel -D /mnt -E ~/.xfstests.exclude + check-parallel -f btrfs -D /mnt -X .exclude -g auto + check-parallel -f udf -D /mnt -E ~/.xfstests.exclude ' exit 1 } +# Only support block device based filesystems with generic mkfs support +# at the moment. +is_supported_fstype() +{ + local fstype=$1 + + case $fstype in + xfs) ;; + ext2|ext3|ext4) ;;Quoted 8 lines
check-parallel external_file argument is a path to a single file containing a list of tests · line 95
Quoted 7 lines
check-parallel +7
-0+ udf) ;; + jfs) ;; + f2fs) ;; + btrfs) ;; + bcachefs) ;; + gfs2) ;; + ocfs2) ;;Extremely minor: Maybe list the test in alphabetical order?
Quoted 4 lines
I copied the list from a case statement somewhere in common/rc that was already in that order. I don't it's worth the time and brain cells to order it, let alone require anyone to maintain that order over the long term. It's just a simple string check....
Yes, that's fine.
--NR
Quoted 2 lines
-Dave.
Signature · 3 lines
Nirjhar Roy Linux Kernel Developer IBM, Bangalore
PATCH 10/28 check-parallel: use PID namespaces for runner process isolation
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
This provides isolation between individual runners so that they cannot see the processes that other test runners have created. This means tools like pkill will only find processes run by the test that is calling it, hence there is no danger that it might kill processes owned by a different test in a different runner context.
This means we need to turn of private pid/mount namespaces inside check itself - nesting private namespaces causes all sorts of weird TEST_DIR mount issues with SELinux disallowing TEST_DIR mounts.
Note that this also regresses generic/504 for check-parallel, because it triggers looking at the init namespace from FSTESTS_ISOL=privatens and this variable is not exported to test subprocesses anymore.
--- check-parallel | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-)
check-parallel +7 -3
@@ -267,14 +267,18 @@ runner_go() rm -f $RESULT_BASE/check.* # Only supports default mkfs parameters right now - wipefs -a $TEST_DEV > /dev/null 2>&1 - yes | mkfs -t $FSTYP $TEST_MKFS_OPTS $TEST_DEV > /dev/null 2>&1 + wipefs -a $TEST_DEV > $me/log 2>&1 + yes | mkfs -t $FSTYP $TEST_MKFS_OPTS $TEST_DEV >> $me/log 2>&1 # export DUMP_CORRUPT_FS=1 # Run the tests in it's own mount namespace, as per the comment below # that precedes making the basedir a private mount. - ./src/nsexec -m ./check $run_section -x unreliable_in_parallel --exact-order ${runner_list[$id]} > $me/log 2>&1 + # + # Similarly, we need to run check in it's own PID namespace so that + # operations like pkill only affect the runner instance, not globally + # kill processes from other check instances. + tools/run_privatens ./check $run_section -x unreliable_in_parallel --exact-order ${runner_list[$id]} >> $me/log 2>&1 wait sleep 1
Signature · 1 line
2.45.2
Re: [PATCH 10/28] check-parallel: use PID namespaces for runner process isolation
Quoted 35 lines · Dave Chinner
From: Dave Chinner <[email protected]>
This provides isolation between individual runners so that they cannot see the processes that other test runners have created. This means tools like pkill will only find processes run by the test that is calling it, hence there is no danger that it might kill processes owned by a different test in a different runner context.
This means we need to turn of private pid/mount namespaces inside check itself - nesting private namespaces causes all sorts of weird
TEST_DIRmount issues with SELinux disallowingTEST_DIRmounts.Note that this also regresses generic/504 for check-parallel, because it triggers looking at the init namespace from FSTESTS_ISOL=privatens and this variable is not exported to test subprocesses anymore.
Signed-off-by Dave Chinner <[email protected]>--- check-parallel | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-)check-parallel +2
-2@@ -267,14 +267,18 @@ runner_go() rm -f $RESULT_BASE/check.* # Only supports default mkfs parameters right now - wipefs -a $TEST_DEV > /dev/null 2>&1 - yes | mkfs -t $FSTYP $TEST_MKFS_OPTS $TEST_DEV > /dev/null 2>&1 + wipefs -a $TEST_DEV > $me/log 2>&1 + yes | mkfs -t $FSTYP $TEST_MKFS_OPTS $TEST_DEV >> $me/log 2>&1
check-parallel runner_go() · line 281
Quoted 10 lines
check-parallel +5
-1# export DUMP_CORRUPT_FS=1 # Run the tests in it's own mount namespace, as per the comment below # that precedes making the basedir a private mount. - ./src/nsexec -m ./check $run_section -x unreliable_in_parallel --exact-order ${runner_list[$id]} > $me/log 2>&1 + # + # Similarly, we need to run check in it's own PID namespace so that + # operations like pkill only affect the runner instance, not globally + # kill processes from other check instances. + tools/run_privatens ./check $run_section -x unreliable_in_parallel --exact-order ${runner_list[$id]} >> $me/log 2>&1
Why aren't we using -s $section and only using $section with check? --NR
Quoted 3 lines
check-parallel +0
-0wait sleep 1
Re: [PATCH 10/28] check-parallel: use PID namespaces for runner process isolation
Quoted 38 lines · Nirjhar Roy (IBM)
Quoted 37 lines · Dave Chinner
From: Dave Chinner <[email protected]>
This provides isolation between individual runners so that they cannot see the processes that other test runners have created. This means tools like pkill will only find processes run by the test that is calling it, hence there is no danger that it might kill processes owned by a different test in a different runner context.
This means we need to turn of private pid/mount namespaces inside check itself - nesting private namespaces causes all sorts of weird
TEST_DIRmount issues with SELinux disallowingTEST_DIRmounts.Note that this also regresses generic/504 for check-parallel, because it triggers looking at the init namespace from FSTESTS_ISOL=privatens and this variable is not exported to test subprocesses anymore.
Signed-off-by Dave Chinner <[email protected]>--- check-parallel | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-)check-parallel +2
-2@@ -267,14 +267,18 @@ runner_go() rm -f $RESULT_BASE/check.* # Only supports default mkfs parameters right now - wipefs -a $TEST_DEV > /dev/null 2>&1 - yes | mkfs -t $FSTYP $TEST_MKFS_OPTS $TEST_DEV > /dev/null 2>&1 + wipefs -a $TEST_DEV > $me/log 2>&1 + yes | mkfs -t $FSTYP $TEST_MKFS_OPTS $TEST_DEV >> $me/log 2>&1 # export DUMP_CORRUPT_FS=1
Quoted 9 lines
check-parallel runner_go() · line 281
Quoted 8 lines
check-parallel +5
-1# Run the tests in it's own mount namespace, as per the comment below # that precedes making the basedir a private mount. - ./src/nsexec -m ./check $run_section -x unreliable_in_parallel --exact-order ${runner_list[$id]} > $me/log 2>&1 + # + # Similarly, we need to run check in it's own PID namespace so that + # operations like pkill only affect the runner instance, not globally + # kill processes from other check instances. + tools/run_privatens ./check $run_section -x unreliable_in_parallel --exact-order ${runner_list[$id]} >> $me/log 2>&1Why aren't we using -s $section and only using $section with check?
It is already encoded in $run_section - the CLI parsing adds the -s for each section that is specified on the command line:
-s) run_section="$run_section -s $2"; shift ;;
-Dave.
Signature · 2 lines
Dave Chinner [email protected]
Re: [PATCH 10/28] check-parallel: use PID namespaces for runner process isolation
Quoted 48 lines · Dave Chinner
Quoted 38 lines · Nirjhar Roy (IBM)
Quoted 37 lines · Dave Chinner
From: Dave Chinner <[email protected]>
This provides isolation between individual runners so that they cannot see the processes that other test runners have created. This means tools like pkill will only find processes run by the test that is calling it, hence there is no danger that it might kill processes owned by a different test in a different runner context.
This means we need to turn of private pid/mount namespaces inside check itself - nesting private namespaces causes all sorts of weird
TEST_DIRmount issues with SELinux disallowingTEST_DIRmounts.Note that this also regresses generic/504 for check-parallel, because it triggers looking at the init namespace from FSTESTS_ISOL=privatens and this variable is not exported to test subprocesses anymore.
Signed-off-by Dave Chinner <[email protected]>--- check-parallel | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-)check-parallel +2
-2@@ -267,14 +267,18 @@ runner_go() rm -f $RESULT_BASE/check.* # Only supports default mkfs parameters right now - wipefs -a $TEST_DEV > /dev/null 2>&1 - yes | mkfs -t $FSTYP $TEST_MKFS_OPTS $TEST_DEV > /dev/null 2>&1 + wipefs -a $TEST_DEV > $me/log 2>&1 + yes | mkfs -t $FSTYP $TEST_MKFS_OPTS $TEST_DEV >> $me/log 2>&1 # export DUMP_CORRUPT_FS=1Quoted 9 lines
check-parallel runner_go() · line 281
Quoted 8 lines
check-parallel +5
-1# Run the tests in it's own mount namespace, as per the comment below # that precedes making the basedir a private mount. - ./src/nsexec -m ./check $run_section -x unreliable_in_parallel --exact-order ${runner_list[$id]} > $me/log 2>&1 + # + # Similarly, we need to run check in it's own PID namespace so that + # operations like pkill only affect the runner instance, not globally + # kill processes from other check instances. + tools/run_privatens ./check $run_section -x unreliable_in_parallel --exact-order ${runner_list[$id]} >> $me/log 2>&1Why aren't we using -s $section and only using $section with check?
Quoted 4 lines
It is already encoded in $run_section - the CLI parsing adds the -s for each section that is specified on the command line:
-s) run_section="$run_section -s $2"; shift ;;
Yeah, right. I realized that later. Thanks
-NR
Quoted 3 lines
-Dave.
Signature · 3 lines
Nirjhar Roy Linux Kernel Developer IBM, Bangalore
PATCH 11/28 check-parallel: initial support for specifying device sizes
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
Rather than hard coding loop device sizes, abstract them out into environment variables with defined default values. This allows a check-parallel wrapper to specify sizes or, in future, for them to be read from a config file. Sizes are specified in human readable values using M/G/T suffixes to indicate the units being specified
Whilst doing this also add support for creating all the external devices that XFS uses during fstests execution. A typical setup will be something like:
TEST_DEV_SIZE=10G
TEST_RTDEV_SIZE=10G
TEST_LOGDEV_SIZE=128M
SCRATCH_DEV_SIZE=20G
SCRATCH_RTDEV_SIZE=20G
SCRATCH_LOGDEV_SIZE=512M
LOGWRITES_DEV_SIZE=2G
--- check-parallel | 35 ++++++++++++++++++++++++++++++----- 1 file changed, 30 insertions(+), 5 deletions(-)
check-parallel +30 -5
@@ -18,6 +18,14 @@ run_section="" tmp=/tmp/check-parallel.$$ +TEST_DEV_SIZE=${TEST_DEV_SIZE:=10G} +TEST_RTDEV_SIZE=${TEST_RTDEV_SIZE:=10G} +TEST_LOGDEV_SIZE=${TEST_LOGDEV_SIZE:=128M} +SCRATCH_DEV_SIZE=${SCRATCH_DEV_SIZE:=20G} +SCRATCH_RTDEV_SIZE=${SCRATCH_RTDEV_SIZE:=20G} +SCRATCH_LOGDEV_SIZE=${SCRATCH_LOGDEV_SIZE:=512M} +LOGWRITES_DEV_SIZE=${LOGWRITES_DEV_SIZE:=2G} + FSTYP= . ./common/exit
@@ -116,7 +124,7 @@ while [ $# -gt 0 ]; do --exact-order) _tl_setup_ordered ;; -n) show_test_list="yes" ;; - -f) is_supported_fstype $2 ; FSTYP=$2; shift ;; + -f) is_supported_fstype $2 ; export FSTYP=$2; shift ;; -s) run_section="$run_section -s $2"; shift ;;
@@ -243,22 +251,35 @@ runner_go() local id=$1 local me=$basedir/runner-$id local _test=$me/test.img + local _test_rt=$me/test-rt.img + local _test_log=$me/test-log.img local _scratch=$me/scratch.img + local _scratch_rt=$me/scratch-rt.img + local _scratch_log=$me/scratch-log.img local _logwrites=$me/logwrites.img local _results=$me/results-$2 mkdir -p $me - xfs_io -f -c 'truncate 2g' $_test - xfs_io -f -c 'truncate 8g' $_scratch - xfs_io -f -c 'truncate 1g' $_logwrites + xfs_io -f -c "truncate $TEST_DEV_SIZE" $_test + xfs_io -f -c "truncate $TEST_RTDEV_SIZE" $_test_rt + xfs_io -f -c "truncate $TEST_LOGDEV_SIZE" $_test_log + xfs_io -f -c "truncate $SCRATCH_DEV_SIZE" $_scratch + xfs_io -f -c "truncate $SCRATCH_RTDEV_SIZE" $_scratch_rt + xfs_io -f -c "truncate $SCRATCH_LOGDEV_SIZE" $_scratch_log + xfs_io -f -c "truncate $LOGWRITES_DEV_SIZE" $_logwrites export TEST_DEV=$(_create_loop_device $_test) + export TEST_RTDEV=$(_create_loop_device $_test_rt) + export TEST_LOGDEV=$(_create_loop_device $_test_log) export TEST_DIR=$me/test + export SCRATCH_DEV=$(_create_loop_device $_scratch) + export SCRATCH_RTDEV=$(_create_loop_device $_scratch_rt) + export SCRATCH_LOGDEV=$(_create_loop_device $_scratch_log) export SCRATCH_MNT=$me/scratch + export LOGWRITES_DEV=$(_create_loop_device $_logwrites) - export FSTYP=xfs export RESULT_BASE=$_results mkdir -p $TEST_DIR
@@ -285,7 +306,11 @@ runner_go() umount -R $TEST_DIR 2> /dev/null umount -R $SCRATCH_MNT 2> /dev/null _destroy_loop_device $TEST_DEV + _destroy_loop_device $TEST_RTDEV + _destroy_loop_device $TEST_LOGDEV _destroy_loop_device $SCRATCH_DEV + _destroy_loop_device $SCRATCH_RTDEV + _destroy_loop_device $SCRATCH_LOGDEV _destroy_loop_device $LOGWRITES_DEV grep -q Failures: $me/log
Signature · 1 line
2.45.2
Re: [PATCH 11/28] check-parallel: initial support for specifying device sizes
Quoted 12 lines · Dave Chinner
From: Dave Chinner <[email protected]>
Rather than hard coding loop device sizes, abstract them out into environment variables with defined default values. This allows a check-parallel wrapper to specify sizes or, in future, for them to be read from a config file. Sizes are specified in human readable values using M/G/T suffixes to indicate the units being specified
Whilst doing this also add support for creating all the external devices that XFS uses during fstests execution. A typical setup will be something like:
Quoted 7 lines
TEST_DEV_SIZE=10G
TEST_RTDEV_SIZE=10G
TEST_LOGDEV_SIZE=128M
SCRATCH_DEV_SIZE=20G
SCRATCH_RTDEV_SIZE=20G
SCRATCH_LOGDEV_SIZE=512M
LOGWRITES_DEV_SIZE=2G
Should we document these new variables in README?
The functionality looks fine to me. I have some refactoring suggestions below (nothing functional).
Quoted 48 lines
Signed-off-by Dave Chinner <[email protected]>--- check-parallel | 35 ++++++++++++++++++++++++++++++----- 1 file changed, 30 insertions(+), 5 deletions(-)check-parallel +13
-1@@ -18,6 +18,14 @@ run_section="" tmp=/tmp/check-parallel.$$ +TEST_DEV_SIZE=${TEST_DEV_SIZE:=10G} +TEST_RTDEV_SIZE=${TEST_RTDEV_SIZE:=10G} +TEST_LOGDEV_SIZE=${TEST_LOGDEV_SIZE:=128M} +SCRATCH_DEV_SIZE=${SCRATCH_DEV_SIZE:=20G} +SCRATCH_RTDEV_SIZE=${SCRATCH_RTDEV_SIZE:=20G} +SCRATCH_LOGDEV_SIZE=${SCRATCH_LOGDEV_SIZE:=512M} +LOGWRITES_DEV_SIZE=${LOGWRITES_DEV_SIZE:=2G} + FSTYP= . ./common/exit@@ -116,7 +124,7 @@ while [ $# -gt 0 ]; do --exact-order) _tl_setup_ordered ;; -n) show_test_list="yes" ;; - -f) is_supported_fstype $2 ; FSTYP=$2; shift ;; + -f) is_supported_fstype $2 ; export FSTYP=$2; shift ;; -s) run_section="$run_section -s $2"; shift ;;@@ -243,22 +251,35 @@ runner_go() local id=$1 local me=$basedir/runner-$id local _test=$me/test.img + local _test_rt=$me/test-rt.img + local _test_log=$me/test-log.img local _scratch=$me/scratch.img + local _scratch_rt=$me/scratch-rt.img + local _scratch_log=$me/scratch-log.img local _logwrites=$me/logwrites.img local _results=$me/results-$2 mkdir -p $me
check-parallel runner_go() · line 270
Quoted 10 lines
check-parallel +7
-3- xfs_io -f -c 'truncate 2g' $_test - xfs_io -f -c 'truncate 8g' $_scratch - xfs_io -f -c 'truncate 1g' $_logwrites + xfs_io -f -c "truncate $TEST_DEV_SIZE" $_test + xfs_io -f -c "truncate $TEST_RTDEV_SIZE" $_test_rt + xfs_io -f -c "truncate $TEST_LOGDEV_SIZE" $_test_log + xfs_io -f -c "truncate $SCRATCH_DEV_SIZE" $_scratch + xfs_io -f -c "truncate $SCRATCH_RTDEV_SIZE" $_scratch_rt + xfs_io -f -c "truncate $SCRATCH_LOGDEV_SIZE" $_scratch_log + xfs_io -f -c "truncate $LOGWRITES_DEV_SIZE" $_logwrites
Do you like the following refactoring:
declare -a devices_map=(
"$TEST_DEV_SIZE:$_test:TEST_DEV" "$TEST_RTDEV_SIZE:$_test_rt:TEST_RTDEV" "$TEST_LOGDEV_SIZE:$_test_log:TEST_LOGDEV" "$SCRATCH_DEV_SIZE:$_scratch:SCRATCH_DEV" "$SCRATCH_RTDEV_SIZE:$_scratch_rt:SCRATCH_RTDEV" "$SCRATCH_LOGDEV_SIZE:$_scratch_log:SCRATCH_LOGDEV" "$LOGWRITES_DEV_SIZE:$_logwrites:LOGWRITES_DEV"
)create_dev_images devices_map
create_dev_images() {
# this executes commands like
# xfs_io -f -c "truncate 10G <test-file.img>"
# for all the devices
local -n dev_list="$1"
for entry in "${dev_list[@]}"; do
IFS=':' read -r -a tuple <<< "$entry"
xfs_io -f -c "truncate ${tuple[0]} ${tuple[1]}"
done
}check-parallel runner_go() · line 282
Quoted 12 lines
check-parallel +6
-0export TEST_DEV=$(_create_loop_device $_test) + export TEST_RTDEV=$(_create_loop_device $_test_rt) + export TEST_LOGDEV=$(_create_loop_device $_test_log) export TEST_DIR=$me/test + export SCRATCH_DEV=$(_create_loop_device $_scratch) + export SCRATCH_RTDEV=$(_create_loop_device $_scratch_rt) + export SCRATCH_LOGDEV=$(_create_loop_device $_scratch_log) export SCRATCH_MNT=$me/scratch + export LOGWRITES_DEV=$(_create_loop_device $_logwrites)
Something similar as above:
create_devices()
{
# This exports all the device related environment variables
# like TEST_DEV, SCRATCH_DEV etc.
local -n dev_list="$1"
IFS=':' read -r -a tuple <<< "$entry"
eval "export ${tuple[2]}=\$(_create_loop_device ${tuple[1]})"}
and then simply callcreate_devices devices_map
export TEST_DIR=$me/test
export SCRATCH_MNT=$me/scratch
Quoted 5 lines
check-parallel +0
-1- export FSTYP=xfs export RESULT_BASE=$_results mkdir -p $TEST_DIR
check-parallel runner_go() · line 314
Quoted 9 lines
check-parallel +4
-0@@ -285,7 +306,11 @@ runner_go() umount -R $TEST_DIR 2> /dev/null umount -R $SCRATCH_MNT 2> /dev/null _destroy_loop_device $TEST_DEV + _destroy_loop_device $TEST_RTDEV + _destroy_loop_device $TEST_LOGDEV _destroy_loop_device $SCRATCH_DEV + _destroy_loop_device $SCRATCH_RTDEV + _destroy_loop_device $SCRATCH_LOGDEV _destroy_loop_device $LOGWRITES_DEV
destroy_devices()
{
# This destroys all the loop devices that were created.
local -n dev_list="$1"
IFS=':' read -r -a tuple <<< "$entry"
eval "_destroy_loop_device \$${tuple[2]}"
}and then calldestroy_devices devices_map.
IMO, this reduces the number of lines of code and adding an entry to devices_map automatically does the devices creation, removal and setting of environment variables without having to manually add them. This is just a suggestion - please let me know what do you think about this.
--NR
Quoted 2 lines
check-parallel +0
-0grep -q Failures: $me/log
Re: [PATCH 11/28] check-parallel: initial support for specifying device sizes
Quoted 13 lines · Nirjhar Roy (IBM)
Quoted 12 lines · Dave Chinner
From: Dave Chinner <[email protected]>
Rather than hard coding loop device sizes, abstract them out into environment variables with defined default values. This allows a check-parallel wrapper to specify sizes or, in future, for them to be read from a config file. Sizes are specified in human readable values using M/G/T suffixes to indicate the units being specified
Whilst doing this also add support for creating all the external devices that XFS uses during fstests execution. A typical setup will be something like:
Quoted 8 lines
Quoted 7 lines
TEST_DEV_SIZE=10G
TEST_RTDEV_SIZE=10G
TEST_LOGDEV_SIZE=128M
SCRATCH_DEV_SIZE=20G
SCRATCH_RTDEV_SIZE=20G
SCRATCH_LOGDEV_SIZE=512M
LOGWRITES_DEV_SIZE=2GShould we document these new variables in README?
Not at this point in time. This patch is just introducing the code to create all the necessary devices for a runner, the sizes of which need to be specified somewhere for it to work.
....
Quoted 38 lines
Quoted 14 lines
check-parallel +4
-0@@ -243,22 +251,35 @@ runner_go() local id=$1 local me=$basedir/runner-$id local _test=$me/test.img + local _test_rt=$me/test-rt.img + local _test_log=$me/test-log.img local _scratch=$me/scratch.img + local _scratch_rt=$me/scratch-rt.img + local _scratch_log=$me/scratch-log.img local _logwrites=$me/logwrites.img local _results=$me/results-$2 mkdir -p $mecheck-parallel runner_go() · line 270
Quoted 10 lines
check-parallel +7
-3- xfs_io -f -c 'truncate 2g' $_test - xfs_io -f -c 'truncate 8g' $_scratch - xfs_io -f -c 'truncate 1g' $_logwrites + xfs_io -f -c "truncate $TEST_DEV_SIZE" $_test + xfs_io -f -c "truncate $TEST_RTDEV_SIZE" $_test_rt + xfs_io -f -c "truncate $TEST_LOGDEV_SIZE" $_test_log + xfs_io -f -c "truncate $SCRATCH_DEV_SIZE" $_scratch + xfs_io -f -c "truncate $SCRATCH_RTDEV_SIZE" $_scratch_rt + xfs_io -f -c "truncate $SCRATCH_LOGDEV_SIZE" $_scratch_log + xfs_io -f -c "truncate $LOGWRITES_DEV_SIZE" $_logwritesDo you like the following refactoring:
declare -a devices_map=(
"$TEST_DEV_SIZE:$_test:TEST_DEV" "$TEST_RTDEV_SIZE:$_test_rt:TEST_RTDEV" "$TEST_LOGDEV_SIZE:$_test_log:TEST_LOGDEV" "$SCRATCH_DEV_SIZE:$_scratch:SCRATCH_DEV" "$SCRATCH_RTDEV_SIZE:$_scratch_rt:SCRATCH_RTDEV" "$SCRATCH_LOGDEV_SIZE:$_scratch_log:SCRATCH_LOGDEV" "$LOGWRITES_DEV_SIZE:$_logwrites:LOGWRITES_DEV")
create_dev_imagesdevices_map
Quoted 10 lines
create_dev_images() {
# this executes commands like
# xfs_io -f -c "truncate 10G <test-file.img>"
# for all the devices
local -n dev_list="$1"
for entry in "${dev_list[@]}"; do
IFS=':' read -r -a tuple <<< "$entry"
xfs_io -f -c "truncate ${tuple[0]} ${tuple[1]}"
done
}My brain hurts just looking at that. That's way too clever and complex for a dumb engineer like myself - it took me several minutes to sorta work out what it does and how it works.
However, I'm still not sure how it works, because the image file variables have different values for every runner context. i.e. we'd need a map per runner, not a single global map. I don't know if that variable is evaluated at declaration time, when it is read into the array, when the array variable is evaluated, how that interacts with background execution, etc. I'm not sure I can work that out without help of the bash man page and watching the output when 'set -x' has been issued.
Compared to reading a bunch of truncate commands in a couple of seconds and being certain the code is doing what it should be doing, adding a device map like this is adding a lot of extra cognitive load and analysis every time I look at this code.
IMO, there's a good reason for writing code that is dumb as a
hammer: everyone knows how to use a hammer. The KISS principle is everyone's friend...
Quoted 18 lines
check-parallel runner_go() · line 314
Quoted 10 lines
check-parallel +4
-0@@ -285,7 +306,11 @@ runner_go() umount -R $TEST_DIR 2> /dev/null umount -R $SCRATCH_MNT 2> /dev/null _destroy_loop_device $TEST_DEV + _destroy_loop_device $TEST_RTDEV + _destroy_loop_device $TEST_LOGDEV _destroy_loop_device $SCRATCH_DEV + _destroy_loop_device $SCRATCH_RTDEV + _destroy_loop_device $SCRATCH_LOGDEV _destroy_loop_device $LOGWRITES_DEV
destroy_devices(){ # This destroys all the loop devices that were created. local -n dev_list="$1" IFS=':' read -r -a tuple <<< "$entry" eval "_destroy_loop_device \$${tuple[2]}" }
Quoted 2 lines
and then call
destroy_devicesdevices_map.
Once everything has settled, I'll be replacing the loop destroy code with a single call to 'losetup -D' from the check-parallel cleanup trap. This will detatch all the image files from all the loop devices in a single command, rather than doing them one at a time in every runner context.
Quoted 5 lines
IMO, this reduces the number of lines of code and adding an entry to
devices_mapautomatically does the devices creation, removal and setting of environment variables without having to manually add them. This is just a suggestion - please let me know what do you think about this.
I can't tell if it works or not without spending a lot more time trying to understand it. I've already spent way too much time trying to understand it well enough to write a coherent response.
I think that about sums it up....
-Dave.
Signature · 2 lines
Dave Chinner [email protected]
PATCH 12/28 config: move config section code to it's own file
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
Move the config section parsing, checking and setup code from common/config to common/config-section so that it can be included directly in contexts where the rest of common/config is not needed.
--- common/config | 382 +--------------------------------------- common/config-sections | 390 +++++++++++++++++++++++++++++++++++++++++ 2 files changed, 392 insertions(+), 380 deletions(-) create mode 100644 common/config-sections
common/config +2 -380
@@ -41,6 +41,7 @@ . common/test_names . common/exit +. common/config-sections # all tests should use a common language setting to prevent golden # output mismatches.
@@ -544,386 +545,7 @@ _source_specific_fs() esac } -known_hosts() -{ - [ "$HOST_CONFIG_DIR" ] || HOST_CONFIG_DIR=`pwd`/configs - - [ -f /etc/xfsqa.config ] && export HOST_OPTIONS=/etc/xfsqa.config - [ -f $HOST_CONFIG_DIR/$HOST ] && export HOST_OPTIONS=$HOST_CONFIG_DIR/$HOST - [ -f $HOST_CONFIG_DIR/$HOST.config ] && export HOST_OPTIONS=$HOST_CONFIG_DIR/$HOST.config -} - -# Returns a list of sections in config file -# Each section starts with the section name in the format -# [section_name1]. Only alphanumeric characters and '_' is allowed -# in the section name otherwise the section will not be resognised. -# Section name must be contained between square brackets. -get_config_sections() { - sed -n -e "s/^\[\([[:alnum:]_-]*\)\]/\1/p" < $1 -} - -if [ ! -f "$HOST_OPTIONS" ]; then - known_hosts -fi - -export HOST_OPTIONS_SECTIONS="-no-sections-" -export OPTIONS_HAVE_SECTIONS=false -if [ -f "$HOST_OPTIONS" ]; then - export HOST_OPTIONS_SECTIONS=`get_config_sections $HOST_OPTIONS` - if [ -z "$HOST_OPTIONS_SECTIONS" ]; then - . $HOST_OPTIONS - export HOST_OPTIONS_SECTIONS="-no-sections-" - else - export OPTIONS_HAVE_SECTIONS=true - fi -fi - -_check_device() -{ - local name=$1 - local dev_needed=$2 - local dev=$3 - - if [ -z "$dev" ]; then - if [ "$dev_needed" == "required" ]; then - _fatal "common/config: $name is required but not defined!" - fi - return 0 - fi - - if [ -b "$dev" ] || ( echo $dev | grep -qE ":|//" ); then - # block device or a network url - return 0 - fi - - case "$FSTYP" in - 9p|fuse|tmpfs|virtiofs|afs) - # 9p, fuse, virtiofs and afs mount tags are just plain strings, - # so anything is allowed tmpfs doesn't use mount source, ignore - ;; - ceph) - # ceph has two different possible syntaxes for mount devices. The - # network URL check above catches the legacy syntax. Check for the - # new-style syntax here. - if ( echo $dev | grep -qEv "=/" ); then - _fatal "common/config: $name ($dev) is not a valid ceph mount string" - fi - ;; - overlay) - if [ ! -d "$dev" ]; then - _fatal "common/config: $name ($dev) is not a directory for overlay" - fi - ;; - ubifs) - if [ ! -c "$dev" ]; then - _fatal "common/config: $name ($dev) is not a character device" - fi - ;; - ceph-fuse) - ;; - *) - _fatal "common/config: $name ($dev) is not a block device or a network filesystem" - esac -} - -# check and return a canonical mount point path -_canonicalize_mountpoint() -{ - local name=$1 - local dir=$2 - - if [ -d "$dir" ]; then - # this follows symlinks and removes all trailing "/"s - readlink -e "$dir" - return 0 - fi - - if [ "$FSTYP" != "overlay" ] || [[ "$name" == OVL_BASE_* ]]; then - _fatal "common/config: $name ($dir) is not a directory" - fi - - # base fs may not be mounted yet, so just check that parent dir - # exists (where base fs will be mounted) because we are going to - # mkdir the overlay mount point dir anyway - local base=`basename $dir` - local parent=`dirname $dir` - parent=`_canonicalize_mountpoint OVL_BASE_$name "$parent"` - - # prepend the overlay mount point to canonical parent path - echo "$parent/$base" -} - -# Enables usage of /dev/disk/by-id/ symlinks to persist target devices -# over reboots -_canonicalize_devices() -{ - if [ "$CANON_DEVS" != "yes" ]; then - return - fi - [ -L "$TEST_DEV" ] && TEST_DEV=$(readlink -e "$TEST_DEV") - [ -L "$SCRATCH_DEV" ] && SCRATCH_DEV=$(readlink -e "$SCRATCH_DEV") - [ -L "$TEST_LOGDEV" ] && TEST_LOGDEV=$(readlink -e "$TEST_LOGDEV") - [ -L "$TEST_RTDEV" ] && TEST_RTDEV=$(readlink -e "$TEST_RTDEV") - [ -L "$SCRATCH_RTDEV" ] && SCRATCH_RTDEV=$(readlink -e "$SCRATCH_RTDEV") - [ -L "$LOGWRITES_DEV" ] && LOGWRITES_DEV=$(readlink -e "$LOGWRITES_DEV") - if [ ! -z "$SCRATCH_DEV_POOL" ]; then - local NEW_SCRATCH_POOL="" - for i in $SCRATCH_DEV_POOL; do - if [ -L $i ]; then - NEW_SCRATCH_POOL="$NEW_SCRATCH_POOL $(readlink -e $i)" - else - NEW_SCRATCH_POOL="$NEW_SCRATCH_POOL $i" - fi - done - SCRATCH_DEV_POOL="$NEW_SCRATCH_POOL" - fi -} - -# On check -overlay, for the non multi section config case, this -# function is called on every test, before init_rc(). -# When SCRATCH/TEST_* vars are defined in config file, config file -# is sourced on every test and this function overrides the vars -# every time. -# When SCRATCH/TEST_* vars are defined in evironment and not -# in config file, this function is called after vars have already -# been overriden in the previous test. -# In that case, TEST_DEV is a directory and not a blockdev/chardev and -# the function will return without overriding the SCRATCH/TEST_* vars. -_overlay_config_override() -{ - # There are 2 options for configuring overlayfs tests: - # - # 1. (legacy) SCRATCH/TEST_DEV point to existing directories - # on an already mounted fs. In this case, the new - # OVL_BASE_SCRATCH/TEST_* vars are set to use the legacy - # vars values (even though they may not be mount points). - # - [ ! -d "$TEST_DEV" ] || export OVL_BASE_TEST_DIR="$TEST_DEV" - [ ! -d "$SCRATCH_DEV" ] || export OVL_BASE_SCRATCH_MNT="$SCRATCH_DEV" - - # Config file may specify base fs type, but we obay -overlay flag - [ "$FSTYP" == overlay ] || export OVL_BASE_FSTYP="$FSTYP" - export FSTYP=overlay - - # 2. SCRATCH/TEST_DEV point to the base fs partitions. In this case, - # the new OVL_BASE_SCRATCH/TEST_DEV/MNT vars are set to the values - # of the configured base fs and SCRATCH/TEST_DEV vars are set to the - # overlayfs base and mount dirs inside base fs mount. - [ -b "$TEST_DEV" ] || [ -c "$TEST_DEV" ] || [ "$OVL_BASE_FSTYP" == tmpfs ] || return 0 - - # Store original base fs vars - export OVL_BASE_TEST_DEV="$TEST_DEV" - export OVL_BASE_TEST_DIR="$TEST_DIR" - # If config does not set MOUNT_OPTIONS, its value may be - # leftover from previous _overlay_config_override, so - # don't use that value for base fs mount - [ "$MOUNT_OPTIONS" != "$OVERLAY_MOUNT_OPTIONS" ] || unset MOUNT_OPTIONS - export OVL_BASE_MOUNT_OPTIONS="$MOUNT_OPTIONS" - - # Set TEST vars to overlay base and mount dirs inside base fs - export TEST_DEV="$OVL_BASE_TEST_DIR" - export TEST_DIR="$OVL_BASE_TEST_DIR/$OVL_MNT" - export MOUNT_OPTIONS="$OVERLAY_MOUNT_OPTIONS" - - [ -b "$SCRATCH_DEV" ] || [ -c "$SCRATCH_DEV" ] || [ "$OVL_BASE_FSTYP" == tmpfs ] || return 0 - - # Store original base fs vars - export OVL_BASE_SCRATCH_DEV="$SCRATCH_DEV" - export OVL_BASE_SCRATCH_MNT="$SCRATCH_MNT" - - # Set SCRATCH vars to overlay base and mount dirs inside base fs - export SCRATCH_DEV="$OVL_BASE_SCRATCH_MNT" - export SCRATCH_MNT="$OVL_BASE_SCRATCH_MNT/$OVL_MNT" - - # Set fsck options, use default if user not set directly. - export FSCK_OPTIONS="$OVERLAY_FSCK_OPTIONS" - [ -z "$FSCK_OPTIONS" ] && _fsck_opts - export IDMAPPED_MOUNTS="$IDMAPPED_MOUNTS" -} - -_overlay_config_restore() -{ - export OVERLAY=true - [ -z "$OVL_BASE_FSTYP" ] || export FSTYP=$OVL_BASE_FSTYP - [ -z "$OVL_BASE_TEST_DEV" ] || export TEST_DEV=$OVL_BASE_TEST_DEV - [ -z "$OVL_BASE_TEST_DIR" ] || export TEST_DIR=$OVL_BASE_TEST_DIR - [ -z "$OVL_BASE_SCRATCH_DEV" ] || export SCRATCH_DEV=$OVL_BASE_SCRATCH_DEV - [ -z "$OVL_BASE_SCRATCH_MNT" ] || export SCRATCH_MNT=$OVL_BASE_SCRATCH_MNT - [ -z "$OVL_BASE_MOUNT_OPTIONS" ] || export MOUNT_OPTIONS=$OVL_BASE_MOUNT_OPTIONS -} - -# Parse config section options. This function will parse all the configuration -# within a single section which name is passed as an argument. For section -# name format see comments in get_config_sections(). -# Empty lines and everything after '#' will be ignored. -# Configuration options should be defined in the format -# -# CONFIG_OPTION=value -# -# This 'CONFIG_OPTION' variable and will be exported as an environment variable. -parse_config_section() { - SECTION=$1 - if ! $OPTIONS_HAVE_SECTIONS; then - return 0 - fi - eval `sed -e 's/[[:space:]]*\=[[:space:]]*/=/g' \ - -e 's/#.*$//' \ - -e 's/[[:space:]]*$//' \ - -e 's/^[[:space:]]*//' \ - -e "s/^\([^=]*\)=\"\?'\?\([^\"']*\)\"\?'\?$/export \1=\"\2\"/" \ - < $HOST_OPTIONS \ - | sed -n -e "/^\[$SECTION\]/,/^\s*\[/{/^[^#].*\=.*/p;}"` -} - -get_next_config() { - if [ ! -z "$CONFIG_INCLUDED" ] && ! $OPTIONS_HAVE_SECTIONS; then - return 0 - fi - - # We might have overriden FSTYP and TEST/SCRATCH vars with overlay values - # in the previous section, so restore them to original values stored in - # OVL_BASE_*. - # We need to do this *before* old FSTYP and MOUNT_OPTIONS are recorded - # and *before* SCRATCH_DEV and MOUNT_OPTIONS are unset - if [ "$FSTYP" == "overlay" ]; then - _overlay_config_restore - fi - - local OLD_FSTYP=$FSTYP - local OLD_MOUNT_OPTIONS=$MOUNT_OPTIONS - local OLD_TEST_FS_MOUNT_OPTS=$TEST_FS_MOUNT_OPTS - local OLD_MKFS_OPTIONS=$MKFS_OPTIONS - local OLD_FSCK_OPTIONS=$FSCK_OPTIONS - local OLD_USE_EXTERNAL=$USE_EXTERNAL - - unset MOUNT_OPTIONS - unset TEST_FS_MOUNT_OPTS - unset MKFS_OPTIONS - unset FSCK_OPTIONS - unset USE_EXTERNAL - - # We might have deduced SCRATCH_DEV from the SCRATCH_DEV_POOL in the previous - # run, so we have to unset it now. - if [ "$SCRATCH_DEV_NOT_SET" == "true" ]; then - unset SCRATCH_DEV - fi - - parse_config_section $1 - if [ ! -z "$OLD_FSTYP" ] && [ $OLD_FSTYP != $FSTYP ]; then - [ -z "$MOUNT_OPTIONS" ] && _mount_opts - [ -z "$TEST_FS_MOUNT_OPTS" ] && _test_mount_opts - [ -z "$MKFS_OPTIONS" ] && _mkfs_opts - [ -z "$FSCK_OPTIONS" ] && _fsck_opts - - # clear the external devices if we are not using them - if [ -z "$USE_EXTERNAL" ]; then - unset TEST_RTDEV - unset TEST_LOGDEV - unset SCRATCH_RTDEV - unset SCRATCH_LOGDEV - fi - else - [ -z "$MOUNT_OPTIONS" ] && export MOUNT_OPTIONS=$OLD_MOUNT_OPTIONS - [ -z "$TEST_FS_MOUNT_OPTS" ] && export TEST_FS_MOUNT_OPTS=$OLD_TEST_FS_MOUNT_OPTS - [ -z "$MKFS_OPTIONS" ] && export MKFS_OPTIONS=$OLD_MKFS_OPTIONS - [ -z "$FSCK_OPTIONS" ] && export FSCK_OPTIONS=$OLD_FSCK_OPTIONS - [ -z "$USE_EXTERNAL" ] && export USE_EXTERNAL=$OLD_USE_EXTERNAL - fi - - # set default RESULT_BASE - if [ -z "$RESULT_BASE" ]; then - export RESULT_BASE="$here/results/" - fi - - if [ "$FSTYP" == "tmpfs" ]; then - if [ -z "$TEST_DEV" ]; then - export TEST_DEV=tmpfs_test - fi - if [ -z "$SCRATCH_DEV" ]; then - export TEST_DEV=tmpfs_scratch - fi - fi - - # Mandatory Config values. - MC="" - [ -z "$EMAIL" ] && MC="$MC EMAIL" - [ -z "$TEST_DIR" ] && MC="$MC TEST_DIR" - [ -z "$TEST_DEV" ] && MC="$MC TEST_DEV" - - if [ -n "$MC" ]; then - echo "Warning: need to define parameters for host $HOST" - echo " or set variables:" - echo " $MC" - _exit 1 - fi - - _check_device TEST_DEV required $TEST_DEV - export TEST_DIR=`_canonicalize_mountpoint TEST_DIR $TEST_DIR` - - # a btrfs tester will set only SCRATCH_DEV_POOL, we will put first of its dev - # to SCRATCH_DEV and rest to SCRATCH_DEV_POOL to maintain the backward compatibility - if [ ! -z "$SCRATCH_DEV_POOL" ]; then - if [ ! -z "$SCRATCH_DEV" ]; then - echo "common/config: Error: \$SCRATCH_DEV ($SCRATCH_DEV) should be unset when \$SCRATCH_DEV_POOL ($SCRATCH_DEV_POOL) is set" - _exit 1 - fi - SCRATCH_DEV=`echo $SCRATCH_DEV_POOL | awk '{print $1}'` - export SCRATCH_DEV - export SCRATCH_DEV_NOT_SET=true - fi - - _check_device SCRATCH_DEV optional $SCRATCH_DEV - export SCRATCH_MNT=`_canonicalize_mountpoint SCRATCH_MNT $SCRATCH_MNT` - - if [ -n "$USE_EXTERNAL" ]; then - _check_device TEST_RTDEV optional $TEST_RTDEV - _check_device TEST_LOGDEV optional $TEST_LOGDEV - _check_device SCRATCH_RTDEV optional $SCRATCH_RTDEV - _check_device SCRATCH_LOGDEV optional $SCRATCH_LOGDEV - fi - - # Override FSTYP from config when running ./check -overlay - # and maybe override base fs TEST/SCRATCH_DEV with overlay base dirs. - # We need to do this *after* default mount options are set by base FSTYP - # and *after* SCRATCH_DEV is deduced from SCRATCH_DEV_POOL - if [ "$OVERLAY" == "true" -o "$FSTYP" == "overlay" ]; then - _overlay_config_override - fi -} - -if [ -z "$CONFIG_INCLUDED" ]; then - get_next_config `echo $HOST_OPTIONS_SECTIONS | cut -f1 -d" "` - export CONFIG_INCLUDED=true - - # Autodetect fs type based on what's on $TEST_DEV unless it's been set - # externally - if [ -z "$FSTYP" ] && [ ! -z "$TEST_DEV" ]; then - FSTYP=`blkid -c /dev/null -s TYPE -o value $TEST_DEV` - fi - FSTYP=${FSTYP:=xfs} - export FSTYP - [ -z "$MOUNT_OPTIONS" ] && _mount_opts - [ -z "$TEST_FS_MOUNT_OPTS" ] && _test_mount_opts - [ -z "$MKFS_OPTIONS" ] && _mkfs_opts - [ -z "$FSCK_OPTIONS" ] && _fsck_opts -else - # We get here for the non multi section case, on every test that sources - # common/rc after re-sourcing the HOST_OPTIONS config file. - # Because of this re-sourcing, we need to re-canonicalize the configured - # mount points and re-override TEST/SCRATCH_DEV overlay vars. - - # canonicalize the mount points - # this follows symlinks and removes all trailing "/"s - export TEST_DIR=`_canonicalize_mountpoint TEST_DIR $TEST_DIR` - export SCRATCH_MNT=`_canonicalize_mountpoint SCRATCH_MNT $SCRATCH_MNT` - - # Override FSTYP from config when running ./check -overlay - # and maybe override base fs TEST/SCRATCH_DEV with overlay base dirs - if [ "$OVERLAY" == "true" -o "$FSTYP" == "overlay" ]; then - _overlay_config_override - fi -fi - +_config_section_setup _canonicalize_devices # mkfs.xfs checks for TEST_DEV before permitting < 300M filesystems. TEST_DIR # and QA_CHECK_FS are also checked by mkfs.xfs, but already exported elsewhere.
common/config-sections new file +390 -0
390 lines not on this page · raw message
Signature · 1 line
2.45.2
Re: [PATCH 12/28] config: move config section code to it's own file
Quoted 5 lines · Dave Chinner
From: Dave Chinner <[email protected]>
Move the config section parsing, checking and setup code from common/config to common/config-section so that it can be included directly in contexts where the rest of common/config is not needed.
This looks okay to me. Just a couple of nits comments below.
Ritesh - this patch addresses one of feedbacks[1] on one of my cleanup patches.
[1] https://lore.kernel.org/all/[email protected]/
Quoted 414 lines
Signed-off-by Dave Chinner <[email protected]>--- common/config | 382 +--------------------------------------- common/config-sections | 390 +++++++++++++++++++++++++++++++++++++++++ 2 files changed, 392 insertions(+), 380 deletions(-) create mode 100644 common/config-sectionscommon/config +2
-380@@ -41,6 +41,7 @@ . common/test_names . common/exit +. common/config-sections # all tests should use a common language setting to prevent golden # output mismatches.@@ -544,386 +545,7 @@ _source_specific_fs() esac } -known_hosts() -{ - [ "$HOST_CONFIG_DIR" ] || HOST_CONFIG_DIR=`pwd`/configs - - [ -f /etc/xfsqa.config ] && export HOST_OPTIONS=/etc/xfsqa.config - [ -f $HOST_CONFIG_DIR/$HOST ] && export HOST_OPTIONS=$HOST_CONFIG_DIR/$HOST - [ -f $HOST_CONFIG_DIR/$HOST.config ] && export HOST_OPTIONS=$HOST_CONFIG_DIR/$HOST.config -} - -# Returns a list of sections in config file -# Each section starts with the section name in the format -# [section_name1]. Only alphanumeric characters and '_' is allowed -# in the section name otherwise the section will not be resognised. -# Section name must be contained between square brackets. -get_config_sections() { - sed -n -e "s/^\[\([[:alnum:]_-]*\)\]/\1/p" < $1 -} - -if [ ! -f "$HOST_OPTIONS" ]; then - known_hosts -fi - -export HOST_OPTIONS_SECTIONS="-no-sections-" -export OPTIONS_HAVE_SECTIONS=false -if [ -f "$HOST_OPTIONS" ]; then - export HOST_OPTIONS_SECTIONS=`get_config_sections $HOST_OPTIONS` - if [ -z "$HOST_OPTIONS_SECTIONS" ]; then - . $HOST_OPTIONS - export HOST_OPTIONS_SECTIONS="-no-sections-" - else - export OPTIONS_HAVE_SECTIONS=true - fi -fi - -_check_device() -{ - local name=$1 - local dev_needed=$2 - local dev=$3 - - if [ -z "$dev" ]; then - if [ "$dev_needed" == "required" ]; then - _fatal "common/config: $name is required but not defined!" - fi - return 0 - fi - - if [ -b "$dev" ] || ( echo $dev | grep -qE ":|//" ); then - # block device or a network url - return 0 - fi - - case "$FSTYP" in - 9p|fuse|tmpfs|virtiofs|afs) - # 9p, fuse, virtiofs and afs mount tags are just plain strings, - # so anything is allowed tmpfs doesn't use mount source, ignore - ;; - ceph) - # ceph has two different possible syntaxes for mount devices. The - # network URL check above catches the legacy syntax. Check for the - # new-style syntax here. - if ( echo $dev | grep -qEv "=/" ); then - _fatal "common/config: $name ($dev) is not a valid ceph mount string" - fi - ;; - overlay) - if [ ! -d "$dev" ]; then - _fatal "common/config: $name ($dev) is not a directory for overlay" - fi - ;; - ubifs) - if [ ! -c "$dev" ]; then - _fatal "common/config: $name ($dev) is not a character device" - fi - ;; - ceph-fuse) - ;; - *) - _fatal "common/config: $name ($dev) is not a block device or a network filesystem" - esac -} - -# check and return a canonical mount point path -_canonicalize_mountpoint() -{ - local name=$1 - local dir=$2 - - if [ -d "$dir" ]; then - # this follows symlinks and removes all trailing "/"s - readlink -e "$dir" - return 0 - fi - - if [ "$FSTYP" != "overlay" ] || [[ "$name" == OVL_BASE_* ]]; then - _fatal "common/config: $name ($dir) is not a directory" - fi - - # base fs may not be mounted yet, so just check that parent dir - # exists (where base fs will be mounted) because we are going to - # mkdir the overlay mount point dir anyway - local base=`basename $dir` - local parent=`dirname $dir` - parent=`_canonicalize_mountpoint OVL_BASE_$name "$parent"` - - # prepend the overlay mount point to canonical parent path - echo "$parent/$base" -} - -# Enables usage of /dev/disk/by-id/ symlinks to persist target devices -# over reboots -_canonicalize_devices() -{ - if [ "$CANON_DEVS" != "yes" ]; then - return - fi - [ -L "$TEST_DEV" ] && TEST_DEV=$(readlink -e "$TEST_DEV") - [ -L "$SCRATCH_DEV" ] && SCRATCH_DEV=$(readlink -e "$SCRATCH_DEV") - [ -L "$TEST_LOGDEV" ] && TEST_LOGDEV=$(readlink -e "$TEST_LOGDEV") - [ -L "$TEST_RTDEV" ] && TEST_RTDEV=$(readlink -e "$TEST_RTDEV") - [ -L "$SCRATCH_RTDEV" ] && SCRATCH_RTDEV=$(readlink -e "$SCRATCH_RTDEV") - [ -L "$LOGWRITES_DEV" ] && LOGWRITES_DEV=$(readlink -e "$LOGWRITES_DEV") - if [ ! -z "$SCRATCH_DEV_POOL" ]; then - local NEW_SCRATCH_POOL="" - for i in $SCRATCH_DEV_POOL; do - if [ -L $i ]; then - NEW_SCRATCH_POOL="$NEW_SCRATCH_POOL $(readlink -e $i)" - else - NEW_SCRATCH_POOL="$NEW_SCRATCH_POOL $i" - fi - done - SCRATCH_DEV_POOL="$NEW_SCRATCH_POOL" - fi -} - -# On check -overlay, for the non multi section config case, this -# function is called on every test, before init_rc(). -# When SCRATCH/TEST_* vars are defined in config file, config file -# is sourced on every test and this function overrides the vars -# every time. -# When SCRATCH/TEST_* vars are defined in evironment and not -# in config file, this function is called after vars have already -# been overriden in the previous test. -# In that case, TEST_DEV is a directory and not a blockdev/chardev and -# the function will return without overriding the SCRATCH/TEST_* vars. -_overlay_config_override() -{ - # There are 2 options for configuring overlayfs tests: - # - # 1. (legacy) SCRATCH/TEST_DEV point to existing directories - # on an already mounted fs. In this case, the new - # OVL_BASE_SCRATCH/TEST_* vars are set to use the legacy - # vars values (even though they may not be mount points). - # - [ ! -d "$TEST_DEV" ] || export OVL_BASE_TEST_DIR="$TEST_DEV" - [ ! -d "$SCRATCH_DEV" ] || export OVL_BASE_SCRATCH_MNT="$SCRATCH_DEV" - - # Config file may specify base fs type, but we obay -overlay flag - [ "$FSTYP" == overlay ] || export OVL_BASE_FSTYP="$FSTYP" - export FSTYP=overlay - - # 2. SCRATCH/TEST_DEV point to the base fs partitions. In this case, - # the new OVL_BASE_SCRATCH/TEST_DEV/MNT vars are set to the values - # of the configured base fs and SCRATCH/TEST_DEV vars are set to the - # overlayfs base and mount dirs inside base fs mount. - [ -b "$TEST_DEV" ] || [ -c "$TEST_DEV" ] || [ "$OVL_BASE_FSTYP" == tmpfs ] || return 0 - - # Store original base fs vars - export OVL_BASE_TEST_DEV="$TEST_DEV" - export OVL_BASE_TEST_DIR="$TEST_DIR" - # If config does not set MOUNT_OPTIONS, its value may be - # leftover from previous _overlay_config_override, so - # don't use that value for base fs mount - [ "$MOUNT_OPTIONS" != "$OVERLAY_MOUNT_OPTIONS" ] || unset MOUNT_OPTIONS - export OVL_BASE_MOUNT_OPTIONS="$MOUNT_OPTIONS" - - # Set TEST vars to overlay base and mount dirs inside base fs - export TEST_DEV="$OVL_BASE_TEST_DIR" - export TEST_DIR="$OVL_BASE_TEST_DIR/$OVL_MNT" - export MOUNT_OPTIONS="$OVERLAY_MOUNT_OPTIONS" - - [ -b "$SCRATCH_DEV" ] || [ -c "$SCRATCH_DEV" ] || [ "$OVL_BASE_FSTYP" == tmpfs ] || return 0 - - # Store original base fs vars - export OVL_BASE_SCRATCH_DEV="$SCRATCH_DEV" - export OVL_BASE_SCRATCH_MNT="$SCRATCH_MNT" - - # Set SCRATCH vars to overlay base and mount dirs inside base fs - export SCRATCH_DEV="$OVL_BASE_SCRATCH_MNT" - export SCRATCH_MNT="$OVL_BASE_SCRATCH_MNT/$OVL_MNT" - - # Set fsck options, use default if user not set directly. - export FSCK_OPTIONS="$OVERLAY_FSCK_OPTIONS" - [ -z "$FSCK_OPTIONS" ] && _fsck_opts - export IDMAPPED_MOUNTS="$IDMAPPED_MOUNTS" -} - -_overlay_config_restore() -{ - export OVERLAY=true - [ -z "$OVL_BASE_FSTYP" ] || export FSTYP=$OVL_BASE_FSTYP - [ -z "$OVL_BASE_TEST_DEV" ] || export TEST_DEV=$OVL_BASE_TEST_DEV - [ -z "$OVL_BASE_TEST_DIR" ] || export TEST_DIR=$OVL_BASE_TEST_DIR - [ -z "$OVL_BASE_SCRATCH_DEV" ] || export SCRATCH_DEV=$OVL_BASE_SCRATCH_DEV - [ -z "$OVL_BASE_SCRATCH_MNT" ] || export SCRATCH_MNT=$OVL_BASE_SCRATCH_MNT - [ -z "$OVL_BASE_MOUNT_OPTIONS" ] || export MOUNT_OPTIONS=$OVL_BASE_MOUNT_OPTIONS -} - -# Parse config section options. This function will parse all the configuration -# within a single section which name is passed as an argument. For section -# name format see comments in get_config_sections(). -# Empty lines and everything after '#' will be ignored. -# Configuration options should be defined in the format -# -# CONFIG_OPTION=value -# -# This 'CONFIG_OPTION' variable and will be exported as an environment variable. -parse_config_section() { - SECTION=$1 - if ! $OPTIONS_HAVE_SECTIONS; then - return 0 - fi - eval `sed -e 's/[[:space:]]*\=[[:space:]]*/=/g' \ - -e 's/#.*$//' \ - -e 's/[[:space:]]*$//' \ - -e 's/^[[:space:]]*//' \ - -e "s/^\([^=]*\)=\"\?'\?\([^\"']*\)\"\?'\?$/export \1=\"\2\"/" \ - < $HOST_OPTIONS \ - | sed -n -e "/^\[$SECTION\]/,/^\s*\[/{/^[^#].*\=.*/p;}"` -} - -get_next_config() { - if [ ! -z "$CONFIG_INCLUDED" ] && ! $OPTIONS_HAVE_SECTIONS; then - return 0 - fi - - # We might have overriden FSTYP and TEST/SCRATCH vars with overlay values - # in the previous section, so restore them to original values stored in - # OVL_BASE_*. - # We need to do this *before* old FSTYP and MOUNT_OPTIONS are recorded - # and *before* SCRATCH_DEV and MOUNT_OPTIONS are unset - if [ "$FSTYP" == "overlay" ]; then - _overlay_config_restore - fi - - local OLD_FSTYP=$FSTYP - local OLD_MOUNT_OPTIONS=$MOUNT_OPTIONS - local OLD_TEST_FS_MOUNT_OPTS=$TEST_FS_MOUNT_OPTS - local OLD_MKFS_OPTIONS=$MKFS_OPTIONS - local OLD_FSCK_OPTIONS=$FSCK_OPTIONS - local OLD_USE_EXTERNAL=$USE_EXTERNAL - - unset MOUNT_OPTIONS - unset TEST_FS_MOUNT_OPTS - unset MKFS_OPTIONS - unset FSCK_OPTIONS - unset USE_EXTERNAL - - # We might have deduced SCRATCH_DEV from the SCRATCH_DEV_POOL in the previous - # run, so we have to unset it now. - if [ "$SCRATCH_DEV_NOT_SET" == "true" ]; then - unset SCRATCH_DEV - fi - - parse_config_section $1 - if [ ! -z "$OLD_FSTYP" ] && [ $OLD_FSTYP != $FSTYP ]; then - [ -z "$MOUNT_OPTIONS" ] && _mount_opts - [ -z "$TEST_FS_MOUNT_OPTS" ] && _test_mount_opts - [ -z "$MKFS_OPTIONS" ] && _mkfs_opts - [ -z "$FSCK_OPTIONS" ] && _fsck_opts - - # clear the external devices if we are not using them - if [ -z "$USE_EXTERNAL" ]; then - unset TEST_RTDEV - unset TEST_LOGDEV - unset SCRATCH_RTDEV - unset SCRATCH_LOGDEV - fi - else - [ -z "$MOUNT_OPTIONS" ] && export MOUNT_OPTIONS=$OLD_MOUNT_OPTIONS - [ -z "$TEST_FS_MOUNT_OPTS" ] && export TEST_FS_MOUNT_OPTS=$OLD_TEST_FS_MOUNT_OPTS - [ -z "$MKFS_OPTIONS" ] && export MKFS_OPTIONS=$OLD_MKFS_OPTIONS - [ -z "$FSCK_OPTIONS" ] && export FSCK_OPTIONS=$OLD_FSCK_OPTIONS - [ -z "$USE_EXTERNAL" ] && export USE_EXTERNAL=$OLD_USE_EXTERNAL - fi - - # set default RESULT_BASE - if [ -z "$RESULT_BASE" ]; then - export RESULT_BASE="$here/results/" - fi - - if [ "$FSTYP" == "tmpfs" ]; then - if [ -z "$TEST_DEV" ]; then - export TEST_DEV=tmpfs_test - fi - if [ -z "$SCRATCH_DEV" ]; then - export TEST_DEV=tmpfs_scratch - fi - fi - - # Mandatory Config values. - MC="" - [ -z "$EMAIL" ] && MC="$MC EMAIL" - [ -z "$TEST_DIR" ] && MC="$MC TEST_DIR" - [ -z "$TEST_DEV" ] && MC="$MC TEST_DEV" - - if [ -n "$MC" ]; then - echo "Warning: need to define parameters for host $HOST" - echo " or set variables:" - echo " $MC" - _exit 1 - fi - - _check_device TEST_DEV required $TEST_DEV - export TEST_DIR=`_canonicalize_mountpoint TEST_DIR $TEST_DIR` - - # a btrfs tester will set only SCRATCH_DEV_POOL, we will put first of its dev - # to SCRATCH_DEV and rest to SCRATCH_DEV_POOL to maintain the backward compatibility - if [ ! -z "$SCRATCH_DEV_POOL" ]; then - if [ ! -z "$SCRATCH_DEV" ]; then - echo "common/config: Error: \$SCRATCH_DEV ($SCRATCH_DEV) should be unset when \$SCRATCH_DEV_POOL ($SCRATCH_DEV_POOL) is set" - _exit 1 - fi - SCRATCH_DEV=`echo $SCRATCH_DEV_POOL | awk '{print $1}'` - export SCRATCH_DEV - export SCRATCH_DEV_NOT_SET=true - fi - - _check_device SCRATCH_DEV optional $SCRATCH_DEV - export SCRATCH_MNT=`_canonicalize_mountpoint SCRATCH_MNT $SCRATCH_MNT` - - if [ -n "$USE_EXTERNAL" ]; then - _check_device TEST_RTDEV optional $TEST_RTDEV - _check_device TEST_LOGDEV optional $TEST_LOGDEV - _check_device SCRATCH_RTDEV optional $SCRATCH_RTDEV - _check_device SCRATCH_LOGDEV optional $SCRATCH_LOGDEV - fi - - # Override FSTYP from config when running ./check -overlay - # and maybe override base fs TEST/SCRATCH_DEV with overlay base dirs. - # We need to do this *after* default mount options are set by base FSTYP - # and *after* SCRATCH_DEV is deduced from SCRATCH_DEV_POOL - if [ "$OVERLAY" == "true" -o "$FSTYP" == "overlay" ]; then - _overlay_config_override - fi -} - -if [ -z "$CONFIG_INCLUDED" ]; then - get_next_config `echo $HOST_OPTIONS_SECTIONS | cut -f1 -d" "` - export CONFIG_INCLUDED=true - - # Autodetect fs type based on what's on $TEST_DEV unless it's been set - # externally - if [ -z "$FSTYP" ] && [ ! -z "$TEST_DEV" ]; then - FSTYP=`blkid -c /dev/null -s TYPE -o value $TEST_DEV` - fi - FSTYP=${FSTYP:=xfs} - export FSTYP - [ -z "$MOUNT_OPTIONS" ] && _mount_opts - [ -z "$TEST_FS_MOUNT_OPTS" ] && _test_mount_opts - [ -z "$MKFS_OPTIONS" ] && _mkfs_opts - [ -z "$FSCK_OPTIONS" ] && _fsck_opts -else - # We get here for the non multi section case, on every test that sources - # common/rc after re-sourcing the HOST_OPTIONS config file. - # Because of this re-sourcing, we need to re-canonicalize the configured - # mount points and re-override TEST/SCRATCH_DEV overlay vars. - - # canonicalize the mount points - # this follows symlinks and removes all trailing "/"s - export TEST_DIR=`_canonicalize_mountpoint TEST_DIR $TEST_DIR` - export SCRATCH_MNT=`_canonicalize_mountpoint SCRATCH_MNT $SCRATCH_MNT` - - # Override FSTYP from config when running ./check -overlay - # and maybe override base fs TEST/SCRATCH_DEV with overlay base dirs - if [ "$OVERLAY" == "true" -o "$FSTYP" == "overlay" ]; then - _overlay_config_override - fi -fi - +_config_section_setup _canonicalize_devices # mkfs.xfs checks for TEST_DEV before permitting < 300M filesystems. TEST_DIR # and QA_CHECK_FS are also checked by mkfs.xfs, but already exported elsewhere.common/config-sections new file +0
-0
common/config-sections line 6
Quoted 6 lines
common/config-sections new file +6
-0@@ -0,0 +1,390 @@ +##/bin/bash +# SPDX-License-Identifier: GPL-2.0 +# Copyright (c) 2000-2003,2006 Silicon Graphics, Inc. All Rights Reserved. +# Copyright (c) 2025 Red Hat, Inc. All Rights Reserved. +# +# Config section parsing and setup code
Since we want to facilitate sourcing of this file independently, should we at least mention in the comments some of the dependant files this file needs to be sourced? For example, this file uses _exit() - so usage of this file requires common/exit to be sourced?
Quoted 36 lines
common/config-sections new file +36
-0+ +_check_device() +{ + local name=$1 + local dev_needed=$2 + local dev=$3 + + if [ -z "$dev" ]; then + if [ "$dev_needed" == "required" ]; then + _fatal "common/config: $name is required but not defined!" + fi + return 0 + fi + + if [ -b "$dev" ] || ( echo $dev | grep -qE ":|//" ); then + # block device or a network url + return 0 + fi + + case "$FSTYP" in + 9p|fuse|tmpfs|virtiofs|afs) + # 9p, fuse, virtiofs and afs mount tags are just plain strings, + # so anything is allowed tmpfs doesn't use mount source, ignore + ;; + ceph) + # ceph has two different possible syntaxes for mount devices. The + # network URL check above catches the legacy syntax. Check for the + # new-style syntax here. + if ( echo $dev | grep -qEv "=/" ); then + _fatal "common/config: $name ($dev) is not a valid ceph mount string" + fi + ;; + overlay) + if [ ! -d "$dev" ]; then + _fatal "common/config: $name ($dev) is not a directory for overlay" + fi
common/config-sections line 52
Quoted 10 lines
common/config-sections new file +10
-0+ ;; + ubifs) + if [ ! -c "$dev" ]; then + _fatal "common/config: $name ($dev) is not a character device" + fi + ;; + ceph-fuse) + ;; + *) + _fatal "common/config: $name ($dev) is not a block device or a network filesystem"
Nit: 80 chars limit exceed.
Quoted 77 lines
common/config-sections new file +77
-0+ esac +} + +# check and return a canonical mount point path +_canonicalize_mountpoint() +{ + local name=$1 + local dir=$2 + + if [ -d "$dir" ]; then + # this follows symlinks and removes all trailing "/"s + readlink -e "$dir" + return 0 + fi + + if [ "$FSTYP" != "overlay" ] || [[ "$name" == OVL_BASE_* ]]; then + _fatal "common/config: $name ($dir) is not a directory" + fi + + # base fs may not be mounted yet, so just check that parent dir + # exists (where base fs will be mounted) because we are going to + # mkdir the overlay mount point dir anyway + local base=`basename $dir` + local parent=`dirname $dir` + parent=`_canonicalize_mountpoint OVL_BASE_$name "$parent"` + + # prepend the overlay mount point to canonical parent path + echo "$parent/$base" +} + +# Enables usage of /dev/disk/by-id/ symlinks to persist target devices +# over reboots +_canonicalize_devices() +{ + if [ "$CANON_DEVS" != "yes" ]; then + return + fi + [ -L "$TEST_DEV" ] && TEST_DEV=$(readlink -e "$TEST_DEV") + [ -L "$SCRATCH_DEV" ] && SCRATCH_DEV=$(readlink -e "$SCRATCH_DEV") + [ -L "$TEST_LOGDEV" ] && TEST_LOGDEV=$(readlink -e "$TEST_LOGDEV") + [ -L "$TEST_RTDEV" ] && TEST_RTDEV=$(readlink -e "$TEST_RTDEV") + [ -L "$SCRATCH_RTDEV" ] && SCRATCH_RTDEV=$(readlink -e "$SCRATCH_RTDEV") + [ -L "$LOGWRITES_DEV" ] && LOGWRITES_DEV=$(readlink -e "$LOGWRITES_DEV") + if [ ! -z "$SCRATCH_DEV_POOL" ]; then + local NEW_SCRATCH_POOL="" + for i in $SCRATCH_DEV_POOL; do + if [ -L $i ]; then + NEW_SCRATCH_POOL="$NEW_SCRATCH_POOL $(readlink -e $i)" + else + NEW_SCRATCH_POOL="$NEW_SCRATCH_POOL $i" + fi + done + SCRATCH_DEV_POOL="$NEW_SCRATCH_POOL" + fi +} + +# On check -overlay, for the non multi section config case, this +# function is called on every test, before init_rc(). +# When SCRATCH/TEST_* vars are defined in config file, config file +# is sourced on every test and this function overrides the vars +# every time. +# When SCRATCH/TEST_* vars are defined in evironment and not +# in config file, this function is called after vars have already +# been overriden in the previous test. +# In that case, TEST_DEV is a directory and not a blockdev/chardev and +# the function will return without overriding the SCRATCH/TEST_* vars. +_overlay_config_override() +{ + # There are 2 options for configuring overlayfs tests: + # + # 1. (legacy) SCRATCH/TEST_DEV point to existing directories + # on an already mounted fs. In this case, the new + # OVL_BASE_SCRATCH/TEST_* vars are set to use the legacy + # vars values (even though they may not be mount points). + # + [ ! -d "$TEST_DEV" ] || export OVL_BASE_TEST_DIR="$TEST_DEV" + [ ! -d "$SCRATCH_DEV" ] || export OVL_BASE_SCRATCH_MNT="$SCRATCH_DEV"
common/config-sections line 139
Quoted 10 lines
common/config-sections new file +10
-0+ + # Config file may specify base fs type, but we obay -overlay flag + [ "$FSTYP" == overlay ] || export OVL_BASE_FSTYP="$FSTYP" + export FSTYP=overlay + + # 2. SCRATCH/TEST_DEV point to the base fs partitions. In this case, + # the new OVL_BASE_SCRATCH/TEST_DEV/MNT vars are set to the values + # of the configured base fs and SCRATCH/TEST_DEV vars are set to the + # overlayfs base and mount dirs inside base fs mount. + [ -b "$TEST_DEV" ] || [ -c "$TEST_DEV" ] || [ "$OVL_BASE_FSTYP" == tmpfs ] || return 0
Nit: 80 chars limit exceeded
Quoted 16 lines
common/config-sections new file +16
-0+ + # Store original base fs vars + export OVL_BASE_TEST_DEV="$TEST_DEV" + export OVL_BASE_TEST_DIR="$TEST_DIR" + # If config does not set MOUNT_OPTIONS, its value may be + # leftover from previous _overlay_config_override, so + # don't use that value for base fs mount + [ "$MOUNT_OPTIONS" != "$OVERLAY_MOUNT_OPTIONS" ] || unset MOUNT_OPTIONS + export OVL_BASE_MOUNT_OPTIONS="$MOUNT_OPTIONS" + + # Set TEST vars to overlay base and mount dirs inside base fs + export TEST_DEV="$OVL_BASE_TEST_DIR" + export TEST_DIR="$OVL_BASE_TEST_DIR/$OVL_MNT" + export MOUNT_OPTIONS="$OVERLAY_MOUNT_OPTIONS" + +
Quoted 1 line
[ -b "$SCRATCH_DEV" ] || [ -c "$SCRATCH_DEV" ] || [ "$OVL_BASE_FSTYP" == tmpfs ] || return 0
Nit: 80 chars limit exceeded
Quoted 171 lines
common/config-sections new file +171
-0+ + # Store original base fs vars + export OVL_BASE_SCRATCH_DEV="$SCRATCH_DEV" + export OVL_BASE_SCRATCH_MNT="$SCRATCH_MNT" + + # Set SCRATCH vars to overlay base and mount dirs inside base fs + export SCRATCH_DEV="$OVL_BASE_SCRATCH_MNT" + export SCRATCH_MNT="$OVL_BASE_SCRATCH_MNT/$OVL_MNT" + + # Set fsck options, use default if user not set directly. + export FSCK_OPTIONS="$OVERLAY_FSCK_OPTIONS" + [ -z "$FSCK_OPTIONS" ] && _fsck_opts + export IDMAPPED_MOUNTS="$IDMAPPED_MOUNTS" +} + +_overlay_config_restore() +{ + export OVERLAY=true + [ -z "$OVL_BASE_FSTYP" ] || export FSTYP=$OVL_BASE_FSTYP + [ -z "$OVL_BASE_TEST_DEV" ] || export TEST_DEV=$OVL_BASE_TEST_DEV + [ -z "$OVL_BASE_TEST_DIR" ] || export TEST_DIR=$OVL_BASE_TEST_DIR + [ -z "$OVL_BASE_SCRATCH_DEV" ] || export SCRATCH_DEV=$OVL_BASE_SCRATCH_DEV + [ -z "$OVL_BASE_SCRATCH_MNT" ] || export SCRATCH_MNT=$OVL_BASE_SCRATCH_MNT + [ -z "$OVL_BASE_MOUNT_OPTIONS" ] || export MOUNT_OPTIONS=$OVL_BASE_MOUNT_OPTIONS +} + +# Returns a list of sections in config file +# Each section starts with the section name in the format +# [section_name1]. Only alphanumeric characters and '_' is allowed +# in the section name otherwise the section will not be resognised. +# Section name must be contained between square brackets. +get_config_sections() { + sed -n -e "s/^\[\([[:alnum:]_-]*\)\]/\1/p" < $1 +} + +# Parse config section options. This function will parse all the configuration +# within a single section which name is passed as an argument. For section +# name format see comments in get_config_sections(). +# Empty lines and everything after '#' will be ignored. +# Configuration options should be defined in the format +# +# CONFIG_OPTION=value +# +# This 'CONFIG_OPTION' variable and will be exported as an environment variable. +parse_config_section() { + SECTION=$1 + if ! $OPTIONS_HAVE_SECTIONS; then + return 0 + fi + eval `sed -e 's/[[:space:]]*\=[[:space:]]*/=/g' \ + -e 's/#.*$//' \ + -e 's/[[:space:]]*$//' \ + -e 's/^[[:space:]]*//' \ + -e "s/^\([^=]*\)=\"\?'\?\([^\"']*\)\"\?'\?$/export \1=\"\2\"/" \ + < $HOST_OPTIONS \ + | sed -n -e "/^\[$SECTION\]/,/^\s*\[/{/^[^#].*\=.*/p;}"` +} + +get_next_config() { + if [ ! -z "$CONFIG_INCLUDED" ] && ! $OPTIONS_HAVE_SECTIONS; then + return 0 + fi + + # We might have overriden FSTYP and TEST/SCRATCH vars with overlay values + # in the previous section, so restore them to original values stored in + # OVL_BASE_*. + # We need to do this *before* old FSTYP and MOUNT_OPTIONS are recorded + # and *before* SCRATCH_DEV and MOUNT_OPTIONS are unset + if [ "$FSTYP" == "overlay" ]; then + _overlay_config_restore + fi + + local OLD_FSTYP=$FSTYP + local OLD_MOUNT_OPTIONS=$MOUNT_OPTIONS + local OLD_TEST_FS_MOUNT_OPTS=$TEST_FS_MOUNT_OPTS + local OLD_MKFS_OPTIONS=$MKFS_OPTIONS + local OLD_FSCK_OPTIONS=$FSCK_OPTIONS + local OLD_USE_EXTERNAL=$USE_EXTERNAL + + unset MOUNT_OPTIONS + unset TEST_FS_MOUNT_OPTS + unset MKFS_OPTIONS + unset FSCK_OPTIONS + unset USE_EXTERNAL + + # We might have deduced SCRATCH_DEV from the SCRATCH_DEV_POOL in the previous + # run, so we have to unset it now. + if [ "$SCRATCH_DEV_NOT_SET" == "true" ]; then + unset SCRATCH_DEV + fi + + parse_config_section $1 + if [ ! -z "$OLD_FSTYP" ] && [ $OLD_FSTYP != $FSTYP ]; then + [ -z "$MOUNT_OPTIONS" ] && _mount_opts + [ -z "$TEST_FS_MOUNT_OPTS" ] && _test_mount_opts + [ -z "$MKFS_OPTIONS" ] && _mkfs_opts + [ -z "$FSCK_OPTIONS" ] && _fsck_opts + + # clear the external devices if we are not using them + if [ -z "$USE_EXTERNAL" ]; then + unset TEST_RTDEV + unset TEST_LOGDEV + unset SCRATCH_RTDEV + unset SCRATCH_LOGDEV + fi + else + [ -z "$MOUNT_OPTIONS" ] && export MOUNT_OPTIONS=$OLD_MOUNT_OPTIONS + [ -z "$TEST_FS_MOUNT_OPTS" ] && export TEST_FS_MOUNT_OPTS=$OLD_TEST_FS_MOUNT_OPTS + [ -z "$MKFS_OPTIONS" ] && export MKFS_OPTIONS=$OLD_MKFS_OPTIONS + [ -z "$FSCK_OPTIONS" ] && export FSCK_OPTIONS=$OLD_FSCK_OPTIONS + [ -z "$USE_EXTERNAL" ] && export USE_EXTERNAL=$OLD_USE_EXTERNAL + fi + + # set default RESULT_BASE + if [ -z "$RESULT_BASE" ]; then + export RESULT_BASE="$here/results/" + fi + + if [ "$FSTYP" == "tmpfs" ]; then + if [ -z "$TEST_DEV" ]; then + export TEST_DEV=tmpfs_test + fi + if [ -z "$SCRATCH_DEV" ]; then + export TEST_DEV=tmpfs_scratch + fi + fi + + # Mandatory Config values. + MC="" + [ -z "$EMAIL" ] && MC="$MC EMAIL" + [ -z "$TEST_DIR" ] && MC="$MC TEST_DIR" + [ -z "$TEST_DEV" ] && MC="$MC TEST_DEV" + + if [ -n "$MC" ]; then + echo "Warning: need to define parameters for host $HOST" + echo " or set variables:" + echo " $MC" + _exit 1 + fi + + _check_device TEST_DEV required $TEST_DEV + export TEST_DIR=`_canonicalize_mountpoint TEST_DIR $TEST_DIR` + + # a btrfs tester will set only SCRATCH_DEV_POOL, we will put first of its dev + # to SCRATCH_DEV and rest to SCRATCH_DEV_POOL to maintain the backward compatibility + if [ ! -z "$SCRATCH_DEV_POOL" ]; then + if [ ! -z "$SCRATCH_DEV" ]; then + echo "common/config: Error: \$SCRATCH_DEV ($SCRATCH_DEV) should be unset when \$SCRATCH_DEV_POOL ($SCRATCH_DEV_POOL) is set" + _exit 1 + fi + SCRATCH_DEV=`echo $SCRATCH_DEV_POOL | awk '{print $1}'` + export SCRATCH_DEV + export SCRATCH_DEV_NOT_SET=true + fi + + _check_device SCRATCH_DEV optional $SCRATCH_DEV + export SCRATCH_MNT=`_canonicalize_mountpoint SCRATCH_MNT $SCRATCH_MNT` + + if [ -n "$USE_EXTERNAL" ]; then + _check_device TEST_RTDEV optional $TEST_RTDEV + _check_device TEST_LOGDEV optional $TEST_LOGDEV + _check_device SCRATCH_RTDEV optional $SCRATCH_RTDEV + _check_device SCRATCH_LOGDEV optional $SCRATCH_LOGDEV + fi + + # Override FSTYP from config when running ./check -overlay + # and maybe override base fs TEST/SCRATCH_DEV with overlay base dirs. + # We need to do this *after* default mount options are set by base FSTYP + # and *after* SCRATCH_DEV is deduced from SCRATCH_DEV_POOL + if [ "$OVERLAY" == "true" -o "$FSTYP" == "overlay" ]; then + _overlay_config_override
common/config-sections line 336
Quoted 10 lines
common/config-sections new file +10
-0+ fi +} + +known_hosts() +{ + [ "$HOST_CONFIG_DIR" ] || HOST_CONFIG_DIR=`pwd`/configs + + [ -f /etc/xfsqa.config ] && export HOST_OPTIONS=/etc/xfsqa.config + [ -f $HOST_CONFIG_DIR/$HOST ] && export HOST_OPTIONS=$HOST_CONFIG_DIR/$HOST + [ -f $HOST_CONFIG_DIR/$HOST.config ] && export HOST_OPTIONS=$HOST_CONFIG_DIR/$HOST.config
Nit: All 3 statements above exceed 80 char limit
--NR
Quoted 54 lines
common/config-sections new file +54
-0+} + +_config_section_setup() +{ + if [ ! -f "$HOST_OPTIONS" ]; then + known_hosts + fi + + export HOST_OPTIONS_SECTIONS="-no-sections-" + export OPTIONS_HAVE_SECTIONS=false + if [ -f "$HOST_OPTIONS" ]; then + export HOST_OPTIONS_SECTIONS=`get_config_sections $HOST_OPTIONS` + if [ -z "$HOST_OPTIONS_SECTIONS" ]; then + . $HOST_OPTIONS + export HOST_OPTIONS_SECTIONS="-no-sections-" + else + export OPTIONS_HAVE_SECTIONS=true + fi + fi + + if [ -z "$CONFIG_INCLUDED" ]; then + get_next_config `echo $HOST_OPTIONS_SECTIONS | cut -f1 -d" "` + export CONFIG_INCLUDED=true + + # Autodetect fs type based on what's on $TEST_DEV unless it's + # been set externally + if [ -z "$FSTYP" ] && [ ! -z "$TEST_DEV" ]; then + FSTYP=`blkid -c /dev/null -s TYPE -o value $TEST_DEV` + fi + FSTYP=${FSTYP:=xfs} + export FSTYP + [ -z "$MOUNT_OPTIONS" ] && _mount_opts + [ -z "$TEST_FS_MOUNT_OPTS" ] && _test_mount_opts + [ -z "$MKFS_OPTIONS" ] && _mkfs_opts + [ -z "$FSCK_OPTIONS" ] && _fsck_opts + else + # We get here for the non multi section case, on every test that + # sources common/rc after re-sourcing the HOST_OPTIONS config + # file. Because of this re-sourcing, we need to re-canonicalize + # the configured mount points and re-override TEST/SCRATCH_DEV + # overlay vars. + + # canonicalize the mount points + # this follows symlinks and removes all trailing "/"s + export TEST_DIR=`_canonicalize_mountpoint TEST_DIR $TEST_DIR` + export SCRATCH_MNT=`_canonicalize_mountpoint SCRATCH_MNT $SCRATCH_MNT` + + # Override FSTYP from config when running ./check -overlay and + # maybe override base fs TEST/SCRATCH_DEV with overlay base dirs + if [ "$OVERLAY" == "true" -o "$FSTYP" == "overlay" ]; then + _overlay_config_override + fi + fi +}
Re: [PATCH 12/28] config: move config section code to it's own file
Quoted 9 lines · Nirjhar Roy
Quoted 5 lines · Dave Chinner
From: Dave Chinner <[email protected]>
Move the config section parsing, checking and setup code from common/config to common/config-section so that it can be included directly in contexts where the rest of common/config is not needed.
This looks okay to me. Just a couple of nits comments below.
Reviewed-by Nirjhar Roy (IBM) <[email protected]>
....
Quoted 8 lines
Quoted 8 lines
common/config-sections new file +2
-0@@ -0,0 +1,390 @@ +##/bin/bash +# SPDX-License-Identifier: GPL-2.0
Quoted 8 lines
common/config-sections line 6
Quoted 4 lines
common/config-sections new file +4
-0+# Copyright (c) 2000-2003,2006 Silicon Graphics, Inc. All Rights Reserved. +# Copyright (c) 2025 Red Hat, Inc. All Rights Reserved. +# +# Config section parsing and setup codeSince we want to facilitate sourcing of this file independently, should we at least mention in the comments some of the dependant files this file needs to be sourced? For example, this file uses
_exit()- so usage of this file requires common/exit to be sourced?
The high level code should have already sourced that file.
Quoted 39 lines
Quoted 39 lines
common/config-sections new file +39
-0+ +_check_device() +{ + local name=$1 + local dev_needed=$2 + local dev=$3 + + if [ -z "$dev" ]; then + if [ "$dev_needed" == "required" ]; then + _fatal "common/config: $name is required but not defined!" + fi + return 0 + fi + + if [ -b "$dev" ] || ( echo $dev | grep -qE ":|//" ); then + # block device or a network url + return 0 + fi + + case "$FSTYP" in + 9p|fuse|tmpfs|virtiofs|afs) + # 9p, fuse, virtiofs and afs mount tags are just plain strings, + # so anything is allowed tmpfs doesn't use mount source, ignore + ;; + ceph) + # ceph has two different possible syntaxes for mount devices. The + # network URL check above catches the legacy syntax. Check for the + # new-style syntax here. + if ( echo $dev | grep -qEv "=/" ); then + _fatal "common/config: $name ($dev) is not a valid ceph mount string" + fi + ;; + overlay) + if [ ! -d "$dev" ]; then + _fatal "common/config: $name ($dev) is not a directory for overlay" + fi + ;; + ubifs) + if [ ! -c "$dev" ]; then
Quoted 8 lines
common/config-sections line 52
Quoted 7 lines
common/config-sections new file +7
-0+ _fatal "common/config: $name ($dev) is not a character device" + fi + ;; + ceph-fuse) + ;; + *) + _fatal "common/config: $name ($dev) is not a block device or a network filesystem"Nit: 80 chars limit exceed.
There are many of these in the code I moved, and there are many, many lines that exceed 80 columns all through fstests. I don't think this needs fixing.
Regardless, this patch is moving code from A to B, so I'm trying to avoid mixing in formatting or bug fixes that would otherwise be impossible to spot in the diff...
-Dave.
Signature · 2 lines
Dave Chinner [email protected]
PATCH 13/28 check-parallel: introduce config file support
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
check-parallel will use the same config file format check does, and use the same code to auto-discover the config file.
The biggest difference is that check-parallel -requires- the use of config sections, and the first section *must* be named "[check-parallel]". This first section is used for defining setup parameters for check parallel - loop device image file sizes, etc.
The second biggest difference is that check-parallel does not allow the config file to define devices. Any section found to contain a device definition such as TEST_DEV or SCRATCH_DEV will result check-parallel terminating with an error.
This config file format works for check-parallel invoking check, too, because once a section is specified on the check command line, it effectively ignores unknown values set in sections that it doesn't run. Hence it effectively skips over the [check-parallel] setup section.
For check-parallel, each config section now defines just the filesystem configuration to be tested; all the usual mount and mkfs options apply, and USE_EXTERNAL must be set for testing external devices.
This commit implements the initial [check-parallel] section support and moves the build in default values for these parameters to the config file setup. This means if the config file does not contain all the necessary parameter values, a default value will be used for it.
--- check-parallel | 14 +++----- common/config-sections | 78 +++++++++++++++++++++++++++++++++++++++--- 2 files changed, 77 insertions(+), 15 deletions(-)
check-parallel +4 -10
@@ -15,22 +15,14 @@ runner_list=() runtimes=() show_test_list= run_section="" +iam="check-parallel" tmp=/tmp/check-parallel.$$ -TEST_DEV_SIZE=${TEST_DEV_SIZE:=10G} -TEST_RTDEV_SIZE=${TEST_RTDEV_SIZE:=10G} -TEST_LOGDEV_SIZE=${TEST_LOGDEV_SIZE:=128M} -SCRATCH_DEV_SIZE=${SCRATCH_DEV_SIZE:=20G} -SCRATCH_RTDEV_SIZE=${SCRATCH_RTDEV_SIZE:=20G} -SCRATCH_LOGDEV_SIZE=${SCRATCH_LOGDEV_SIZE:=512M} -LOGWRITES_DEV_SIZE=${LOGWRITES_DEV_SIZE:=2G} - -FSTYP= - . ./common/exit . ./common/test_names . ./common/test_list +. ./common/config-sections usage() {
@@ -332,6 +324,8 @@ cleanup() trap "cleanup; exit" HUP INT QUIT TERM +_config_setup_parallel + split_runner_list if [ -n "$show_test_list" ]; then echo Time ordered test list:
common/config-sections +73 -5
102 lines not on this page · raw message
Signature · 1 line
2.45.2
Re: [PATCH 13/28] check-parallel: introduce config file support
Quoted 136 lines · Dave Chinner
From: Dave Chinner <[email protected]>
check-parallel will use the same config file format check does, and use the same code to auto-discover the config file.
The biggest difference is that check-parallel -requires- the use of config sections, and the first section *must* be named "[check-parallel]". This first section is used for defining setup parameters for check parallel - loop device image file sizes, etc.
The second biggest difference is that check-parallel does not allow the config file to define devices. Any section found to contain a device definition such as
TEST_DEVorSCRATCH_DEVwill result check-parallel terminating with an error.This config file format works for check-parallel invoking check, too, because once a section is specified on the check command line, it effectively ignores unknown values set in sections that it doesn't run. Hence it effectively skips over the [check-parallel] setup section.
For check-parallel, each config section now defines just the filesystem configuration to be tested; all the usual mount and mkfs options apply, and
USE_EXTERNALmust be set for testing external devices.This commit implements the initial [check-parallel] section support and moves the build in default values for these parameters to the config file setup. This means if the config file does not contain all the necessary parameter values, a default value will be used for it.
Signed-off-by Dave Chinner <[email protected]>--- check-parallel | 14 +++----- common/config-sections | 78 +++++++++++++++++++++++++++++++++++++++--- 2 files changed, 77 insertions(+), 15 deletions(-)check-parallel +4
-10@@ -15,22 +15,14 @@ runner_list=() runtimes=() show_test_list= run_section="" +iam="check-parallel" tmp=/tmp/check-parallel.$$ -TEST_DEV_SIZE=${TEST_DEV_SIZE:=10G} -TEST_RTDEV_SIZE=${TEST_RTDEV_SIZE:=10G} -TEST_LOGDEV_SIZE=${TEST_LOGDEV_SIZE:=128M} -SCRATCH_DEV_SIZE=${SCRATCH_DEV_SIZE:=20G} -SCRATCH_RTDEV_SIZE=${SCRATCH_RTDEV_SIZE:=20G} -SCRATCH_LOGDEV_SIZE=${SCRATCH_LOGDEV_SIZE:=512M} -LOGWRITES_DEV_SIZE=${LOGWRITES_DEV_SIZE:=2G} - -FSTYP= - . ./common/exit . ./common/test_names . ./common/test_list +. ./common/config-sections usage() {@@ -332,6 +324,8 @@ cleanup() trap "cleanup; exit" HUP INT QUIT TERM +_config_setup_parallel + split_runner_list if [ -n "$show_test_list" ]; then echo Time ordered test list:common/config-sections +25
-5@@ -329,6 +329,7 @@ get_next_config() { known_hosts() { + [ -z "$HOST" ] && export HOST=`hostname -s` [ "$HOST_CONFIG_DIR" ] || HOST_CONFIG_DIR=`pwd`/configs [ -f /etc/xfsqa.config ] && export HOST_OPTIONS=/etc/xfsqa.config@@ -336,7 +337,7 @@ known_hosts() [ -f $HOST_CONFIG_DIR/$HOST.config ] && export HOST_OPTIONS=$HOST_CONFIG_DIR/$HOST.config } -_config_section_setup() +_config_file_setup() { if [ ! -f "$HOST_OPTIONS" ]; then known_hosts@@ -346,13 +347,32 @@ _config_section_setup() export OPTIONS_HAVE_SECTIONS=false if [ -f "$HOST_OPTIONS" ]; then export HOST_OPTIONS_SECTIONS=`get_config_sections $HOST_OPTIONS` - if [ -z "$HOST_OPTIONS_SECTIONS" ]; then - . $HOST_OPTIONS - export HOST_OPTIONS_SECTIONS="-no-sections-" - else + if [ -n "$HOST_OPTIONS_SECTIONS" ]; then export OPTIONS_HAVE_SECTIONS=true fi fi +} + +_config_section_setup() +{ + if [ "$iam" == "check-parallel" ]; then + echo "$iam: incorrect config file format chosen!" + exit 1; + fi + + _config_file_setup + + # If we don't have sections, source the options from the config file. + # Otherwise, strip sections that should not be run by check that may be + # present in the config file + if [ "$OPTIONS_HAVE_SECTIONS" != "true" ]; then + . $HOST_OPTIONS + export HOST_OPTIONS_SECTIONS="-no-sections-" + else + export HOST_OPTIONS_SECTIONS=$(echo $HOST_OPTIONS_SECTIONS | \ + sed -e 's/check-parallel//') + fi + if [ -z "$CONFIG_INCLUDED" ]; then get_next_config `echo $HOST_OPTIONS_SECTIONS | cut -f1 -d" "`
common/config-sections _config_section_setup() · line 415
Quoted 8 lines
common/config-sections +5
-0@@ -388,3 +408,51 @@ _config_section_setup() fi fi } + +# check-parallel config files must: +# 1. have config sections defined +# 2. use the first config section for check-parallel setup +# 3. not define any physical device parameter in any section
Nit: Maybe some documentation of the above restrictions and some new variables introduced (like SCRATCH_DEV_SIZE, ..)in the README?
common/config-sections _config_section_setup() · line 428
Quoted 12 lines
common/config-sections +12
-0+# If all these are true, then we read the first section that defines +# the check-parallel config parameters and continue onwards. +_config_setup_parallel() +{ + if [ "$iam" != "check-parallel" ]; then + echo "$iam: incorrect config file format chosen!" + exit 1; + fi + + _config_file_setup + + if [ "$OPTIONS_HAVE_SECTIONS" != "true" ]; then
Most of the places the way OPTIONS_HAVE_SECTIONS is being used is as if $OPTIONS_HAVE_SECTIONS; then
...
fi
so, when OPTIONS_HAVE_SECTIONS=true, /bin/true is executed and the exit code of /bin/true is always a success and it is not a string comparison with "true" or "false". /bin/true always succeeds and hence if OPTIONS_HAVE_SECTIONS=true, then
if $OPTIONS_HAVE_SECTIONS is always executed. In the same way /bin/false always has a failure code upon exit.
So maybe we should change this to
if $OPTIONS_HAVE_SECTIONS; then - just to have consistency with the rest of the code?
common/config-sections _config_section_setup() · line 439
Quoted 11 lines
common/config-sections +11
-0+ echo "$iam config file has no sections!" + exit 1; + fi + + local first_section=`echo $HOST_OPTIONS_SECTIONS | cut -f1 -d" "` + if [ "$first_section" != "$iam" ]; then + echo "$iam config file has no [$iam] section" + exit 1 + fi + + grep DEV $HOST_OPTIONS |grep -qv SIZE
This will incorrectly fail the check-parallel invocation even if we have the devices defined in the comment section - so something like the following is causing a failure
local.config >>
[check-parallel]
TEST_DEV_SIZE=2G
#TEST_DEV=/dev/loop0
Maybe we should try to ignore the lines that begin with '#' something like - grep -v '^[[:space:]]*#' $HOST_OPTIONS | grep DEV | grep -qv SIZE ?
Another case:
Let us consider another local.config file:
[check-parallel]
TEST_DEV_SIZE=2G
[xfs_4k]
TEST_DEV=/dev/loop0
TEST_DIR=/mnt1/test
SCRATCH_DEV=/dev/loop1
SCRATCH_MNT=/mnt1/scratch
The above file runs fine ./check -s xfs_4k selftest/001
However, with check-parallel, it will fail and it is expected according to the design. But is there any specific reason to fail when the above configuration is perfectly suitable to run check-parallel (we have check-parallel, section defined)? So
./check-parallel -s check-parralel -D /mnt1 -x stress -t 1 selftest/001 check-parallel config file has devices defined
Instead of grepping on the entire $HOST_OPTIONS, how about we only grep on the check-parallel section? In this way we parse/export the environment variables only in check-parallel section and ignore the sections that have devices defined and/or check-parallel incompatible. ./check -s xfs_4k works anyway with [check-parallel] section defined. ./check -s check-parallel will fail with *DEVs not defined which is self explanatory.
Some something like:get_env_section $first_section | grep -v '^[[:space:]]*#' | grep DEV | grep -qv SIZE
get_env_section()
{
# code taken from parse_config_section
SECTION=$1
echo `sed -e 's/[[:space:]]*\=[[:space:]]*/=/g' \
-e 's/#.*$//' \
-e 's/[[:space:]]*$//' \
-e 's/^[[:space:]]*//' \
-e "s/^\([^=]*\)=\"\?'\?\([^\"']*\)\"\?'\?$/export \1=\"\2\"/" \
< $HOST_OPTIONS \
| sed -n -e "/^\[$SECTION\]/,/^\s*\[/{/^[^#].*\=.*/p;}"`
}The value that we are getting out of the the above suggestion modification is that we can re-use local.config file between check and check-parallel.
Please let me know what do you think of the above?
common/config-sections _config_section_setup() · line 440
Quoted 1 line
common/config-sections +1
-0+ if [ $? -ne 1 ]; then
+ echo "$iam config file has
Quoted 1 line
devices defined"
+ exit 1
+ fi
+
+ # we only need to
Quoted 1 line
pull in the config parameters here and set defaults
+ # if they are
common/config-sections _config_section_setup() · line 447
Quoted 2 lines
not set after pulling in the config values.
common/config-sections +1
-0+ parse_config_section $1
What value is $1 holding? _config_setup_parallel is being called only from check-parallel without any parameters passed, right? Did you mean _config_setup_parallel $first_section ?
--NR
Quoted 1 line
+
+ TEST_DEV_SIZE=${TEST_DEV_SIZE:=10G}
+ TEST_RTDEV_SIZE=${TES
Quoted 1 line
T_RTDEV_SIZE:=10G}
+ TEST_LOGDEV_SIZE=${TEST_LOGDEV_SIZE:=128M}
+
Quoted 1 line
SCRATCH_DEV_SIZE=${SCRATCH_DEV_SIZE:=20G}
+ SCRATCH_RTDEV_SIZE=${SC
Quoted 1 line
RATCH_RTDEV_SIZE:=20G}
+ SCRATCH_LOGDEV_SIZE=${SCRATCH_LOGDEV_SI
Quoted 1 line
ZE:=512M}
+ LOGWRITES_DEV_SIZE=${LOGWRITES_DEV_SIZE:=2G}
+
+ FSTYP
Quoted 1 line
=${FSTYP:=xfs}
+}
Re: [PATCH 13/28] check-parallel: introduce config file support
Quoted 35 lines · Nirjhar Roy
Quoted 34 lines · Dave Chinner
From: Dave Chinner <[email protected]>
check-parallel will use the same config file format check does, and use the same code to auto-discover the config file.
The biggest difference is that check-parallel -requires- the use of config sections, and the first section *must* be named "[check-parallel]". This first section is used for defining setup parameters for check parallel - loop device image file sizes, etc.
The second biggest difference is that check-parallel does not allow the config file to define devices. Any section found to contain a device definition such as
TEST_DEVorSCRATCH_DEVwill result check-parallel terminating with an error.This config file format works for check-parallel invoking check, too, because once a section is specified on the check command line, it effectively ignores unknown values set in sections that it doesn't run. Hence it effectively skips over the [check-parallel] setup section.
For check-parallel, each config section now defines just the filesystem configuration to be tested; all the usual mount and mkfs options apply, and
USE_EXTERNALmust be set for testing external devices.This commit implements the initial [check-parallel] section support and moves the build in default values for these parameters to the config file setup. This means if the config file does not contain all the necessary parameter values, a default value will be used for it.
Signed-off-by Dave Chinner <[email protected]>
....
Quoted 11 lines
common/config-sections _config_section_setup() · line 415
Quoted 9 lines
common/config-sections +5
-0@@ -388,3 +408,51 @@ _config_section_setup() fi fi } + +# check-parallel config files must: +# 1. have config sections defined +# 2. use the first config section for check-parallel setup +# 3. not define any physical device parameter in any sectionNit: Maybe some documentation of the above restrictions and some new variables introduced (like
SCRATCH_DEV_SIZE, ..)in the README?
Eventually, yes. Right now stuff is still in a state of flux, so there is no point in adding extra documentation that has to be kept up to date.
Quoted 14 lines
common/config-sections _config_section_setup() · line 428
Quoted 12 lines
common/config-sections +12
-0+# If all these are true, then we read the first section that defines +# the check-parallel config parameters and continue onwards. +_config_setup_parallel() +{ + if [ "$iam" != "check-parallel" ]; then + echo "$iam: incorrect config file format chosen!" + exit 1; + fi + + _config_file_setup + + if [ "$OPTIONS_HAVE_SECTIONS" != "true" ]; thenMost of the places the way
OPTIONS_HAVE_SECTIONSis being used is as if $OPTIONS_HAVE_SECTIONS; then
Quoted 6 lines
...fi
so, when OPTIONS_HAVE_SECTIONS=true,
/bin/trueis executed and the exit code of/bin/trueis always a success and it is not a string comparison with "true" or "false".
Yes, I know.
Executing /bin/true requires creating a new process just to return a successful exit code. This takes several milliseconds of CPU time to create and tear down a new process context.
OTOH, doing a string comparison in the shell parser takes a couple of microseconds of CPU time....
Which is faster and burns less energy?
Quoted 23 lines
/bin/truealways succeeds and hence if OPTIONS_HAVE_SECTIONS=true, then
if $OPTIONS_HAVE_SECTIONS is always executed. In the same way/bin/falsealways has a failure code upon exit.
So maybe we should change this to
if $OPTIONS_HAVE_SECTIONS; then - just to have consistency with the rest of the code?common/config-sections _config_section_setup() · line 439
Quoted 11 lines
common/config-sections +11
-0+ echo "$iam config file has no sections!" + exit 1; + fi + + local first_section=`echo $HOST_OPTIONS_SECTIONS | cut -f1 -d" "` + if [ "$first_section" != "$iam" ]; then + echo "$iam config file has no [$iam] section" + exit 1 + fi + + grep DEV $HOST_OPTIONS |grep -qv SIZEThis will incorrectly fail the check-parallel invocation even if we have the devices defined in the comment section - so something like the following is causing a failure
Quoted 4 lines
local.config >>
[check-parallel]
TEST_DEV_SIZE=2G
#TEST_DEV=/dev/loop0
Don't put devices in the check-parallel config file.
Quoted 11 lines
Maybe we should try to ignore the lines that begin with '#' something like - grep -v '^[[:space:]]*#' $HOST_OPTIONS | grep DEV | grep -qv SIZE ?
Another case:
Let us consider another local.config file:
[check-parallel]
TEST_DEV_SIZE=2G
Quoted 7 lines
[
xfs_4k]
TEST_DEV=/dev/loop0
TEST_DIR=/mnt1/test
SCRATCH_DEV=/dev/loop1
SCRATCH_MNT=/mnt1/scratchThe above file runs fine ./check -s
xfs_4kselftest/001
Yes, but it is an invalid check-parallel config file because it has devices defined in it.
If check-parallel allows this, then when it runs a check instance it will override the environment defines set by the check-parallel runner and every runner will try to use the same devices and mount points.
It just doesn't work.
Quoted 8 lines
However, with check-parallel, it will fail and it is expected according to the design. But is there any specific reason to fail when the above configuration is perfectly suitable to run check-parallel (we have check-parallel, section defined)? So
./check-parallel -s check-parralel -D /mnt1 -x stress -t 1 selftest/001 check-parallel config file has devices defined
Quoted 7 lines
Instead of grepping on the entire $HOST_OPTIONS, how about we only grep on the check-parallel section? In this way we parse/export the environment variables only in check-parallel section and ignore the sections that have devices defined and/or check-parallel incompatible. ./check -s
xfs_4kworks anyway with [check-parallel] section defined. ./check -s check-parallel will fail with *DEVs not defined which is self explanatory.
I think you may have misunderstood the direction check-parallel is going in.
I don't want check and check-parallel to share the same config file. What I need is for them to use the same file format and hence be able to share parser code.
Later in the patchset I remove the dependency that check-parallel has on check, and at that point the check-parallel config file no longer needs to work with check.
However, there is an intermediate period in the patchset where check-parallel calls check, and so the config file for check-parallel has to be compatible with check for at least that short period in time.
To make that work, the check-parallel config file cannot have any devices or mount points defined anywhere in it, otherwise check will override the values that check-parallel has provided via the environment. We don't need to over-engineer this check - if there are any known devices defined in the config file, abort.
If it gets a false positive, I don't care because once check-parallel no longer calls check, the "no devices in check-parallel config file" rule becomes irrelevant because the config files are no longer shared. At this point the config file supports only the subset of parameters that check-parallel defines as valid, and no more. This set of valid parameters for check-parallel will be enumerated in future patches.
Hence all we have to do at this point is make sure that the section parser continues to work correctly for both test runners, not try to invent the One True Config File To Rule Them All....
Quoted 4 lines
common/config-sections _config_section_setup() · line 447
Quoted 1 line
common/config-sections +1
-0+ parse_config_section $1What value is $1 holding?
_config_setup_parallelis being called only from check-parallel without any parameters passed, right? Did you mean_config_setup_parallel$first_section ?
Yup. Fixed.
-Dave.
Signature · 2 lines
Dave Chinner [email protected]
PATCH 14/28 fstests: further separate sourcing common/rc and common/config from initialisation
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
The sourcing of common/rc still causes code to be run, partially because it sources common/config and partially because there is some in-line code amongst all the function definitions inside common/rc
This is messy, and re-sourcing those files also does an awful lot of setup work that isn't actually required.
common/config only needs to be included once - everything that scripts then depend on should be exported by it, and hence it should only be included once from check/check-parallel to set up all the environmental parameters for the entire run.
common/rc also only needs to be included once per context, but it does not need to directly include common config nor does it need to run init_rc in each individual test context.
Seperate out this mess. Include common/config directly where needed and only use it to set up the environment. Move all the code that is in common/config to common/rc so that common/config is not needed for any purpose other than setting up the initial environment. Move the initialisation functions to the scripts that include common/config.
Config file and config section parsing can be run directly from check and/or check-parallel; this is not needed for every context that needs to know how what XFS_MKFS_PROG is set to...
Similarly, include common/rc only once, and only call init_rc or _source_specific_fs() from the contexts that actually need that code to be run.
--- check | 23 ++--- common/config | 212 -------------------------------------------- common/preamble | 18 +++- common/rc | 227 ++++++++++++++++++++++++++++++++++++++++++++---- 4 files changed, 240 insertions(+), 240 deletions(-)
check +12 -11
@@ -46,6 +46,9 @@ rm -f $tmp.list $tmp.tmp $tmp.grep $here/$iam.out $tmp.report.* $tmp.arglist . ./common/exit . ./common/test_names . ./common/test_list +. ./common/config +. ./common/config-sections +. ./common/rc usage() {
@@ -183,15 +186,17 @@ while [ $# -gt 0 ]; do shift done -# we need common/rc, that also sources common/config. We need to source it -# after processing args, overlay needs FSTYP set before sourcing common/config -if ! . ./common/rc; then - echo "check: failed to source common/rc" - exit 1 -fi - +# now we have done argument parsing, overlay has FSTYP set and we can now +# start processing the config files and setting up devices. +_config_section_setup +_canonicalize_devices init_rc +if [ ! -z "$REPORT_LIST" ]; then + . ./common/report + _assert_report_list +fi + # If the test config specified a soak test duration, see if there are any # unit suffixes that need converting to an integer seconds count. if [ -n "$SOAK_DURATION" ]; then
@@ -553,10 +558,6 @@ function run_section() status=1 exit fi - # Previous FSTYP derived from TEST_DEV could be changed, source - # common/rc again with correct FSTYP to get FSTYP specific configs, - # e.g. common/xfs - . common/rc _tl_prepare_test_list elif [ "$OLD_TEST_FS_MOUNT_OPTS" != "$TEST_FS_MOUNT_OPTS" ]; then # Unmount TEST_DEV to apply the updated mount options.
common/config +0 -212
221 lines not on this page · raw message
common/preamble +17 -1
@@ -51,7 +51,23 @@ _begin_fstest() . ./common/exit . ./common/rc - init_rc + + # Explicitly source the filesystem specific functions the test may need. + # This opens the door for template-file based functionality using + # function redfinition (e.g. to provide _scratch_mkfs()), rather than + # having everyting FSTYP specific being implemented in common/rc with + # massive case statements. + _source_specific_fs $FSTYP + + # Always mount the test device because there many feature checks (i.e. + # _requires_....() functions) that assume the TEST_DIR is mounted. Lots + # of tests do not call _require_test to actually mount the test device + # first, so if we don't mount the test device then the _requires... + # checks are not probing the correct filesystem for support. + _check_if_dev_already_mounted $TEST_DEV $TEST_DIR + if [ $? -eq 1 ]; then + _test_mount || _fail "Cannot mount $TEST_DEV on $TEST_DIR" + fi # remove previous $seqres.full before test rm -f $seqres.full $seqres.hints
common/rc +211 -16
256 lines not on this page · raw message
Signature · 1 line
2.45.2
Re: [PATCH 14/28] fstests: further separate sourcing common/rc and common/config from initialisation
Quoted 332 lines · Dave Chinner
From: Dave Chinner <[email protected]>
The sourcing of common/rc still causes code to be run, partially because it sources common/config and partially because there is some in-line code amongst all the function definitions inside common/rc
This is messy, and re-sourcing those files also does an awful lot of setup work that isn't actually required.
common/config only needs to be included once - everything that scripts then depend on should be exported by it, and hence it should only be included once from check/check-parallel to set up all the environmental parameters for the entire run.
common/rc also only needs to be included once per context, but it does not need to directly include common config nor does it need to run
init_rcin each individual test context.Seperate out this mess. Include common/config directly where needed and only use it to set up the environment. Move all the code that is in common/config to common/rc so that common/config is not needed for any purpose other than setting up the initial environment. Move the initialisation functions to the scripts that include common/config.
Config file and config section parsing can be run directly from check and/or check-parallel; this is not needed for every context that needs to know how what
XFS_MKFS_PROGis set to...Similarly, include common/rc only once, and only call
init_rcor_source_specific_fs()from the contexts that actually need that code to be run.Signed-off-by Dave Chinner <[email protected]>--- check | 23 ++--- common/config | 212 -------------------------------------------- common/preamble | 18 +++- common/rc | 227 ++++++++++++++++++++++++++++++++++++++++++++---- 4 files changed, 240 insertions(+), 240 deletions(-)check +12
-11@@ -46,6 +46,9 @@ rm -f $tmp.list $tmp.tmp $tmp.grep $here/$iam.out $tmp.report.* $tmp.arglist . ./common/exit . ./common/test_names . ./common/test_list +. ./common/config +. ./common/config-sections +. ./common/rc usage() {@@ -183,15 +186,17 @@ while [ $# -gt 0 ]; do shift done -# we need common/rc, that also sources common/config. We need to source it -# after processing args, overlay needs FSTYP set before sourcing common/config -if ! . ./common/rc; then - echo "check: failed to source common/rc" - exit 1 -fi - +# now we have done argument parsing, overlay has FSTYP set and we can now +# start processing the config files and setting up devices. +_config_section_setup +_canonicalize_devices init_rc +if [ ! -z "$REPORT_LIST" ]; then + . ./common/report + _assert_report_list +fi + # If the test config specified a soak test duration, see if there are any # unit suffixes that need converting to an integer seconds count. if [ -n "$SOAK_DURATION" ]; then@@ -553,10 +558,6 @@ function run_section() status=1 exit fi - # Previous FSTYP derived from TEST_DEV could be changed, source - # common/rc again with correct FSTYP to get FSTYP specific configs, - # e.g. common/xfs - . common/rc _tl_prepare_test_list elif [ "$OLD_TEST_FS_MOUNT_OPTS" != "$TEST_FS_MOUNT_OPTS" ]; then # Unmount TEST_DEV to apply the updated mount options.common/config +0
-212@@ -41,7 +41,6 @@ . common/test_names . common/exit -. common/config-sections # all tests should use a common language setting to prevent golden # output mismatches.@@ -342,214 +341,3 @@ if [ -x /usr/sbin/selinuxenabled ] && /usr/sbin/selinuxenabled; then : ${SELINUX_MOUNT_OPTIONS:="-o context=$(stat -c %C /)"} export SELINUX_MOUNT_OPTIONS fi - -_common_mount_opts() -{ - case $FSTYP in - 9p) - echo $PLAN9_MOUNT_OPTIONS - ;; - fuse) - echo $FUSE_MOUNT_OPTIONS - ;; - xfs) - echo $XFS_MOUNT_OPTIONS - ;; - udf) - echo $UDF_MOUNT_OPTIONS - ;; - nfs) - echo $NFS_MOUNT_OPTIONS - ;; - afs) - echo $AFS_MOUNT_OPTIONS - ;; - cifs) - echo $CIFS_MOUNT_OPTIONS - ;; - ceph) - echo $CEPHFS_MOUNT_OPTIONS - ;; - glusterfs) - echo $GLUSTERFS_MOUNT_OPTIONS - ;; - overlay) - echo $OVERLAY_MOUNT_OPTIONS - ;; - ext2|ext3|ext4) - # acls & xattrs aren't turned on by default on ext$FOO - echo "-o acl,user_xattr $EXT_MOUNT_OPTIONS" - ;; - f2fs) - echo "-o acl,user_xattr $F2FS_MOUNT_OPTIONS" - ;; - reiser4) - # acls & xattrs aren't supported by reiser4 - echo $REISER4_MOUNT_OPTIONS - ;; - gfs2) - # acls aren't turned on by default on gfs2 - echo "-o acl $GFS2_MOUNT_OPTIONS" - ;; - tmpfs) - # We need to specify the size at mount, use 1G by default - echo "-o size=1G $TMPFS_MOUNT_OPTIONS" - ;; - ubifs) - echo $UBIFS_MOUNT_OPTIONS - ;; - *) - ;; - esac -} - -_mount_opts() -{ - export MOUNT_OPTIONS=$(_common_mount_opts) -} - -_test_mount_opts() -{ - export TEST_FS_MOUNT_OPTS=$(_common_mount_opts) -} - -_mkfs_opts() -{ - case $FSTYP in - xfs) - export MKFS_OPTIONS=$XFS_MKFS_OPTIONS - ;; - udf) - [ ! -z "$udf_fsize" ] && \ - UDF_MKFS_OPTIONS="$UDF_MKFS_OPTIONS -s $udf_fsize" - export MKFS_OPTIONS=$UDF_MKFS_OPTIONS - ;; - nfs) - export MKFS_OPTIONS=$NFS_MKFS_OPTIONS - ;; - afs) - export MKFS_OPTIONS=$AFS_MKFS_OPTIONS - ;; - cifs) - export MKFS_OPTIONS=$CIFS_MKFS_OPTIONS - ;; - ceph) - export MKFS_OPTIONS=$CEPHFS_MKFS_OPTIONS - ;; - reiser4) - export MKFS_OPTIONS=$REISER4_MKFS_OPTIONS - ;; - gfs2) - export MKFS_OPTIONS="$GFS2_MKFS_OPTIONS -O -p lock_nolock" - ;; - jfs) - export MKFS_OPTIONS="$JFS_MKFS_OPTIONS -q" - ;; - f2fs) - export MKFS_OPTIONS="$F2FS_MKFS_OPTIONS" - ;; - btrfs) - export MKFS_OPTIONS="$BTRFS_MKFS_OPTIONS" - ;; - bcachefs) - export MKFS_OPTIONS=$BCACHEFS_MKFS_OPTIONS - ;; - *) - ;; - esac -} - -_fsck_opts() -{ - case $FSTYP in - ext2|ext3|ext4) - export FSCK_OPTIONS="-nf" - ;; - reiser*) - export FSCK_OPTIONS="--yes" - ;; - f2fs) - export FSCK_OPTIONS="" - ;; - *) - export FSCK_OPTIONS="-n" - ;; - esac -} - -# check necessary running dependences then source sepcific fs helpers -_source_specific_fs() -{ - local fs=$1 - - if [ -z "$fs" ];then - fs=$FSTYP - fi - - case "$fs" in - xfs) - [ "$XFS_LOGPRINT_PROG" = "" ] && _fatal "xfs_logprint not found" - [ "$XFS_REPAIR_PROG" = "" ] && _fatal "xfs_repair not found" - [ "$XFS_DB_PROG" = "" ] && _fatal "xfs_db not found" - [ "$MKFS_XFS_PROG" = "" ] && _fatal "mkfs_xfs not found" - [ "$XFS_INFO_PROG" = "" ] && _fatal "xfs_info not found" - - . ./common/xfs - ;; - udf) - [ "$MKFS_UDF_PROG" = "" ] && _fatal "mkfs_udf/mkudffs not found" - ;; - btrfs) - [ "$MKFS_BTRFS_PROG" = "" ] && _fatal "mkfs.btrfs not found" - - . ./common/btrfs - ;; - ext4) - [ "$MKFS_EXT4_PROG" = "" ] && _fatal "mkfs.ext4 not found" - . ./common/ext4 - ;; - ext2|ext3) - . ./common/ext4 - ;; - f2fs) - [ "$MKFS_F2FS_PROG" = "" ] && _fatal "mkfs.f2fs not found" - ;; - nfs) - . ./common/nfs - ;; - afs) - ;; - cifs) - ;; - 9p) - ;; - fuse) - ;; - ceph) - . ./common/ceph - ;; - glusterfs) - ;; - overlay) - . ./common/overlay - ;; - reiser4) - [ "$MKFS_REISER4_PROG" = "" ] && _fatal "mkfs.reiser4 not found" - ;; - pvfs2) - ;; - ubifs) - [ "$UBIUPDATEVOL_PROG" = "" ] && _fatal "ubiupdatevol not found" - . ./common/ubifs - ;; - esac -} - -_config_section_setup -_canonicalize_devices -# mkfs.xfs checks for TEST_DEV before permitting < 300M filesystems. TEST_DIR -# and QA_CHECK_FS are also checked by mkfs.xfs, but already exported elsewhere. -export TEST_DEV - -# make sure this script returns success -/bin/truecommon/preamble +5
-1@@ -51,7 +51,23 @@ _begin_fstest() . ./common/exit . ./common/rc - init_rc + + # Explicitly source the filesystem specific functions the test may need. + # This opens the door for template-file based functionality using + # function redfinition (e.g. to provide _scratch_mkfs()), rather than + # having everyting FSTYP specific being implemented in common/rc with
common/preamble _begin_fstest() · line 68
Quoted 10 lines
common/preamble +10
-0+ # massive case statements. + _source_specific_fs $FSTYP + + # Always mount the test device because there many feature checks (i.e. + # _requires_....() functions) that assume the TEST_DIR is mounted. Lots + # of tests do not call _require_test to actually mount the test device + # first, so if we don't mount the test device then the _requires... + # checks are not probing the correct filesystem for support. + _check_if_dev_already_mounted $TEST_DEV $TEST_DIR + if [ $? -eq 1 ]; then
_check_if_dev_already_mounted can return 1 (not mounted) or 2 (wrong mount) - in that case shouldn't we check for $? -ne 0 ?
Also, TEST_DEV is always mounted by "check", right? Is this just a paranoic check?
Quoted 260 lines
common/preamble +2
-0+ _test_mount || _fail "Cannot mount $TEST_DEV on $TEST_DIR" + fi # remove previous $seqres.full before test rm -f $seqres.full $seqres.hintscommon/rc +201
-16@@ -2,10 +2,11 @@ # SPDX-License-Identifier: GPL-2.0+ # Copyright (c) 2000-2006 Silicon Graphics, Inc. All Rights Reserved. -. common/config - BC="$(type -P bc)" || BC= +# make sure we have a standard umask +umask 022 + # Don't use sync(1) directly if at all possible. In most cases we only need to # sync the fs under test, so we use syncfs if it is supported to prevent # disturbance of other tests that may be running concurrently.@@ -246,17 +247,6 @@ _log_err() echo "(see $seqres.full for details)" } -# make sure we have a standard umask -umask 022 - -# check for correct setup and source the $FSTYP specific functions now -_source_specific_fs $FSTYP - -if [ ! -z "$REPORT_LIST" ]; then - . ./common/report - _assert_report_list -fi - _get_filesize() { stat -c %s "$1"@@ -4934,6 +4924,8 @@ init_rc() _exit 1 fi + _source_specific_fs $FSTYP + # if $TEST_DEV is not mounted, mount it now as XFS if [ -z "`_fs_type $TEST_DEV`" ] then@@ -4973,6 +4965,11 @@ init_rc() # it is supported. $XFS_IO_PROG -i -c quit 2>/dev/null && \ export XFS_IO_PROG="$XFS_IO_PROG -i" + + # mkfs.xfs checks for TEST_DEV before permitting < 300M filesystems. + # TEST_DIR and QA_CHECK_FS are also checked by mkfs.xfs, but already + # exported elsewhere. + export TEST_DEV } # get real device path name by following link@@ -5844,6 +5841,204 @@ _require_program() { _have_program "$1" || _notrun "$tag required" } -################################################################################ -# make sure this script returns success -/bin/true +_common_mount_opts() +{ + case $FSTYP in + 9p) + echo $PLAN9_MOUNT_OPTIONS + ;; + fuse) + echo $FUSE_MOUNT_OPTIONS + ;; + xfs) + echo $XFS_MOUNT_OPTIONS + ;; + udf) + echo $UDF_MOUNT_OPTIONS + ;; + nfs) + echo $NFS_MOUNT_OPTIONS + ;; + afs) + echo $AFS_MOUNT_OPTIONS + ;; + cifs) + echo $CIFS_MOUNT_OPTIONS + ;; + ceph) + echo $CEPHFS_MOUNT_OPTIONS + ;; + glusterfs) + echo $GLUSTERFS_MOUNT_OPTIONS + ;; + overlay) + echo $OVERLAY_MOUNT_OPTIONS + ;; + ext2|ext3|ext4) + # acls & xattrs aren't turned on by default on ext$FOO + echo "-o acl,user_xattr $EXT_MOUNT_OPTIONS" + ;; + f2fs) + echo "-o acl,user_xattr $F2FS_MOUNT_OPTIONS" + ;; + reiser4) + # acls & xattrs aren't supported by reiser4 + echo $REISER4_MOUNT_OPTIONS + ;; + gfs2) + # acls aren't turned on by default on gfs2 + echo "-o acl $GFS2_MOUNT_OPTIONS" + ;; + tmpfs) + # We need to specify the size at mount, use 1G by default + echo "-o size=1G $TMPFS_MOUNT_OPTIONS" + ;; + ubifs) + echo $UBIFS_MOUNT_OPTIONS + ;; + *) + ;; + esac +} + +_mount_opts() +{ + export MOUNT_OPTIONS=$(_common_mount_opts) +} + +_test_mount_opts() +{ + export TEST_FS_MOUNT_OPTS=$(_common_mount_opts) +} + +_mkfs_opts() +{ + case $FSTYP in + xfs) + export MKFS_OPTIONS=$XFS_MKFS_OPTIONS + ;; + udf) + [ ! -z "$udf_fsize" ] && \ + UDF_MKFS_OPTIONS="$UDF_MKFS_OPTIONS -s $udf_fsize" + export MKFS_OPTIONS=$UDF_MKFS_OPTIONS + ;; + nfs) + export MKFS_OPTIONS=$NFS_MKFS_OPTIONS + ;; + afs) + export MKFS_OPTIONS=$AFS_MKFS_OPTIONS + ;; + cifs) + export MKFS_OPTIONS=$CIFS_MKFS_OPTIONS + ;; + ceph) + export MKFS_OPTIONS=$CEPHFS_MKFS_OPTIONS + ;; + reiser4) + export MKFS_OPTIONS=$REISER4_MKFS_OPTIONS + ;; + gfs2) + export MKFS_OPTIONS="$GFS2_MKFS_OPTIONS -O -p lock_nolock" + ;; + jfs) + export MKFS_OPTIONS="$JFS_MKFS_OPTIONS -q" + ;; + f2fs) + export MKFS_OPTIONS="$F2FS_MKFS_OPTIONS" + ;; + btrfs) + export MKFS_OPTIONS="$BTRFS_MKFS_OPTIONS" + ;; + bcachefs) + export MKFS_OPTIONS=$BCACHEFS_MKFS_OPTIONS + ;; + *) + ;; + esac +} + +_fsck_opts() +{ + case $FSTYP in + ext2|ext3|ext4) + export FSCK_OPTIONS="-nf" + ;; + reiser*) + export FSCK_OPTIONS="--yes" + ;; + f2fs) + export FSCK_OPTIONS="" + ;; + *) + export FSCK_OPTIONS="-n" + ;; + esac +} + +# check necessary running dependences then source sepcific fs helpers +_source_specific_fs() +{ + local fs=$1 + + if [ -z "$fs" ];then + fs=$FSTYP + fi + + case "$fs" in + xfs) + [ "$XFS_LOGPRINT_PROG" = "" ] && _fatal "xfs_logprint not found" + [ "$XFS_REPAIR_PROG" = "" ] && _fatal "xfs_repair not found" + [ "$XFS_DB_PROG" = "" ] && _fatal "xfs_db not found" + [ "$MKFS_XFS_PROG" = "" ] && _fatal "mkfs_xfs not found" + [ "$XFS_INFO_PROG" = "" ] && _fatal "xfs_info not found" + + . ./common/xfs + ;; + udf) + [ "$MKFS_UDF_PROG" = "" ] && _fatal "mkfs_udf/mkudffs not found" + ;; + btrfs) + [ "$MKFS_BTRFS_PROG" = "" ] && _fatal "mkfs.btrfs not found" + + . ./common/btrfs + ;; + ext4) + [ "$MKFS_EXT4_PROG" = "" ] && _fatal "mkfs.ext4 not found" + . ./common/ext4 + ;; + ext2|ext3) + . ./common/ext4 + ;; + f2fs) + [ "$MKFS_F2FS_PROG" = "" ] && _fatal "mkfs.f2fs not found" + ;; + nfs) + . ./common/nfs + ;; + afs) + ;; + cifs) + ;; + 9p) + ;; + fuse) + ;; + ceph) + . ./common/ceph + ;; + glusterfs) + ;; + overlay) + . ./common/overlay + ;; + reiser4)
common/rc _require_program() { · line 6044
Quoted 10 lines
common/rc +10
-0+ [ "$MKFS_REISER4_PROG" = "" ] && _fatal "mkfs.reiser4 not found" + ;; + pvfs2) + ;; + ubifs) + [ "$UBIUPDATEVOL_PROG" = "" ] && _fatal "ubiupdatevol not found" + . ./common/ubifs + ;; + esac +}
So the logic behind moving the above functions from common/config to common/rc is because we want the configuration work done only once (functions in common/config) and so we include common/config only once per check/check-parallel context and the above functions (like _source_specific_fs(), _fsck_opts() etc) are needed for each test - so we factor them into common/rc and include common/rc per test instead of common/config which is not needed to be sourced for each test, am I correct? --NR
PATCH 15/28 check-parallel: de-batch test execution
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
To improve how check-parallel runs tests, it needs to run tests directly from the runner threads. We currently batch them based on runtime before we execture any tests, but this results in runner 0 always having a test list with runtime longer than the test list for runner N.
As a result, we can end up with higher numbered runners finishing all their tests before runner 0 has even finished the first test it was given to run. Hence we end up with check-parallel starting with maximum concurrency, but the test concurrency reduces as the run goes on.
To fix this, we need a dynamic test list such that each runner only needs to be scheduled to run a single test at a time. When they have finished the current test, they can pop the next test to run off the time ordered stack and execute that. Hence test runners won't stop running until there are no more tests to run, hence maximising concurrency across the entire test run.
To do this, we first need a test list mechanism that is safe for concurrent destacking from multiple test runners. We place the test list in a temporary file, then use file locks to serialise access to the temporary file.
We order the list in the test file from lowest runtime to highest. This means that running tests from longest to shortest runtime destacks from the end fo the file. This means that the next test to run is always the last line fo the file and we can simply use truncation based mechanisms to consume the test during destacking.
Running tests individually via check like this is inefficient as there is a lot of check setup and initialisation overhead. However, by increasing the utilisation of the test runner threads, overall runtime of check-parallel does not increase with this change. Reduction of this repeated overhead will also be addressed in future patches.
--- check-parallel | 75 +++++++++++++++++++++++++++++--------------------- 1 file changed, 43 insertions(+), 32 deletions(-)
check-parallel +43 -32
@@ -18,6 +18,7 @@ run_section="" iam="check-parallel" tmp=/tmp/check-parallel.$$ +test_list="$tmp.test_list" . ./common/exit . ./common/test_names
@@ -150,9 +151,6 @@ if [ -d "$basedir/runner-0/" ]; then prev_results=`ls -tr $basedir/runner-0/ | grep results | tail -1` fi -_tl_prepare_test_list -_tl_strip_test_list - # grab all previously run tests and order them from highest runtime to lowest # We are going to try to run the longer tests first, hopefully so we can avoid # massive thundering herds trying to run lots of really short tests in parallel
@@ -198,22 +196,22 @@ if ! $_tl_randomise -a ! $_tl_exact_order; then fi fi -# split the list amongst N runners -split_runner_list() +# Grab the next test to be run from the tail of the file. +# Returns an empty string if there is no tests remaining to run. +# File operations are run under flock so concurrent gets are serialised against +# each other. +get_next_test() { - local ix - local rx - local -a _list=( $_tl_tests ) - for ((ix = 0; ix < ${#_list[*]}; ix++)); do - seq="${_list[$ix]}" - rx=$((ix % $runners)) - if ! _tl_expunge_test $seq; then - runner_list[$rx]+="${_list[$ix]} " - fi - #echo $seq - done + local test= + + flock 99 + test=$(tail -1 $test_list) + sed -i "\,$test,d" $test_list + flock -u 99 + echo $test } + _create_loop_device() { local file=$1 dev
@@ -240,6 +238,8 @@ _destroy_loop_device() runner_go() { + exec 99<>$tmp.test_list_lock + local id=$1 local me=$basedir/runner-$id local _test=$me/test.img
@@ -250,6 +250,7 @@ runner_go() local _scratch_log=$me/scratch-log.img local _logwrites=$me/logwrites.img local _results=$me/results-$2 + local test_to_run=$(get_next_test) mkdir -p $me
@@ -291,7 +292,15 @@ runner_go() # Similarly, we need to run check in it's own PID namespace so that # operations like pkill only affect the runner instance, not globally # kill processes from other check instances. - tools/run_privatens ./check $run_section -x unreliable_in_parallel --exact-order ${runner_list[$id]} >> $me/log 2>&1 + while [ -n "$test_to_run" ]; do + echo "Runner $id: running test $test_to_run" + unset FSTESTS_ISOL + if ! _tl_expunge_test $test_to_run; then + tools/run_privatens ./check $run_section $test_to_run >> $me/log 2>&1 + fi + + test_to_run=$(get_next_test) + done wait sleep 1
@@ -320,20 +329,32 @@ cleanup() umount -R $basedir/*/test 2> /dev/null umount -R $basedir/*/scratch 2> /dev/null losetup --detach-all + rm -rf $tmp.* } trap "cleanup; exit" HUP INT QUIT TERM _config_setup_parallel -split_runner_list +_tl_setup_exclude_group "unreliable_in_parallel" +_tl_prepare_test_list +_tl_strip_test_list + +if ! $_tl_randomise -a ! $_tl_exact_order; then + if [ -f $basedir/runner-0/$prev_results/check.time ]; then + time_order_test_list + fi +fi + +# reverse the order of tests so that the get_next_test() can pull from the file +# tail rather than the head. +echo $_tl_tests |sed -e 's/ /\n/g' | tac > $test_list if [ -n "$show_test_list" ]; then echo Time ordered test list: - echo $_tl_tests - echo + cat $test_list + exit 0 fi - # Each parallel test runner needs to only see it's own mount points. If we # leave the basedir as shared, then all tests see all mounts and then we get # mount propagation issues cropping up. For example, cloning a new mount
@@ -349,20 +370,10 @@ mount --make-private $basedir now=`date +%Y-%m-%d-%H:%M:%S` for ((i = 0; i < $runners; i++)); do - - if [ -n "$show_test_list" ]; then - echo "Runner $i: ${runner_list[$i]}" - else - runner_go $i $now & - fi - + runner_go $i $now & done; wait -if [ -n "$show_test_list" ]; then - exit 0 -fi - echo -n "Tests run: " grep Ran $basedir/*/log | sed -e 's,^.*:,,' -e 's, ,\n,g' | sort | uniq | wc -l
Signature · 1 line
2.45.2
Re: [PATCH 15/28] check-parallel: de-batch test execution
Quoted 158 lines · Dave Chinner
From: Dave Chinner <[email protected]>
To improve how check-parallel runs tests, it needs to run tests directly from the runner threads. We currently batch them based on runtime before we execture any tests, but this results in runner 0 always having a test list with runtime longer than the test list for runner N.
As a result, we can end up with higher numbered runners finishing all their tests before runner 0 has even finished the first test it was given to run. Hence we end up with check-parallel starting with maximum concurrency, but the test concurrency reduces as the run goes on.
To fix this, we need a dynamic test list such that each runner only needs to be scheduled to run a single test at a time. When they have finished the current test, they can pop the next test to run off the time ordered stack and execute that. Hence test runners won't stop running until there are no more tests to run, hence maximising concurrency across the entire test run.
To do this, we first need a test list mechanism that is safe for concurrent destacking from multiple test runners. We place the test list in a temporary file, then use file locks to serialise access to the temporary file.
We order the list in the test file from lowest runtime to highest. This means that running tests from longest to shortest runtime destacks from the end fo the file. This means that the next test to run is always the last line fo the file and we can simply use truncation based mechanisms to consume the test during destacking.
Running tests individually via check like this is inefficient as there is a lot of check setup and initialisation overhead. However, by increasing the utilisation of the test runner threads, overall runtime of check-parallel does not increase with this change. Reduction of this repeated overhead will also be addressed in future patches.
Signed-off-by Dave Chinner <[email protected]>--- check-parallel | 75 +++++++++++++++++++++++++++++--------------------- 1 file changed, 43 insertions(+), 32 deletions(-)check-parallel +36
-18@@ -18,6 +18,7 @@ run_section="" iam="check-parallel" tmp=/tmp/check-parallel.$$ +test_list="$tmp.test_list" . ./common/exit . ./common/test_names@@ -150,9 +151,6 @@ if [ -d "$basedir/runner-0/" ]; then prev_results=`ls -tr $basedir/runner-0/ | grep results | tail -1` fi -_tl_prepare_test_list -_tl_strip_test_list - # grab all previously run tests and order them from highest runtime to lowest # We are going to try to run the longer tests first, hopefully so we can avoid # massive thundering herds trying to run lots of really short tests in parallel@@ -198,22 +196,22 @@ if ! $_tl_randomise -a ! $_tl_exact_order; then fi fi -# split the list amongst N runners -split_runner_list() +# Grab the next test to be run from the tail of the file. +# Returns an empty string if there is no tests remaining to run. +# File operations are run under flock so concurrent gets are serialised against +# each other. +get_next_test() { - local ix - local rx - local -a _list=( $_tl_tests ) - for ((ix = 0; ix < ${#_list[*]}; ix++)); do - seq="${_list[$ix]}" - rx=$((ix % $runners)) - if ! _tl_expunge_test $seq; then - runner_list[$rx]+="${_list[$ix]} " - fi - #echo $seq - done + local test= + + flock 99 + test=$(tail -1 $test_list) + sed -i "\,$test,d" $test_list + flock -u 99 + echo $test } + _create_loop_device() { local file=$1 dev@@ -240,6 +238,8 @@ _destroy_loop_device() runner_go() { + exec 99<>$tmp.test_list_lock + local id=$1 local me=$basedir/runner-$id local _test=$me/test.img@@ -250,6 +250,7 @@ runner_go() local _scratch_log=$me/scratch-log.img local _logwrites=$me/logwrites.img local _results=$me/results-$2 + local test_to_run=$(get_next_test) mkdir -p $me@@ -291,7 +292,15 @@ runner_go() # Similarly, we need to run check in it's own PID namespace so that # operations like pkill only affect the runner instance, not globally # kill processes from other check instances. - tools/run_privatens ./check $run_section -x unreliable_in_parallel --exact-order ${runner_list[$id]} >> $me/log 2>&1 + while [ -n "$test_to_run" ]; do + echo "Runner $id: running test $test_to_run" + unset FSTESTS_ISOL + if ! _tl_expunge_test $test_to_run; then + tools/run_privatens ./check $run_section $test_to_run >> $me/log 2>&1 + fi + + test_to_run=$(get_next_test) + done wait sleep 1@@ -320,20 +329,32 @@ cleanup() umount -R $basedir/*/test 2> /dev/null umount -R $basedir/*/scratch 2> /dev/null losetup --detach-all + rm -rf $tmp.* } trap "cleanup; exit" HUP INT QUIT TERM _config_setup_parallel -split_runner_list +_tl_setup_exclude_group "unreliable_in_parallel" +_tl_prepare_test_list +_tl_strip_test_list + +if ! $_tl_randomise -a ! $_tl_exact_order; then + if [ -f $basedir/runner-0/$prev_results/check.time ]; then + time_order_test_list + fi +fi
check-parallel cleanup() · line 355
Quoted 10 lines
check-parallel +6
-2+ +# reverse the order of tests so that the get_next_test() can pull from the file +# tail rather than the head. +echo $_tl_tests |sed -e 's/ /\n/g' | tac > $test_list if [ -n "$show_test_list" ]; then echo Time ordered test list: - echo $_tl_tests - echo + cat $test_list + exit 0
_exit 0?
Looks good otherwise. I think this change will extract the maximum concurrency since we are kind of uniformly distrubuting the long running tests instead of flooding runner-0 with the top n slow running tests.
--NR
Quoted 28 lines
check-parallel +1
-12fi - # Each parallel test runner needs to only see it's own mount points. If we # leave the basedir as shared, then all tests see all mounts and then we get # mount propagation issues cropping up. For example, cloning a new mount@@ -349,20 +370,10 @@ mount --make-private $basedir now=`date +%Y-%m-%d-%H:%M:%S` for ((i = 0; i < $runners; i++)); do - - if [ -n "$show_test_list" ]; then - echo "Runner $i: ${runner_list[$i]}" - else - runner_go $i $now & - fi - + runner_go $i $now & done; wait -if [ -n "$show_test_list" ]; then - exit 0 -fi - echo -n "Tests run: " grep Ran $basedir/*/log | sed -e 's,^.*:,,' -e 's, ,\n,g' | sort | uniq | wc -l
PATCH 16/28 check-parallel: run sections directly
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
Currently we pass the section through to check for it to apply and run. However, we do not reconfigure the test or external devices in check-parallel after the initial setup, so thie results in devices that may be incorrectly configured for the given section config we want to run.
To fix this, we need to iterate the sections directly in check-parallel and reconfigure the runner device setup between each section that is run. This allows the test device and external devices to be set up correctly for each section config.
As the test list is consumed as we walk it, we need to reset the test list for each section that we run.
We still pass the config section through to check so that it can also source the correct config from the section we are running.
We also add section exclusion support so that we skip sections the same way that check currently does.
--- check-parallel | 118 ++++++++++++++++++++++++++++------------- common/config-sections | 3 ++ 2 files changed, 84 insertions(+), 37 deletions(-)
check-parallel +81 -37
@@ -15,6 +15,7 @@ runner_list=() runtimes=() show_test_list= run_section="" +exclude_section="" iam="check-parallel" tmp=/tmp/check-parallel.$$
@@ -119,7 +120,8 @@ while [ $# -gt 0 ]; do -f) is_supported_fstype $2 ; export FSTYP=$2; shift ;; - -s) run_section="$run_section -s $2"; shift ;; + -s) run_section="$run_section $2"; shift ;; + -S) exclude_section="$exclude_section $2"; shift ;; -*) usage ;; *) # not an argument, we've got tests now.
@@ -202,11 +204,12 @@ fi # each other. get_next_test() { + local test_file="$test_list.$1" local test= flock 99 - test=$(tail -1 $test_list) - sed -i "\,$test,d" $test_list + test=$(tail -1 $test_file) + sed -i "\,$test,d" $test_file flock -u 99 echo $test }
@@ -236,10 +239,34 @@ _destroy_loop_device() losetup -d $dev || _fail "Cannot destroy loop device $dev" } -runner_go() +run_tests() { + local section="$1" + exec 99<>$tmp.test_list_lock + local test_to_run=$(get_next_test $section) + + # Run the tests in it's own mount namespace, as per the comment below + # that precedes making the basedir a private mount. + # + # Similarly, we need to run check in it's own PID namespace so that + # operations like pkill only affect the runner instance, not globally + # kill processes from other check instances. + while [ -n "$test_to_run" ]; do + echo -n " $test_to_run " + unset FSTESTS_ISOL + if ! _tl_expunge_test $test_to_run; then + tools/run_privatens ./check -s $section $test_to_run >> $me/log 2>&1 + fi + + test_to_run=$(get_next_test $section) + done +} + +runner_go() +{ + local id=$1 local me=$basedir/runner-$id local _test=$me/test.img
@@ -250,7 +277,7 @@ runner_go() local _scratch_log=$me/scratch-log.img local _logwrites=$me/logwrites.img local _results=$me/results-$2 - local test_to_run=$(get_next_test) + local section=$3 mkdir -p $me
@@ -286,21 +313,7 @@ runner_go() # export DUMP_CORRUPT_FS=1 - # Run the tests in it's own mount namespace, as per the comment below - # that precedes making the basedir a private mount. - # - # Similarly, we need to run check in it's own PID namespace so that - # operations like pkill only affect the runner instance, not globally - # kill processes from other check instances. - while [ -n "$test_to_run" ]; do - echo "Runner $id: running test $test_to_run" - unset FSTESTS_ISOL - if ! _tl_expunge_test $test_to_run; then - tools/run_privatens ./check $run_section $test_to_run >> $me/log 2>&1 - fi - - test_to_run=$(get_next_test) - done + run_tests $section wait sleep 1
@@ -322,6 +335,44 @@ runner_go() } +run_section() +{ + local section="$1" + local now="$2" + local i + + echo $run_section |grep -qw $section || return + echo $exclude_section |grep -qw $section && return + + echo + echo Running section: $section + echo + + parse_config_section $section + + # set up consumable test list first + cp $test_list $test_list.$section + for ((i = 0; i < $runners; i++)); do + runner_go $i $now $section & + done + wait + + echo + echo Section: $section + echo -n "Tests run: " + grep Ran $basedir/*/log | sed -e 's,^.*:,,' -e 's, ,\n,g' | sort | uniq | wc -l + + echo -n "Tests _notrun: " + grep "^Not run" $basedir/*/log | uniq | sed -e 's,^.*:,,' -e 's, ,\n,g' -e 's,^\n,,' | wc -l + + echo -n "Failure count: " + grep Failures: $basedir/*/log | uniq | sed -e "s/^.*Failures://" -e "s,\([0-9]\) \([gx]\),\1\n \2,g" |wc -l + echo + + echo Ten slowest tests - runtime in seconds: + cat $basedir/*/results-$now/check.time | sort -k 2 -nr | head -10 +} + cleanup() { killall -INT -q check
@@ -336,6 +387,8 @@ trap "cleanup; exit" HUP INT QUIT TERM _config_setup_parallel +run_section=${run_section:="$HOST_OPTIONS_SECTIONS"} + _tl_setup_exclude_group "unreliable_in_parallel" _tl_prepare_test_list _tl_strip_test_list
@@ -369,23 +422,14 @@ fi mount --make-private $basedir now=`date +%Y-%m-%d-%H:%M:%S` -for ((i = 0; i < $runners; i++)); do - runner_go $i $now & -done; -wait - -echo -n "Tests run: " -grep Ran $basedir/*/log | sed -e 's,^.*:,,' -e 's, ,\n,g' | sort | uniq | wc -l - -echo -n "Tests _notrun: " -grep "^Not run" $basedir/*/log | uniq | sed -e 's,^.*:,,' -e 's, ,\n,g' -e 's,^\n,,' | wc -l - -echo -n "Failure count: " -grep Failures: $basedir/*/log | uniq | sed -e "s/^.*Failures://" -e "s,\([0-9]\) \([gx]\),\1\n \2,g" |wc -l -echo - -echo Ten slowest tests - runtime in seconds: -cat $basedir/*/results-$now/check.time | sort -k 2 -nr | head -10 +for section in $HOST_OPTIONS_SECTIONS; do + run_section $section $now + if [ "$sum_bad" != 0 ] && [ "$istop" = true ]; then + interrupt=false + status=`expr $sum_bad != 0` + exit + fi +done echo echo Cleanup on Aisle 5?
common/config-sections +3 -0
@@ -436,6 +436,9 @@ _config_setup_parallel() exit 1 fi + # strip check-parallel from the sections to run + export HOST_OPTIONS_SECTIONS=`echo $HOST_OPTIONS_SECTIONS | sed -e "s/$iam//"` + grep DEV $HOST_OPTIONS |grep -qv SIZE if [ $? -ne 1 ]; then echo "$iam config file has devices defined"
Signature · 1 line
2.45.2
Re: [PATCH 16/28] check-parallel: run sections directly
Quoted 4 lines · Dave Chinner
From: Dave Chinner <[email protected]>
Currently we pass the section through to check for it to apply and run. However, we do not reconfigure the test or external devices in
If we are passing ./check -s <s1> -s <s2>, the devices will be reconfigured, right? I was able to understand the implementation in this patch but I am not able to figure out the reason. Could please explain "results in devices that may be incorrectly configured for the given section config we want to run."
Basically instead of check issuing multiple ./check -s <s1> -s <s2> ..., what we are doing is we are issuing ./check -s <s1> # one section to ./check and each such ./check invocation is being made in parallel - but what is wrong with multiple ./check -s <s1> -s <s2> in parallel?
Quoted 229 lines
check-parallel after the initial setup, so thie results in devices that may be incorrectly configured for the given section config we want to run.
To fix this, we need to iterate the sections directly in check-parallel and reconfigure the runner device setup between each section that is run. This allows the test device and external devices to be set up correctly for each section config.
As the test list is consumed as we walk it, we need to reset the test list for each section that we run.
We still pass the config section through to check so that it can also source the correct config from the section we are running.
We also add section exclusion support so that we skip sections the same way that check currently does.
Signed-off-by Dave Chinner <[email protected]>--- check-parallel | 118 ++++++++++++++++++++++++++++------------- common/config-sections | 3 ++ 2 files changed, 84 insertions(+), 37 deletions(-)check-parallel +81
-37@@ -15,6 +15,7 @@ runner_list=() runtimes=() show_test_list= run_section="" +exclude_section="" iam="check-parallel" tmp=/tmp/check-parallel.$$@@ -119,7 +120,8 @@ while [ $# -gt 0 ]; do -f) is_supported_fstype $2 ; export FSTYP=$2; shift ;; - -s) run_section="$run_section -s $2"; shift ;; + -s) run_section="$run_section $2"; shift ;; + -S) exclude_section="$exclude_section $2"; shift ;; -*) usage ;; *) # not an argument, we've got tests now.@@ -202,11 +204,12 @@ fi # each other. get_next_test() { + local test_file="$test_list.$1" local test= flock 99 - test=$(tail -1 $test_list) - sed -i "\,$test,d" $test_list + test=$(tail -1 $test_file) + sed -i "\,$test,d" $test_file flock -u 99 echo $test }@@ -236,10 +239,34 @@ _destroy_loop_device() losetup -d $dev || _fail "Cannot destroy loop device $dev" } -runner_go() +run_tests() { + local section="$1" + exec 99<>$tmp.test_list_lock + local test_to_run=$(get_next_test $section) + + # Run the tests in it's own mount namespace, as per the comment below + # that precedes making the basedir a private mount. + # + # Similarly, we need to run check in it's own PID namespace so that + # operations like pkill only affect the runner instance, not globally + # kill processes from other check instances. + while [ -n "$test_to_run" ]; do + echo -n " $test_to_run " + unset FSTESTS_ISOL + if ! _tl_expunge_test $test_to_run; then + tools/run_privatens ./check -s $section $test_to_run >> $me/log 2>&1 + fi + + test_to_run=$(get_next_test $section) + done +} + +runner_go() +{ + local id=$1 local me=$basedir/runner-$id local _test=$me/test.img@@ -250,7 +277,7 @@ runner_go() local _scratch_log=$me/scratch-log.img local _logwrites=$me/logwrites.img local _results=$me/results-$2 - local test_to_run=$(get_next_test) + local section=$3 mkdir -p $me@@ -286,21 +313,7 @@ runner_go() # export DUMP_CORRUPT_FS=1 - # Run the tests in it's own mount namespace, as per the comment below - # that precedes making the basedir a private mount. - # - # Similarly, we need to run check in it's own PID namespace so that - # operations like pkill only affect the runner instance, not globally - # kill processes from other check instances. - while [ -n "$test_to_run" ]; do - echo "Runner $id: running test $test_to_run" - unset FSTESTS_ISOL - if ! _tl_expunge_test $test_to_run; then - tools/run_privatens ./check $run_section $test_to_run >> $me/log 2>&1 - fi - - test_to_run=$(get_next_test) - done + run_tests $section wait sleep 1@@ -322,6 +335,44 @@ runner_go() } +run_section() +{ + local section="$1" + local now="$2" + local i + + echo $run_section |grep -qw $section || return + echo $exclude_section |grep -qw $section && return + + echo + echo Running section: $section + echo + + parse_config_section $section + + # set up consumable test list first + cp $test_list $test_list.$section + for ((i = 0; i < $runners; i++)); do + runner_go $i $now $section & + done + wait + + echo + echo Section: $section + echo -n "Tests run: " + grep Ran $basedir/*/log | sed -e 's,^.*:,,' -e 's, ,\n,g' | sort | uniq | wc -l + + echo -n "Tests _notrun: " + grep "^Not run" $basedir/*/log | uniq | sed -e 's,^.*:,,' -e 's, ,\n,g' -e 's,^\n,,' | wc -l + + echo -n "Failure count: " + grep Failures: $basedir/*/log | uniq | sed -e "s/^.*Failures://" -e "s,\([0-9]\) \([gx]\),\1\n \2,g" |wc -l + echo + + echo Ten slowest tests - runtime in seconds: + cat $basedir/*/results-$now/check.time | sort -k 2 -nr | head -10 +} + cleanup() { killall -INT -q check@@ -336,6 +387,8 @@ trap "cleanup; exit" HUP INT QUIT TERM _config_setup_parallel +run_section=${run_section:="$HOST_OPTIONS_SECTIONS"} + _tl_setup_exclude_group "unreliable_in_parallel" _tl_prepare_test_list _tl_strip_test_list@@ -369,23 +422,14 @@ fi mount --make-private $basedir now=`date +%Y-%m-%d-%H:%M:%S` -for ((i = 0; i < $runners; i++)); do - runner_go $i $now & -done; -wait - -echo -n "Tests run: " -grep Ran $basedir/*/log | sed -e 's,^.*:,,' -e 's, ,\n,g' | sort | uniq | wc -l - -echo -n "Tests _notrun: " -grep "^Not run" $basedir/*/log | uniq | sed -e 's,^.*:,,' -e 's, ,\n,g' -e 's,^\n,,' | wc -l - -echo -n "Failure count: " -grep Failures: $basedir/*/log | uniq | sed -e "s/^.*Failures://" -e "s,\([0-9]\) \([gx]\),\1\n \2,g" |wc -l -echo - -echo Ten slowest tests - runtime in seconds: -cat $basedir/*/results-$now/check.time | sort -k 2 -nr | head -10 +for section in $HOST_OPTIONS_SECTIONS; do + run_section $section $now + if [ "$sum_bad" != 0 ] && [ "$istop" = true ]; then + interrupt=false + status=`expr $sum_bad != 0` + exit + fi +done echo echo Cleanup on Aisle 5?common/config-sections +3
-0@@ -436,6 +436,9 @@ _config_setup_parallel() exit 1 fi + # strip check-parallel from the sections to run + export HOST_OPTIONS_SECTIONS=`echo $HOST_OPTIONS_SECTIONS | sed -e "s/$iam//"` + grep DEV $HOST_OPTIONS |grep -qv SIZE if [ $? -ne 1 ]; then echo "$iam config file has devices defined"
PATCH 17/28 check-parallel: rebuild test list when FSTYP changes
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
When a config section changes FSTYP, we do not rebuild the test list that is to be run. This means that if we go from xfs to ext4, the ext4 section will still try to run all the XFS tests, and won't try to run the ext4 tests.
To fix this for check-parallel, check if the FSTYP has changed and if it has strip the old tests from the test list and add the new tests back to it.
This is effectively a zero-day bug in the check config section support code.
While there, remain the internal _tl_file variable and the filename used to hold command line tests. The check-parallel change uncovered that it was using the same filename ($tmp.test_list) as the test list code was using and they were stepping on each other.
--- check-parallel | 81 +++++++++++++++++++++++++++++------------------- common/test_list | 8 ++--- 2 files changed, 53 insertions(+), 36 deletions(-)
check-parallel +49 -32
@@ -175,7 +175,6 @@ time_order_test_list() local rx=0 local ix local jx - #set -x for ((ix = 0; ix < ${#_list[*]}; ix++)); do echo $_tl_tests | grep -q ${_list[$ix]} if [ $? == 0 ]; then
@@ -198,6 +197,26 @@ if ! $_tl_randomise -a ! $_tl_exact_order; then fi fi +# Build the test list from the test sepcification and FSTYP settings. This can +# be called at any time to regenerate the list as the include/exclude lists +# generated from the command line are retained separately to the test list +# itself. +setup_test_list() +{ + _tl_prepare_test_list + _tl_strip_test_list + + if ! $_tl_randomise -a ! $_tl_exact_order; then + if [ -f $basedir/runner-0/$prev_results/check.time ]; then + time_order_test_list + fi + fi + + # reverse the order of tests so that the get_next_test() can pull from the file + # tail rather than the head. + echo $_tl_tests |sed -e 's/ /\n/g' | tac > $test_list +} + # Grab the next test to be run from the tail of the file. # Returns an empty string if there is no tests remaining to run. # File operations are run under flock so concurrent gets are serialised against
@@ -242,6 +261,7 @@ _destroy_loop_device() run_tests() { local section="$1" + local logfile="$2" exec 99<>$tmp.test_list_lock
@@ -257,7 +277,7 @@ run_tests() echo -n " $test_to_run " unset FSTESTS_ISOL if ! _tl_expunge_test $test_to_run; then - tools/run_privatens ./check -s $section $test_to_run >> $me/log 2>&1 + tools/run_privatens ./check -s $section $test_to_run >> $logfile 2>&1 fi test_to_run=$(get_next_test $section)
@@ -308,12 +328,12 @@ runner_go() rm -f $RESULT_BASE/check.* # Only supports default mkfs parameters right now - wipefs -a $TEST_DEV > $me/log 2>&1 - yes | mkfs -t $FSTYP $TEST_MKFS_OPTS $TEST_DEV >> $me/log 2>&1 + wipefs -a $TEST_DEV >> $_results/log 2>&1 + yes | mkfs -t $FSTYP $TEST_MKFS_OPTS $TEST_DEV >> $_results/log 2>&1 # export DUMP_CORRUPT_FS=1 - run_tests $section + run_tests $section $_results/log wait sleep 1
@@ -327,10 +347,10 @@ runner_go() _destroy_loop_device $SCRATCH_LOGDEV _destroy_loop_device $LOGWRITES_DEV - grep -q Failures: $me/log + grep -q Failures: $_results/log if [ $? -eq 0 ]; then echo -n "Runner $id Failures: " - grep Failures: $me/log | uniq | sed -e "s/^.*Failures://" + grep Failures: $_results/log | uniq | sed -e "s/^.*Failures://" fi }
@@ -339,19 +359,33 @@ run_section() { local section="$1" local now="$2" + local results="$basedir/*/results-$now" local i echo $run_section |grep -qw $section || return echo $exclude_section |grep -qw $section && return echo - echo Running section: $section + echo Running section: $section $now echo parse_config_section $section - # set up consumable test list first + # update the test list if necessary, then set up + # the consumable test list for this section to use. + if [ "$last_fstyp" != "$FSTYP" ]; then + setup_test_list + last_fstyp=$FSTYP + fi + echo FS Type: $FSTYP + + if [ -n "$show_test_list" ]; then + echo Test list to run: + cat $test_list + return + fi cp $test_list $test_list.$section + for ((i = 0; i < $runners; i++)); do runner_go $i $now $section & done
@@ -360,17 +394,18 @@ run_section() echo echo Section: $section echo -n "Tests run: " - grep Ran $basedir/*/log | sed -e 's,^.*:,,' -e 's, ,\n,g' | sort | uniq | wc -l + grep Ran $results/log | sed -e 's,^.*:,,' -e 's, ,\n,g' | sort | uniq | wc -l echo -n "Tests _notrun: " - grep "^Not run" $basedir/*/log | uniq | sed -e 's,^.*:,,' -e 's, ,\n,g' -e 's,^\n,,' | wc -l + grep "^Not run" $results/log | uniq | sed -e 's,^.*:,,' -e 's, ,\n,g' -e 's,^\n,,' | wc -l echo -n "Failure count: " - grep Failures: $basedir/*/log | uniq | sed -e "s/^.*Failures://" -e "s,\([0-9]\) \([gx]\),\1\n \2,g" |wc -l + grep Failures: $results/log | uniq | sed -e "s/^.*Failures://" -e "s,\([0-9]\) \([gx]\),\1\n \2,g" |wc -l echo echo Ten slowest tests - runtime in seconds: - cat $basedir/*/results-$now/check.time | sort -k 2 -nr | head -10 + cat $results/check.time | sort -k 2 -nr | head -10 + } cleanup()
@@ -386,27 +421,8 @@ cleanup() trap "cleanup; exit" HUP INT QUIT TERM _config_setup_parallel - run_section=${run_section:="$HOST_OPTIONS_SECTIONS"} - _tl_setup_exclude_group "unreliable_in_parallel" -_tl_prepare_test_list -_tl_strip_test_list - -if ! $_tl_randomise -a ! $_tl_exact_order; then - if [ -f $basedir/runner-0/$prev_results/check.time ]; then - time_order_test_list - fi -fi - -# reverse the order of tests so that the get_next_test() can pull from the file -# tail rather than the head. -echo $_tl_tests |sed -e 's/ /\n/g' | tac > $test_list -if [ -n "$show_test_list" ]; then - echo Time ordered test list: - cat $test_list - exit 0 -fi # Each parallel test runner needs to only see it's own mount points. If we # leave the basedir as shared, then all tests see all mounts and then we get
@@ -422,6 +438,7 @@ fi mount --make-private $basedir now=`date +%Y-%m-%d-%H:%M:%S` +last_fstyp= for section in $HOST_OPTIONS_SECTIONS; do run_section $section $now if [ "$sum_bad" != 0 ] && [ "$istop" = true ]; then
common/test_list +4 -4
@@ -20,7 +20,7 @@ _XGROUP_LIST= _tl_exact_order=false _tl_randomise=false _tl_have_test_args=false -_tl_file="$tmp.test_list" +_tl_cli_tests="$tmp._tl_cli_tests" _tl_exclude_tests=() _tl_tests=
@@ -122,8 +122,8 @@ _tl_prepare_test_list() { unset _tl_tests # Tests specified on the command line - if [ -s $_tl_file ]; then - cat $_tl_file > $tmp.list + if [ -s $_tl_cli_tests ]; then + cat $_tl_cli_tests > $tmp.list else touch $tmp.list fi
@@ -281,7 +281,7 @@ _tl_setup_cli() if grep -Eq "^$test_name" $group_file; then # in group file ... OK echo $_tl_src_dir/$test_dir/$test_name \ - >> $_tl_file + >> $_tl_cli_tests _tl_have_test_args=true else # oops
Signature · 1 line
2.45.2
Re: [PATCH 17/28] check-parallel: rebuild test list when FSTYP changes
Quoted 11 lines · Dave Chinner
From: Dave Chinner <[email protected]>
When a config section changes FSTYP, we do not rebuild the test list that is to be run. This means that if we go from xfs to ext4, the ext4 section will still try to run all the XFS tests, and won't try to run the ext4 tests.
To fix this for check-parallel, check if the FSTYP has changed and if it has strip the old tests from the test list and add the new tests back to it.
Quoted 7 lines
This is effectively a zero-day bug in the check config section support code.
While there, remain the internal
_tl_filevariable and the filename used to hold command line tests. The check-parallel change uncovered that it was using the same filename ($tmp.test_list) as the test list code was using and they were stepping on each other.
Right.
Quoted 20 lines
Signed-off-by Dave Chinner <[email protected]>--- check-parallel | 81 +++++++++++++++++++++++++++++------------------- common/test_list | 8 ++--- 2 files changed, 53 insertions(+), 36 deletions(-)check-parallel +0
-1@@ -175,7 +175,6 @@ time_order_test_list() local rx=0 local ix local jx - #set -x for ((ix = 0; ix < ${#_list[*]}; ix++)); do echo $_tl_tests | grep -q ${_list[$ix]} if [ $? == 0 ]; then
check-parallel if ! $_tl_randomise -a ! $_tl_exact_order; then · line 200
Quoted 4 lines
check-parallel +1
-0@@ -198,6 +197,26 @@ if ! $_tl_randomise -a ! $_tl_exact_order; then fi fi +# Build the test list from the test sepcification and FSTYP settings. This can
typo: specification
Quoted 69 lines
check-parallel +26
-6+# be called at any time to regenerate the list as the include/exclude lists +# generated from the command line are retained separately to the test list +# itself. +setup_test_list() +{ + _tl_prepare_test_list + _tl_strip_test_list + + if ! $_tl_randomise -a ! $_tl_exact_order; then + if [ -f $basedir/runner-0/$prev_results/check.time ]; then + time_order_test_list + fi + fi + + # reverse the order of tests so that the get_next_test() can pull from the file + # tail rather than the head. + echo $_tl_tests |sed -e 's/ /\n/g' | tac > $test_list +} + # Grab the next test to be run from the tail of the file. # Returns an empty string if there is no tests remaining to run. # File operations are run under flock so concurrent gets are serialised against@@ -242,6 +261,7 @@ _destroy_loop_device() run_tests() { local section="$1" + local logfile="$2" exec 99<>$tmp.test_list_lock@@ -257,7 +277,7 @@ run_tests() echo -n " $test_to_run " unset FSTESTS_ISOL if ! _tl_expunge_test $test_to_run; then - tools/run_privatens ./check -s $section $test_to_run >> $me/log 2>&1 + tools/run_privatens ./check -s $section $test_to_run >> $logfile 2>&1 fi test_to_run=$(get_next_test $section)@@ -308,12 +328,12 @@ runner_go() rm -f $RESULT_BASE/check.* # Only supports default mkfs parameters right now - wipefs -a $TEST_DEV > $me/log 2>&1 - yes | mkfs -t $FSTYP $TEST_MKFS_OPTS $TEST_DEV >> $me/log 2>&1 + wipefs -a $TEST_DEV >> $_results/log 2>&1 + yes | mkfs -t $FSTYP $TEST_MKFS_OPTS $TEST_DEV >> $_results/log 2>&1 # export DUMP_CORRUPT_FS=1 - run_tests $section + run_tests $section $_results/log wait sleep 1@@ -327,10 +347,10 @@ runner_go() _destroy_loop_device $SCRATCH_LOGDEV _destroy_loop_device $LOGWRITES_DEV - grep -q Failures: $me/log + grep -q Failures: $_results/log if [ $? -eq 0 ]; then echo -n "Runner $id Failures: " - grep Failures: $me/log | uniq | sed -e "s/^.*Failures://" + grep Failures: $_results/log | uniq | sed -e "s/^.*Failures://" fi }
check-parallel run_section() · line 362
Quoted 4 lines
check-parallel +1
-0@@ -339,19 +359,33 @@ run_section() { local section="$1" local now="$2" + local results="$basedir/*/results-$now"
This results points to all the individual _results for result-"$now" and _results is for each runner id and an associated timestamp.
So if we have 2 invocations at 2 different timestamps(t1 and t2) with 2 test runners, we will have $basedir/runner-0/result-t1
$basedir/runner-1/result-t1
$basedir/runner-0/result-t2
$basedir/runner-1/result-t2
I feel grouping the results for each invocation, i.e for each timestamp together looks more convenient, doesn't it. What I am saying is the following:
Instead of the above if we can have
$basedir/r-t1/runner-0/result
$basedir/r-t1/runner-1/result
$basedir/r-t2/runner-0/result
$basedir/r-t2/runner-1/result
So all the files/results under one invocation are under one direction i.e, $basedir/r-t<n>/* where t<n> is the timestamp we are looking for.
Please let me know what do you think of the above and if it has any pitfalls?
Quoted 83 lines
check-parallel +20
-25local i echo $run_section |grep -qw $section || return echo $exclude_section |grep -qw $section && return echo - echo Running section: $section + echo Running section: $section $now echo parse_config_section $section - # set up consumable test list first + # update the test list if necessary, then set up + # the consumable test list for this section to use. + if [ "$last_fstyp" != "$FSTYP" ]; then + setup_test_list + last_fstyp=$FSTYP + fi + echo FS Type: $FSTYP + + if [ -n "$show_test_list" ]; then + echo Test list to run: + cat $test_list + return + fi cp $test_list $test_list.$section + for ((i = 0; i < $runners; i++)); do runner_go $i $now $section & done@@ -360,17 +394,18 @@ run_section() echo echo Section: $section echo -n "Tests run: " - grep Ran $basedir/*/log | sed -e 's,^.*:,,' -e 's, ,\n,g' | sort | uniq | wc -l + grep Ran $results/log | sed -e 's,^.*:,,' -e 's, ,\n,g' | sort | uniq | wc -l echo -n "Tests _notrun: " - grep "^Not run" $basedir/*/log | uniq | sed -e 's,^.*:,,' -e 's, ,\n,g' -e 's,^\n,,' | wc -l + grep "^Not run" $results/log | uniq | sed -e 's,^.*:,,' -e 's, ,\n,g' -e 's,^\n,,' | wc -l echo -n "Failure count: " - grep Failures: $basedir/*/log | uniq | sed -e "s/^.*Failures://" -e "s,\([0-9]\) \([gx]\),\1\n \2,g" |wc -l + grep Failures: $results/log | uniq | sed -e "s/^.*Failures://" -e "s,\([0-9]\) \([gx]\),\1\n \2,g" |wc -l echo echo Ten slowest tests - runtime in seconds: - cat $basedir/*/results-$now/check.time | sort -k 2 -nr | head -10 + cat $results/check.time | sort -k 2 -nr | head -10 + } cleanup()@@ -386,27 +421,8 @@ cleanup() trap "cleanup; exit" HUP INT QUIT TERM _config_setup_parallel - run_section=${run_section:="$HOST_OPTIONS_SECTIONS"} - _tl_setup_exclude_group "unreliable_in_parallel" -_tl_prepare_test_list -_tl_strip_test_list - -if ! $_tl_randomise -a ! $_tl_exact_order; then - if [ -f $basedir/runner-0/$prev_results/check.time ]; then - time_order_test_list - fi -fi - -# reverse the order of tests so that the get_next_test() can pull from the file -# tail rather than the head. -echo $_tl_tests |sed -e 's/ /\n/g' | tac > $test_list -if [ -n "$show_test_list" ]; then - echo Time ordered test list: - cat $test_list - exit 0 -fi # Each parallel test runner needs to only see it's own mount points. If we # leave the basedir as shared, then all tests see all mounts and then we get
check-parallel fi · line 441
Quoted 4 lines
check-parallel +1
-0@@ -422,6 +438,7 @@ fi mount --make-private $basedir now=`date +%Y-%m-%d-%H:%M:%S` +last_fstyp=
Minor: Can we make last_fstyp somehow local to this file? The reason why I am asking is that, we
have some similar logic in run_section() of check ("$OLD_FSTYP" != "$FSTYP" -> recreate TEST_DEV)
and maybe in an unlikely event if somebody tends to re-use this variable name unknowingly - haven't thought about how would this script behave though, but just in case. --NR
Quoted 36 lines
check-parallel +0
-0for section in $HOST_OPTIONS_SECTIONS; do run_section $section $now if [ "$sum_bad" != 0 ] && [ "$istop" = true ]; thencommon/test_list +4
-4@@ -20,7 +20,7 @@ _XGROUP_LIST= _tl_exact_order=false _tl_randomise=false _tl_have_test_args=false -_tl_file="$tmp.test_list" +_tl_cli_tests="$tmp._tl_cli_tests" _tl_exclude_tests=() _tl_tests=@@ -122,8 +122,8 @@ _tl_prepare_test_list() { unset _tl_tests # Tests specified on the command line - if [ -s $_tl_file ]; then - cat $_tl_file > $tmp.list + if [ -s $_tl_cli_tests ]; then + cat $_tl_cli_tests > $tmp.list else touch $tmp.list fi@@ -281,7 +281,7 @@ _tl_setup_cli() if grep -Eq "^$test_name" $group_file; then # in group file ... OK echo $_tl_src_dir/$test_dir/$test_name \ - >> $_tl_file + >> $_tl_cli_tests _tl_have_test_args=true else # oops
PATCH 18/28 check-parallel: create a "results-latest" symlink
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
check-parallel ensures that it doesn't overwrite results by creating date-stamped results directories. This, however, makes it harder to easy find the results for the test that is currently running or has just completed.
To solve this problem, maintain a symlink to the latest results directory so that it can always be found by the same name.
Also, if this symlink exists and is valid, use it as the source for runtime ordering data for the next run instead of trying to find it via ls and sort based ordering.
--- check-parallel | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)
check-parallel +6 -1
@@ -149,7 +149,9 @@ if [[ $runners -le 0 || $runners -gt 1024 ]]; then usage fi -if [ -d "$basedir/runner-0/" ]; then +if [ -L $basedir/runner-0/latest-result ]; then + prev_results="latest-result" +elif [ -d "$basedir/runner-0/" ]; then prev_results=`ls -tr $basedir/runner-0/ | grep results | tail -1` fi
@@ -325,6 +327,9 @@ runner_go() mkdir -p $TEST_DIR mkdir -p $SCRATCH_MNT mkdir -p $RESULT_BASE + rm -f $me/latest-result + ln -s $_results $me/latest-result + rm -f $RESULT_BASE/check.* # Only supports default mkfs parameters right now
Signature · 1 line
2.45.2
Re: [PATCH 18/28] check-parallel: create a "results-latest" symlink
Quoted 7 lines · Dave Chinner
From: Dave Chinner <[email protected]>
check-parallel ensures that it doesn't overwrite results by creating date-stamped results directories. This, however, makes it harder to easy find the results for the test that is currently running or has just completed.
Quoted 6 lines
To solve this problem, maintain a symlink to the latest results directory so that it can always be found by the same name.
Also, if this symlink exists and is valid, use it as the source for runtime ordering data for the next run instead of trying to find it via ls and sort based ordering.
This looks good and useful. However, if we can have the above information in the README and in the comment section of the script it will be really helpful for someone who is new.
--NR
Quoted 30 lines
Signed-off-by Dave Chinner <[email protected]>--- check-parallel | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)check-parallel +6
-1@@ -149,7 +149,9 @@ if [[ $runners -le 0 || $runners -gt 1024 ]]; then usage fi -if [ -d "$basedir/runner-0/" ]; then +if [ -L $basedir/runner-0/latest-result ]; then + prev_results="latest-result" +elif [ -d "$basedir/runner-0/" ]; then prev_results=`ls -tr $basedir/runner-0/ | grep results | tail -1` fi@@ -325,6 +327,9 @@ runner_go() mkdir -p $TEST_DIR mkdir -p $SCRATCH_MNT mkdir -p $RESULT_BASE + rm -f $me/latest-result + ln -s $_results $me/latest-result + rm -f $RESULT_BASE/check.* # Only supports default mkfs parameters right now
PATCH 19/28 check: factor test running
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
Rework the code that check uses to run an individual test, separating the executing of the test from the various pre- and post- test processing operations that are specific to check results processing.
This essentially encapsulates the test running and result tracking in it's own file, leaving just the section iteration and reporting to the caller that is running the tests. The caller needs to define the runseq() function that actually executes the test and some environment variables (e.g. REPORT_DIR) so that the test execution code will run correctly and stash results and reports in the correct location.
This greatly simplifies the check script as a big chunk of code dedicated to simply running a test and gathering the results is completely abstracted away. This makes it clearer what check is now doing at a high level in terms of iterating sections and generating reports.
--- check | 396 ++++++----------------------------------------- common/test_exec | 352 +++++++++++++++++++++++++++++++++++++++++ 2 files changed, 398 insertions(+), 350 deletions(-) create mode 100644 common/test_exec
check +46 -350
@@ -8,13 +8,9 @@ tmp=/tmp/$$ status=0 needwrap=true needsum=true -try=() sum_bad=0 -bad=() -notrun=() interrupt=true diff="diff -u" -showme=false export here=`pwd` brief_test_summary=false do_report=false
@@ -38,14 +34,12 @@ export QA_CHECK_FS=${QA_CHECK_FS:=true} # number of diff lines from a failed test, 0 for whole output export DIFF_LENGTH=${DIFF_LENGTH:=10} -# by default don't output timestamps -timestamp=${TIMESTAMP:=false} - rm -f $tmp.list $tmp.tmp $tmp.grep $here/$iam.out $tmp.report.* $tmp.arglist . ./common/exit . ./common/test_names . ./common/test_list +. ./common/test_exec . ./common/config . ./common/config-sections . ./common/rc
@@ -122,12 +116,6 @@ examples: exit 1 } -_timestamp() -{ - local now=`date "+%T"` - echo -n " [$now]" -} - # Process command arguments first. while [ $# -gt 0 ]; do case "$1" in
@@ -155,7 +143,7 @@ while [ $# -gt 0 ]; do -l) diff="diff" ;; -udiff) diff="$diff -u" ;; - -n) showme=true ;; + -n) _te_dry_run="true" ;; -i) iterations=$2; shift ;; -I) iterations=$2; istop=true; shift ;; -T) timestamp=true ;;
@@ -197,6 +185,9 @@ if [ ! -z "$REPORT_LIST" ]; then _assert_report_list fi +# by default don't output timestamps +_te_emit_timestamps=${TIMESTAMP:=} + # If the test config specified a soak test duration, see if there are any # unit suffixes that need converting to an integer seconds count. if [ -n "$SOAK_DURATION" ]; then
@@ -227,13 +218,6 @@ then exit 1 fi -_wipe_counters() -{ - try=() - notrun=() - bad=() -} - _global_log() { echo "$1" >> $check.log if $OPTIONS_HAVE_SECTIONS; then
@@ -252,11 +236,11 @@ _wrapup() check="$RESULT_BASE/check" $interrupt && sect_stop=`_wallclock` - if $showme && $needwrap; then + if [ "$_te_dry_run" == "true" ] && $needwrap; then if $do_report; then - # $showme = all selected tests are notrun (no tries) - _make_section_report "$section" "${#notrun[*]}" "0" \ - "${#notrun[*]}" \ + # $_te_dry_run = all selected tests are notrun (no tries) + _make_section_report "$section" "${#_te_notrun[*]}" "0" \ + "${#_te_notrun[*]}" \ "$((sect_stop - sect_start))" fi needwrap=false
@@ -276,6 +260,7 @@ _wrapup() cp $check.time ${REPORT_DIR}/check.time fi fi + set +x _global_log "" _global_log "Kernel version: $(uname -r)"
@@ -283,12 +268,12 @@ _wrapup() echo "SECTION -- $section" >>$tmp.summary echo "=========================" >>$tmp.summary - if ((${#try[*]} > 0)); then + if ((${#_te_try[*]} > 0)); then if [ $brief_test_summary == "false" ]; then - echo "Ran: ${try[*]}" - echo "Ran: ${try[*]}" >>$tmp.summary + echo "Ran: ${_te_try[*]}" + echo "Ran: ${_te_try[*]}" >>$tmp.summary fi - _global_log "Ran: ${try[*]}" + _global_log "Ran: ${_te_try[*]}" fi $interrupt && echo "Interrupted!" | tee -a $check.log
@@ -297,30 +282,30 @@ _wrapup() ${REPORT_DIR}/check.log fi - if ((${#notrun[*]} > 0)); then + if ((${#_te_notrun[*]} > 0)); then if [ $brief_test_summary == "false" ]; then - echo "Not run: ${notrun[*]}" - echo "Not run: ${notrun[*]}" >>$tmp.summary + echo "Not run: ${_te_notrun[*]}" + echo "Not run: ${_te_notrun[*]}" >>$tmp.summary fi - _global_log "Not run: ${notrun[*]}" + _global_log "Not run: ${_te_notrun[*]}" fi - if ((${#bad[*]} > 0)); then - echo "Failures: ${bad[*]}" - echo "Failed ${#bad[*]} of ${#try[*]} tests" - _global_log "Failures: ${bad[*]}" - _global_log "Failed ${#bad[*]} of ${#try[*]} tests" - echo "Failures: ${bad[*]}" >>$tmp.summary - echo "Failed ${#bad[*]} of ${#try[*]} tests" >>$tmp.summary + if ((${#_te_bad[*]} > 0)); then + echo "Failures: ${_te_bad[*]}" + echo "Failed ${#_te_bad[*]} of ${#_te_try[*]} tests" + _global_log "Failures: ${_te_bad[*]}" + _global_log "Failed ${#_te_bad[*]} of ${#_te_try[*]} tests" + echo "Failures: ${_te_bad[*]}" >>$tmp.summary + echo "Failed ${#_te_bad[*]} of ${#_te_try[*]} tests" >>$tmp.summary else - echo "Passed all ${#try[*]} tests" - _global_log "Passed all ${#try[*]} tests" - echo "Passed all ${#try[*]} tests" >>$tmp.summary + echo "Passed all ${#_te_try[*]} tests" + _global_log "Passed all ${#_te_try[*]} tests" + echo "Passed all ${#_te_try[*]} tests" >>$tmp.summary fi echo "" >>$tmp.summary if $do_report; then - _make_section_report "$section" "${#try[*]}" \ - "${#bad[*]}" "${#notrun[*]}" \ + _make_section_report "$section" "${#_te_try[*]}" \ + "${#_te_bad[*]}" "${#_te_notrun[*]}" \ "$((sect_stop - sect_start))" fi
@@ -338,8 +323,8 @@ _wrapup() needwrap=false fi - sum_bad=`expr $sum_bad + ${#bad[*]}` - _wipe_counters + sum_bad=`expr $sum_bad + ${#_te_bad[*]}` + _te_wipe_counters if ! $OPTIONS_HAVE_SECTIONS; then rm -f $tmp.* fi
@@ -348,7 +333,7 @@ _wrapup() _summary() { _wrapup - if $showme; then + if [ "$_te_dry_run" == "true" ]; then : elif $needsum; then count=`wc -L $tmp.summary | cut -f1 -d" "`
@@ -358,95 +343,6 @@ _summary() rm -f $tmp.* } -_check_filesystems() -{ - local ret=0 - - if [ -f ${RESULT_DIR}/require_test ]; then - if ! _check_test_fs ; then - ret=1 - echo "Trying to repair broken TEST_DEV file system" - _repair_test_fs - _test_mount - fi - rm -f ${RESULT_DIR}/require_test* - else - _test_unmount 2> /dev/null - fi - if [ -f ${RESULT_DIR}/require_scratch ]; then - _check_scratch_fs || ret=1 - rm -f ${RESULT_DIR}/require_scratch* - fi - _scratch_unmount 2> /dev/null - return $ret -} - -# retain files which would be overwritten in subsequent reruns of the same test -_stash_fail_loop_files() { - local seq_prefix="${REPORT_DIR}/${1}" - local cp_suffix="$2" - - for i in ".full" ".dmesg" ".out.bad" ".notrun" ".core" ".hints"; do - rm -f "${seq_prefix}${i}${cp_suffix}" - if [ -f "${seq_prefix}${i}" ]; then - cp "${seq_prefix}${i}" "${seq_prefix}${i}${cp_suffix}" - fi - done -} - -# Retain in @bad / @notrun the result of the just-run @test_seq. @try array -# entries are added prior to execution. -_stash_test_status() { - local test_seq="$1" - local test_status="$2" - - if $do_report && [[ $test_status != "expunge" ]]; then - _make_testcase_report "$section" "$test_seq" \ - "$test_status" "$((stop - start))" - fi - - if ((${#loop_status[*]} > 0)); then - # continuing or completing rerun-on-failure loop - _stash_fail_loop_files "$test_seq" ".rerun${#loop_status[*]}" - loop_status+=("$test_status") - if ((${#loop_status[*]} > loop_on_fail)); then - printf "%s aggregate results across %d runs: " \ - "$test_seq" "${#loop_status[*]}" - awk "BEGIN { - n=split(\"${loop_status[*]}\", arr);"' - for (i = 1; i <= n; i++) - stats[arr[i]]++; - for (x in stats) - printf("%s=%d (%.1f%%)", - (i-- > n ? x : ", " x), - stats[x], 100 * stats[x] / n); - }' - echo - loop_status=() - fi - return # only stash @bad result for initial failure in loop - fi - - case "$test_status" in - fail) - if ((loop_on_fail > 0)); then - # initial failure, start rerun-on-failure loop - _stash_fail_loop_files "$test_seq" ".rerun0" - loop_status+=("$test_status") - fi - bad+=("$test_seq") - ;; - list|notrun) - notrun+=("$test_seq") - ;; - pass|expunge) - ;; - *) - echo "Unexpected test $test_seq status: $test_status" - ;; - esac -} - # Can we run systemd scopes? HAVE_SYSTEMD_SCOPES= systemctl reset-failed "fstests-check" &>/dev/null
@@ -454,11 +350,7 @@ systemd-run --quiet --unit "fstests-check" --scope bash -c "exit 77" &> /dev/nul test $? -eq 77 && HAVE_SYSTEMD_SCOPES=yes # Make the check script unattractive to the OOM killer... -OOM_SCORE_ADJ="/proc/self/oom_score_adj" -function _adjust_oom_score() { - test -w "${OOM_SCORE_ADJ}" && echo "$1" > "${OOM_SCORE_ADJ}" -} -_adjust_oom_score -500 +_te_adjust_oom_score -500 # ...and make the tests themselves somewhat more attractive to it, so that if # the system runs out of memory it'll be the test that gets killed and not the
@@ -472,7 +364,7 @@ _adjust_oom_score -500 # when systemd tells them to terminate (e.g. programs stuck in D state when # systemd sends SIGKILL), so we use reset-failed to tear down the scope. _run_seq() { - local cmd=(bash -c "test -w ${OOM_SCORE_ADJ} && echo 250 > ${OOM_SCORE_ADJ}; exec ./$seq") + local cmd=(bash -c "test -w ${_te_oom_score_adj} && echo 250 > ${_te_oom_score_adj}; exec ./$seq") local res if [ -n "${HAVE_SYSTEMD_SCOPES}" ]; then
@@ -617,222 +509,26 @@ function run_section() seqres="$check" _check_test_fs - loop_status=() # track rerun-on-failure state - local tc_status ix + if $OPTIONS_HAVE_SECTIONS; then + REPORT_DIR="$RESULT_BASE/$section" + else + REPORT_DIR="$RESULT_BASE" + fi + + local ix local -a _list=( $_tl_tests ) - for ((ix = 0; ix < ${#_list[*]}; !${#loop_status[*]} && ix++)); do + for ((ix = 0; ix < ${#_list[*]}; !${#_te_loop_status[*]} && ix++)); do seq="${_list[$ix]}" - # the filename for the test and the name output are different. - # we don't include the tests/ directory in the name output. - export seqnum=$(_tl_strip_src_dir $seq) - group=${seqnum%%/*} - if $OPTIONS_HAVE_SECTIONS; then - REPORT_DIR="$RESULT_BASE/$section" - else - REPORT_DIR="$RESULT_BASE" - fi - export RESULT_DIR="$REPORT_DIR/$group" - seqres="$REPORT_DIR/$seqnum" - # Generate the entire section report with whatever test results # we have so far. Leave the $sect_time parameter empty so that # it's a little more obvious that this test run is incomplete. if $do_report; then - _make_section_report "$section" "${#try[*]}" \ - "${#bad[*]}" "${#notrun[*]}" \ + _make_section_report "$section" "${#_te_try[*]}" \ + "${#_te_bad[*]}" "${#_te_notrun[*]}" \ "" &> /dev/null fi - - echo -n "$seqnum" - - if $showme; then - if _tl_expunge_test $seqnum; then - tc_status="expunge" - else - echo - start=0 - stop=0 - tc_status="list" - fi - _stash_test_status "$seqnum" "$tc_status" - continue - fi - - tc_status="pass" - if [ ! -f $seq ]; then - echo " - no such test?" - _stash_test_status "$seqnum" "$tc_status" - continue - fi - - # really going to try and run this one - mkdir -p $RESULT_DIR - rm -f ${RESULT_DIR}/require_scratch* - rm -f ${RESULT_DIR}/require_test* - rm -f $seqres.out.bad $seqres.hints - - # check if we really should run it - if _tl_expunge_test $seqnum; then - tc_status="expunge" - _stash_test_status "$seqnum" "$tc_status" - continue - fi - - # record that we really tried to run this test. - if ((!${#loop_status[*]})); then - try+=("$seqnum") - fi - - awk 'BEGIN {lasttime=" "} \ - $1 == "'$seqnum'" {lasttime=" " $2 "s ... "; exit} \ - END {printf "%s", lasttime}' "$check.time" - rm -f core $seqres.notrun - - start=`_wallclock` - $timestamp && _timestamp - [ ! -x $seq ] && chmod u+x $seq # ensure we can run it - $LOGGER_PROG "run xfstest $seqnum" - if [ -w /dev/kmsg ]; then - export date_time=`date +"%F %T"` - echo "run fstests $seqnum at $date_time" > /dev/kmsg - # _check_dmesg depends on this log in dmesg - touch ${RESULT_DIR}/check_dmesg - rm -f ${RESULT_DIR}/dmesg_filter - fi - _try_wipe_scratch_devs > /dev/null 2>&1 - - # clear the WARN_ONCE state to allow a potential problem - # to be reported for each test - (echo 1 > $DEBUGFS_MNT/clear_warn_once) > /dev/null 2>&1 - - test_start_time="$(date +"%F %T")" - if [ "$DUMP_OUTPUT" = true ]; then - _run_seq 2>&1 | tee $tmp.out - # Because $? would get tee's return code - sts=${PIPESTATUS[0]} - else - _run_seq >$tmp.out 2>&1 - sts=$? - fi - - # If someone sets kernel.core_pattern or kernel.core_uses_pid, - # coredumps generated by fstests might have a longer name than - # just "core". Use globbing to find the most common patterns, - # assuming there are no other coredump capture packages set up. - local cores=0 - for i in core core.*; do - test -f "$i" || continue - if ((cores++ == 0)); then - _dump_err_cont "[dumped core]" - fi - (_adjust_oom_score 250; _save_coredump "$i") - tc_status="fail" - done - - if [ -f $seqres.notrun ]; then - $timestamp && _timestamp - stop=`_wallclock` - $timestamp || echo -n "[not run] " - $timestamp && echo " [not run]" && \ - echo -n " $seqnum -- " - cat $seqres.notrun - tc_status="notrun" - _stash_test_status "$seqnum" "$tc_status" - - # Unmount the scratch fs so that we can wipe the scratch - # dev state prior to the next test run. - _scratch_unmount 2> /dev/null - continue; - fi - - if [ $sts -ne 0 ]; then - _dump_err_cont "[failed, exit status $sts]" - _test_unmount 2> /dev/null - _scratch_unmount 2> /dev/null - rm -f ${RESULT_DIR}/require_test* - rm -f ${RESULT_DIR}/require_scratch* - # Even though we failed, there may be something interesting in - # dmesg which can help debugging. - _check_dmesg - tc_status="fail" - else - # The test apparently passed, so check for corruption - # and log messages that shouldn't be there. Run the - # checking tools from a subshell with adjusted OOM - # score so that the OOM killer will target them instead - # of the check script itself. - (_adjust_oom_score 250; _check_filesystems) || tc_status="fail" - _check_dmesg || tc_status="fail" - - # Save any coredumps from the post-test fs checks - for i in core core.*; do - test -f "$i" || continue - if ((cores++ == 0)); then - _dump_err_cont "[dumped core]" - fi - (_adjust_oom_score 250; _save_coredump "$i") - tc_status="fail" - done - fi - - # Reload the module after each test to check for leaks or - # other problems. - if [ -n "${TEST_FS_MODULE_RELOAD}" ]; then - _test_unmount 2> /dev/null - _scratch_unmount 2> /dev/null - modprobe -r fs-$FSTYP - modprobe fs-$FSTYP - fi - - # Scan for memory leaks after every test so that associating - # a leak to a particular test will be as accurate as possible. - _check_kmemleak || tc_status="fail" - - # test ends after all checks are done. - $timestamp && _timestamp - stop=`_wallclock` - - if [ ! -f $seq.out ]; then - _dump_err "no qualified output" - tc_status="fail" - _stash_test_status "$seqnum" "$tc_status" - continue; - fi - - # coreutils 8.16+ changed quote formats in error messages - # from `foo' to 'foo'. Filter old versions to match the new - # version. - sed -i "s/\`/\'/g" $tmp.out - if diff $seq.out $tmp.out >/dev/null 2>&1 ; then - if [ "$tc_status" != "fail" ]; then - echo "$seqnum `expr $stop - $start`" >>$tmp.time - echo -n " `expr $stop - $start`s" - fi - echo "" - else - _dump_err "- output mismatch (see $seqres.out.bad)" - mv $tmp.out $seqres.out.bad - $diff $seq.out $seqres.out.bad | { - if test "$DIFF_LENGTH" -le 0; then - cat - else - head -n "$DIFF_LENGTH" - echo "..." - echo "(Run '$diff $here/$seq.out $seqres.out.bad'" \ - " to see the entire diff)" - fi; } | sed -e 's/^\(.\)/ \1/' - tc_status="fail" - fi - if [ -f $seqres.hints ]; then - if [ "$tc_status" == "fail" ]; then - echo - cat $seqres.hints - else - rm -f $seqres.hints - fi - fi - _stash_test_status "$seqnum" "$tc_status" + _te_run_test $seq done # Reset these three variables so that unmount output doesn't get
common/test_exec new file +352 -0
352 lines not on this page · raw message
Signature · 1 line
2.45.2
Re: [PATCH 19/28] check: factor test running
Quoted 11 lines · Dave Chinner
From: Dave Chinner <[email protected]>
Rework the code that check uses to run an individual test, separating the executing of the test from the various pre- and post- test processing operations that are specific to check results processing.
This essentially encapsulates the test running and result tracking in it's own file, leaving just the section iteration and reporting to the caller that is running the tests. The caller needs to define the
runseq()function that actually executes the test and some
I couldn't understand the usage of runseq() function? Can you please elaborate this a little bit more?
Quoted 315 lines
environment variables (e.g.
REPORT_DIR) so that the test execution code will run correctly and stash results and reports in the correct location.This greatly simplifies the check script as a big chunk of code dedicated to simply running a test and gathering the results is completely abstracted away. This makes it clearer what check is now doing at a high level in terms of iterating sections and generating reports.
Signed-off-by Dave Chinner <[email protected]>--- check | 396 ++++++----------------------------------------- common/test_exec | 352 +++++++++++++++++++++++++++++++++++++++++ 2 files changed, 398 insertions(+), 350 deletions(-) create mode 100644 common/test_execcheck +34
-142@@ -8,13 +8,9 @@ tmp=/tmp/$$ status=0 needwrap=true needsum=true -try=() sum_bad=0 -bad=() -notrun=() interrupt=true diff="diff -u" -showme=false export here=`pwd` brief_test_summary=false do_report=false@@ -38,14 +34,12 @@ export QA_CHECK_FS=${QA_CHECK_FS:=true} # number of diff lines from a failed test, 0 for whole output export DIFF_LENGTH=${DIFF_LENGTH:=10} -# by default don't output timestamps -timestamp=${TIMESTAMP:=false} - rm -f $tmp.list $tmp.tmp $tmp.grep $here/$iam.out $tmp.report.* $tmp.arglist . ./common/exit . ./common/test_names . ./common/test_list +. ./common/test_exec . ./common/config . ./common/config-sections . ./common/rc@@ -122,12 +116,6 @@ examples: exit 1 } -_timestamp() -{ - local now=`date "+%T"` - echo -n " [$now]" -} - # Process command arguments first. while [ $# -gt 0 ]; do case "$1" in@@ -155,7 +143,7 @@ while [ $# -gt 0 ]; do -l) diff="diff" ;; -udiff) diff="$diff -u" ;; - -n) showme=true ;; + -n) _te_dry_run="true" ;; -i) iterations=$2; shift ;; -I) iterations=$2; istop=true; shift ;; -T) timestamp=true ;;@@ -197,6 +185,9 @@ if [ ! -z "$REPORT_LIST" ]; then _assert_report_list fi +# by default don't output timestamps +_te_emit_timestamps=${TIMESTAMP:=} + # If the test config specified a soak test duration, see if there are any # unit suffixes that need converting to an integer seconds count. if [ -n "$SOAK_DURATION" ]; then@@ -227,13 +218,6 @@ then exit 1 fi -_wipe_counters() -{ - try=() - notrun=() - bad=() -} - _global_log() { echo "$1" >> $check.log if $OPTIONS_HAVE_SECTIONS; then@@ -252,11 +236,11 @@ _wrapup() check="$RESULT_BASE/check" $interrupt && sect_stop=`_wallclock` - if $showme && $needwrap; then + if [ "$_te_dry_run" == "true" ] && $needwrap; then if $do_report; then - # $showme = all selected tests are notrun (no tries) - _make_section_report "$section" "${#notrun[*]}" "0" \ - "${#notrun[*]}" \ + # $_te_dry_run = all selected tests are notrun (no tries) + _make_section_report "$section" "${#_te_notrun[*]}" "0" \ + "${#_te_notrun[*]}" \ "$((sect_stop - sect_start))" fi needwrap=false@@ -276,6 +260,7 @@ _wrapup() cp $check.time ${REPORT_DIR}/check.time fi fi + set +x _global_log "" _global_log "Kernel version: $(uname -r)"@@ -283,12 +268,12 @@ _wrapup() echo "SECTION -- $section" >>$tmp.summary echo "=========================" >>$tmp.summary - if ((${#try[*]} > 0)); then + if ((${#_te_try[*]} > 0)); then if [ $brief_test_summary == "false" ]; then - echo "Ran: ${try[*]}" - echo "Ran: ${try[*]}" >>$tmp.summary + echo "Ran: ${_te_try[*]}" + echo "Ran: ${_te_try[*]}" >>$tmp.summary fi - _global_log "Ran: ${try[*]}" + _global_log "Ran: ${_te_try[*]}" fi $interrupt && echo "Interrupted!" | tee -a $check.log@@ -297,30 +282,30 @@ _wrapup() ${REPORT_DIR}/check.log fi - if ((${#notrun[*]} > 0)); then + if ((${#_te_notrun[*]} > 0)); then if [ $brief_test_summary == "false" ]; then - echo "Not run: ${notrun[*]}" - echo "Not run: ${notrun[*]}" >>$tmp.summary + echo "Not run: ${_te_notrun[*]}" + echo "Not run: ${_te_notrun[*]}" >>$tmp.summary fi - _global_log "Not run: ${notrun[*]}" + _global_log "Not run: ${_te_notrun[*]}" fi - if ((${#bad[*]} > 0)); then - echo "Failures: ${bad[*]}" - echo "Failed ${#bad[*]} of ${#try[*]} tests" - _global_log "Failures: ${bad[*]}" - _global_log "Failed ${#bad[*]} of ${#try[*]} tests" - echo "Failures: ${bad[*]}" >>$tmp.summary - echo "Failed ${#bad[*]} of ${#try[*]} tests" >>$tmp.summary + if ((${#_te_bad[*]} > 0)); then + echo "Failures: ${_te_bad[*]}" + echo "Failed ${#_te_bad[*]} of ${#_te_try[*]} tests" + _global_log "Failures: ${_te_bad[*]}" + _global_log "Failed ${#_te_bad[*]} of ${#_te_try[*]} tests" + echo "Failures: ${_te_bad[*]}" >>$tmp.summary + echo "Failed ${#_te_bad[*]} of ${#_te_try[*]} tests" >>$tmp.summary else - echo "Passed all ${#try[*]} tests" - _global_log "Passed all ${#try[*]} tests" - echo "Passed all ${#try[*]} tests" >>$tmp.summary + echo "Passed all ${#_te_try[*]} tests" + _global_log "Passed all ${#_te_try[*]} tests" + echo "Passed all ${#_te_try[*]} tests" >>$tmp.summary fi echo "" >>$tmp.summary if $do_report; then - _make_section_report "$section" "${#try[*]}" \ - "${#bad[*]}" "${#notrun[*]}" \ + _make_section_report "$section" "${#_te_try[*]}" \ + "${#_te_bad[*]}" "${#_te_notrun[*]}" \ "$((sect_stop - sect_start))" fi@@ -338,8 +323,8 @@ _wrapup() needwrap=false fi - sum_bad=`expr $sum_bad + ${#bad[*]}` - _wipe_counters + sum_bad=`expr $sum_bad + ${#_te_bad[*]}` + _te_wipe_counters if ! $OPTIONS_HAVE_SECTIONS; then rm -f $tmp.* fi@@ -348,7 +333,7 @@ _wrapup() _summary() { _wrapup - if $showme; then + if [ "$_te_dry_run" == "true" ]; then : elif $needsum; then count=`wc -L $tmp.summary | cut -f1 -d" "`@@ -358,95 +343,6 @@ _summary() rm -f $tmp.* } -_check_filesystems() -{ - local ret=0 - - if [ -f ${RESULT_DIR}/require_test ]; then - if ! _check_test_fs ; then - ret=1 - echo "Trying to repair broken TEST_DEV file system" - _repair_test_fs - _test_mount - fi - rm -f ${RESULT_DIR}/require_test* - else - _test_unmount 2> /dev/null - fi - if [ -f ${RESULT_DIR}/require_scratch ]; then - _check_scratch_fs || ret=1 - rm -f ${RESULT_DIR}/require_scratch* - fi - _scratch_unmount 2> /dev/null - return $ret -} - -# retain files which would be overwritten in subsequent reruns of the same test -_stash_fail_loop_files() { - local seq_prefix="${REPORT_DIR}/${1}" - local cp_suffix="$2" - - for i in ".full" ".dmesg" ".out.bad" ".notrun" ".core" ".hints"; do - rm -f "${seq_prefix}${i}${cp_suffix}" - if [ -f "${seq_prefix}${i}" ]; then - cp "${seq_prefix}${i}" "${seq_prefix}${i}${cp_suffix}" - fi - done -} - -# Retain in @bad / @notrun the result of the just-run @test_seq. @try array -# entries are added prior to execution. -_stash_test_status() { - local test_seq="$1" - local test_status="$2" - - if $do_report && [[ $test_status != "expunge" ]]; then - _make_testcase_report "$section" "$test_seq" \ - "$test_status" "$((stop - start))" - fi - - if ((${#loop_status[*]} > 0)); then - # continuing or completing rerun-on-failure loop - _stash_fail_loop_files "$test_seq" ".rerun${#loop_status[*]}" - loop_status+=("$test_status") - if ((${#loop_status[*]} > loop_on_fail)); then - printf "%s aggregate results across %d runs: " \ - "$test_seq" "${#loop_status[*]}" - awk "BEGIN { - n=split(\"${loop_status[*]}\", arr);"' - for (i = 1; i <= n; i++) - stats[arr[i]]++; - for (x in stats) - printf("%s=%d (%.1f%%)", - (i-- > n ? x : ", " x), - stats[x], 100 * stats[x] / n); - }' - echo - loop_status=() - fi - return # only stash @bad result for initial failure in loop - fi - - case "$test_status" in - fail) - if ((loop_on_fail > 0)); then - # initial failure, start rerun-on-failure loop - _stash_fail_loop_files "$test_seq" ".rerun0" - loop_status+=("$test_status") - fi - bad+=("$test_seq") - ;; - list|notrun) - notrun+=("$test_seq") - ;; - pass|expunge) - ;; - *) - echo "Unexpected test $test_seq status: $test_status" - ;; - esac -} - # Can we run systemd scopes? HAVE_SYSTEMD_SCOPES= systemctl reset-failed "fstests-check" &>/dev/null@@ -454,11 +350,7 @@ systemd-run --quiet --unit "fstests-check" --scope bash -c "exit 77" &> /dev/nul test $? -eq 77 && HAVE_SYSTEMD_SCOPES=yes # Make the check script unattractive to the OOM killer... -OOM_SCORE_ADJ="/proc/self/oom_score_adj" -function _adjust_oom_score() { - test -w "${OOM_SCORE_ADJ}" && echo "$1" > "${OOM_SCORE_ADJ}" -} -_adjust_oom_score -500 +_te_adjust_oom_score -500 # ...and make the tests themselves somewhat more attractive to it, so that if # the system runs out of memory it'll be the test that gets killed and not the
check _adjust_oom_score -500 · line 367
Quoted 5 lines
check +1
-1@@ -472,7 +364,7 @@ _adjust_oom_score -500 # when systemd tells them to terminate (e.g. programs stuck in D state when # systemd sends SIGKILL), so we use reset-failed to tear down the scope. _run_seq() { - local cmd=(bash -c "test -w ${OOM_SCORE_ADJ} && echo 250 > ${OOM_SCORE_ADJ}; exec ./$seq") + local cmd=(bash -c "test -w ${_te_oom_score_adj} && echo 250 > ${_te_oom_score_adj}; exec ./$seq")
nit: 80 chars exceeded
Quoted 326 lines
check +11
-207local res if [ -n "${HAVE_SYSTEMD_SCOPES}" ]; then@@ -617,222 +509,26 @@ function run_section() seqres="$check" _check_test_fs - loop_status=() # track rerun-on-failure state - local tc_status ix + if $OPTIONS_HAVE_SECTIONS; then + REPORT_DIR="$RESULT_BASE/$section" + else + REPORT_DIR="$RESULT_BASE" + fi + + local ix local -a _list=( $_tl_tests ) - for ((ix = 0; ix < ${#_list[*]}; !${#loop_status[*]} && ix++)); do + for ((ix = 0; ix < ${#_list[*]}; !${#_te_loop_status[*]} && ix++)); do seq="${_list[$ix]}" - # the filename for the test and the name output are different. - # we don't include the tests/ directory in the name output. - export seqnum=$(_tl_strip_src_dir $seq) - group=${seqnum%%/*} - if $OPTIONS_HAVE_SECTIONS; then - REPORT_DIR="$RESULT_BASE/$section" - else - REPORT_DIR="$RESULT_BASE" - fi - export RESULT_DIR="$REPORT_DIR/$group" - seqres="$REPORT_DIR/$seqnum" - # Generate the entire section report with whatever test results # we have so far. Leave the $sect_time parameter empty so that # it's a little more obvious that this test run is incomplete. if $do_report; then - _make_section_report "$section" "${#try[*]}" \ - "${#bad[*]}" "${#notrun[*]}" \ + _make_section_report "$section" "${#_te_try[*]}" \ + "${#_te_bad[*]}" "${#_te_notrun[*]}" \ "" &> /dev/null fi - - echo -n "$seqnum" - - if $showme; then - if _tl_expunge_test $seqnum; then - tc_status="expunge" - else - echo - start=0 - stop=0 - tc_status="list" - fi - _stash_test_status "$seqnum" "$tc_status" - continue - fi - - tc_status="pass" - if [ ! -f $seq ]; then - echo " - no such test?" - _stash_test_status "$seqnum" "$tc_status" - continue - fi - - # really going to try and run this one - mkdir -p $RESULT_DIR - rm -f ${RESULT_DIR}/require_scratch* - rm -f ${RESULT_DIR}/require_test* - rm -f $seqres.out.bad $seqres.hints - - # check if we really should run it - if _tl_expunge_test $seqnum; then - tc_status="expunge" - _stash_test_status "$seqnum" "$tc_status" - continue - fi - - # record that we really tried to run this test. - if ((!${#loop_status[*]})); then - try+=("$seqnum") - fi - - awk 'BEGIN {lasttime=" "} \ - $1 == "'$seqnum'" {lasttime=" " $2 "s ... "; exit} \ - END {printf "%s", lasttime}' "$check.time" - rm -f core $seqres.notrun - - start=`_wallclock` - $timestamp && _timestamp - [ ! -x $seq ] && chmod u+x $seq # ensure we can run it - $LOGGER_PROG "run xfstest $seqnum" - if [ -w /dev/kmsg ]; then - export date_time=`date +"%F %T"` - echo "run fstests $seqnum at $date_time" > /dev/kmsg - # _check_dmesg depends on this log in dmesg - touch ${RESULT_DIR}/check_dmesg - rm -f ${RESULT_DIR}/dmesg_filter - fi - _try_wipe_scratch_devs > /dev/null 2>&1 - - # clear the WARN_ONCE state to allow a potential problem - # to be reported for each test - (echo 1 > $DEBUGFS_MNT/clear_warn_once) > /dev/null 2>&1 - - test_start_time="$(date +"%F %T")" - if [ "$DUMP_OUTPUT" = true ]; then - _run_seq 2>&1 | tee $tmp.out - # Because $? would get tee's return code - sts=${PIPESTATUS[0]} - else - _run_seq >$tmp.out 2>&1 - sts=$? - fi - - # If someone sets kernel.core_pattern or kernel.core_uses_pid, - # coredumps generated by fstests might have a longer name than - # just "core". Use globbing to find the most common patterns, - # assuming there are no other coredump capture packages set up. - local cores=0 - for i in core core.*; do - test -f "$i" || continue - if ((cores++ == 0)); then - _dump_err_cont "[dumped core]" - fi - (_adjust_oom_score 250; _save_coredump "$i") - tc_status="fail" - done - - if [ -f $seqres.notrun ]; then - $timestamp && _timestamp - stop=`_wallclock` - $timestamp || echo -n "[not run] " - $timestamp && echo " [not run]" && \ - echo -n " $seqnum -- " - cat $seqres.notrun - tc_status="notrun" - _stash_test_status "$seqnum" "$tc_status" - - # Unmount the scratch fs so that we can wipe the scratch - # dev state prior to the next test run. - _scratch_unmount 2> /dev/null - continue; - fi - - if [ $sts -ne 0 ]; then - _dump_err_cont "[failed, exit status $sts]" - _test_unmount 2> /dev/null - _scratch_unmount 2> /dev/null - rm -f ${RESULT_DIR}/require_test* - rm -f ${RESULT_DIR}/require_scratch* - # Even though we failed, there may be something interesting in - # dmesg which can help debugging. - _check_dmesg - tc_status="fail" - else - # The test apparently passed, so check for corruption - # and log messages that shouldn't be there. Run the - # checking tools from a subshell with adjusted OOM - # score so that the OOM killer will target them instead - # of the check script itself. - (_adjust_oom_score 250; _check_filesystems) || tc_status="fail" - _check_dmesg || tc_status="fail" - - # Save any coredumps from the post-test fs checks - for i in core core.*; do - test -f "$i" || continue - if ((cores++ == 0)); then - _dump_err_cont "[dumped core]" - fi - (_adjust_oom_score 250; _save_coredump "$i") - tc_status="fail" - done - fi - - # Reload the module after each test to check for leaks or - # other problems. - if [ -n "${TEST_FS_MODULE_RELOAD}" ]; then - _test_unmount 2> /dev/null - _scratch_unmount 2> /dev/null - modprobe -r fs-$FSTYP - modprobe fs-$FSTYP - fi - - # Scan for memory leaks after every test so that associating - # a leak to a particular test will be as accurate as possible. - _check_kmemleak || tc_status="fail" - - # test ends after all checks are done. - $timestamp && _timestamp - stop=`_wallclock` - - if [ ! -f $seq.out ]; then - _dump_err "no qualified output" - tc_status="fail" - _stash_test_status "$seqnum" "$tc_status" - continue; - fi - - # coreutils 8.16+ changed quote formats in error messages - # from `foo' to 'foo'. Filter old versions to match the new - # version. - sed -i "s/\`/\'/g" $tmp.out - if diff $seq.out $tmp.out >/dev/null 2>&1 ; then - if [ "$tc_status" != "fail" ]; then - echo "$seqnum `expr $stop - $start`" >>$tmp.time - echo -n " `expr $stop - $start`s" - fi - echo "" - else - _dump_err "- output mismatch (see $seqres.out.bad)" - mv $tmp.out $seqres.out.bad - $diff $seq.out $seqres.out.bad | { - if test "$DIFF_LENGTH" -le 0; then - cat - else - head -n "$DIFF_LENGTH" - echo "..." - echo "(Run '$diff $here/$seq.out $seqres.out.bad'" \ - " to see the entire diff)" - fi; } | sed -e 's/^\(.\)/ \1/' - tc_status="fail" - fi - if [ -f $seqres.hints ]; then - if [ "$tc_status" == "fail" ]; then - echo - cat $seqres.hints - else - rm -f $seqres.hints - fi - fi - _stash_test_status "$seqnum" "$tc_status" + _te_run_test $seq done # Reset these three variables so that unmount output doesn't getcommon/test_exec new file +83
-0@@ -0,0 +1,352 @@ +##/bin/bash +# SPDX-License-Identifier: GPL-2.0+ +# Copyright (c) 2000-2002,2006 Silicon Graphics, Inc. All Rights Reserved. +# Copyright (c) 2025 Red Hat, Inc. All Rights Reserved. +# +# Test execution functions +# +# This file contains the functions to run a test and capture the results. The +# caller context must source all the dependencies this code requires, as well +# as provide certain global variables and define certain +# functions to run and track the test status of the test being run. +# +# Any function or variable that is public should have a "_te_" prefix. + +# test status tracking variables. These are externally visible so that the +# caller can do it's own test reporting based on the tracking provided by these +# variables. +_te_try=() +_te_bad=() +_te_notrun=() +_te_loop_status=() +_te_emit_timestamps="" +_te_dry_run="" + +_te_wipe_counters() +{ + _te_try=() + _te_notrun=() + _te_bad=() + _te_loop_status=() +} + +_te_oom_score_adj="/proc/self/oom_score_adj" +_te_adjust_oom_score() { + test -w "${_te_oom_score_adj}" && echo "$1" > "${_te_oom_score_adj}" +} + +_te_timestamp() +{ + if [ "$_te_emit_timestamps" == "true" ]; then + local now=`date "+%T"` + echo -n " [$now]" + fi +} + +_te_check_filesystems() +{ + local ret=0 + + if [ -f ${RESULT_DIR}/require_test ]; then + if ! _check_test_fs ; then + ret=1 + echo "Trying to repair broken TEST_DEV file system" + _repair_test_fs + _test_mount + fi + rm -f ${RESULT_DIR}/require_test* + else + _test_unmount 2> /dev/null + fi + if [ -f ${RESULT_DIR}/require_scratch ]; then + _check_scratch_fs || ret=1 + rm -f ${RESULT_DIR}/require_scratch* + fi + _scratch_unmount 2> /dev/null + return $ret +} + +# retain files which would be overwritten in subsequent reruns of the same test +_te_stash_fail_loop_files() { + local seq_prefix="${REPORT_DIR}/${1}" + local cp_suffix="$2" + + for i in ".full" ".dmesg" ".out.bad" ".notrun" ".core" ".hints"; do + rm -f "${seq_prefix}${i}${cp_suffix}" + if [ -f "${seq_prefix}${i}" ]; then + cp "${seq_prefix}${i}" "${seq_prefix}${i}${cp_suffix}" + fi + done +} + +# Retain in @bad / @notrun the result of the just-run @test_seq. @try array +# entries are added prior to execution.
common/test_exec line 93
Quoted 10 lines
common/test_exec new file +10
-0+_te_stash_test_status() { + local test_seq="$1" + local test_status="$2" + + if $do_report && [[ $test_status != "expunge" ]]; then + _make_testcase_report "$section" "$test_seq" \ + "$test_status" "$((stop - start))" + fi + + if ((${#loop_status[*]} > 0)); then
Shouldn't the above be _te_loop_status?
Quoted 17 lines
common/test_exec new file +17
-0+ # continuing or completing rerun-on-failure loop + _te_stash_fail_loop_files "$test_seq" ".rerun${#loop_status[*]}" + _te_loop_status+=("$test_status") + if ((${#_te_loop_status[*]} > loop_on_fail)); then + printf "%s aggregate results across %d runs: " \ + "$test_seq" "${#loop_status[*]}" + awk "BEGIN { + n=split(\"${loop_status[*]}\", arr);"' + for (i = 1; i <= n; i++) + stats[arr[i]]++; + for (x in stats) + printf("%s=%d (%.1f%%)", + (i-- > n ? x : ", " x), + stats[x], 100 * stats[x] / n); + }' + echo + _te_loop_status=()
common/test_exec line 120
Quoted 10 lines
common/test_exec new file +10
-0+ fi + return # only stash @bad result for initial failure in loop + fi + + case "$test_status" in + fail) + if ((loop_on_fail > 0)); then + # initial failure, start rerun-on-failure loop + _te_stash_fail_loop_files "$test_seq" ".rerun0" + loop_status+=("$test_status")
Same: Shouldn't the above be _te_loop_status? We should only use _te_loop_status, isn't it?
Quoted 14 lines
common/test_exec new file +14
-0+ fi + _te_bad+=("$test_seq") + ;; + list|notrun) + _te_notrun+=("$test_seq") + ;; + pass|expunge) + ;; + *) + echo "Unexpected test $test_seq status: $test_status" + ;; + esac +} +
common/test_exec line 144
Quoted 10 lines
common/test_exec new file +10
-0+# Run a test. +# +# This currently relies on the caller defining global variables for test +# reporting and status tracking: +# +# REPORT_DIR +# tmp +# +# The caller also needs to define the functions _run_seq and _kill_seq +# for executing and killing specific test binaries.
I couldn't understand this part. _run_seq() is already defined, isn't it? --NR
Quoted 208 lines
common/test_exec new file +208
-0+# +_te_run_test() +{ + local seq="$1" + local tc_status="pass" + local start + local stop + local sts + + # the filename for the test and the name output are different. + # we don't include the tests/ directory in the name output. + export seqnum=$(_tl_strip_src_dir $seq) + group=${seqnum%%/*} + export RESULT_DIR="$REPORT_DIR/$group" + seqres="$REPORT_DIR/$seqnum" + + echo -n "$seqnum" + + if [ "$_te_dry_run" == "true" ]; then + if _tl_expunge_test $seqnum; then + tc_status="expunge" + else + echo + start=0 + stop=0 + tc_status="list" + fi + _te_stash_test_status "$seqnum" "$tc_status" + return + fi + + if [ ! -f $seq ]; then + echo " - no such test?" + _te_stash_test_status "$seqnum" "$tc_status" + return + fi + + # really going to try and run this one + mkdir -p $RESULT_DIR + rm -f ${RESULT_DIR}/require_scratch* + rm -f ${RESULT_DIR}/require_test* + rm -f $seqres.out.bad $seqres.hints + + # check if we really should run it + if _tl_expunge_test $seqnum; then + tc_status="expunge" + _te_stash_test_status "$seqnum" "$tc_status" + return + fi + + # record that we really tried to run this test. + if ((!${#loop_status[*]})); then + _te_try+=("$seqnum") + fi + + awk 'BEGIN {lasttime=" "} \ + $1 == "'$seqnum'" {lasttime=" " $2 "s ... "; exit} \ + END {printf "%s", lasttime}' "$check.time" + rm -f core $seqres.notrun + + start=`_wallclock` + _te_timestamp + [ ! -x $seq ] && chmod u+x $seq # ensure we can run it + $LOGGER_PROG "run xfstest $seqnum" + if [ -w /dev/kmsg ]; then + export date_time=`date +"%F %T"` + echo "run fstests $seqnum at $date_time" > /dev/kmsg + # _check_dmesg depends on this log in dmesg + touch ${RESULT_DIR}/check_dmesg + rm -f ${RESULT_DIR}/dmesg_filter + fi + _try_wipe_scratch_devs > /dev/null 2>&1 + + # clear the WARN_ONCE state to allow a potential problem + # to be reported for each test + (echo 1 > $DEBUGFS_MNT/clear_warn_once) > /dev/null 2>&1 + + test_start_time="$(date +"%F %T")" + if [ "$DUMP_OUTPUT" = true ]; then + _run_seq 2>&1 | tee $tmp.out + # Because $? would get tee's return code + sts=${PIPESTATUS[0]} + else + _run_seq >$tmp.out 2>&1 + sts=$? + fi + + # If someone sets kernel.core_pattern or kernel.core_uses_pid, + # coredumps generated by fstests might have a longer name than + # just "core". Use globbing to find the most common patterns, + # assuming there are no other coredump capture packages set up. + local cores=0 + for i in core core.*; do + test -f "$i" || continue + if ((cores++ == 0)); then + _dump_err_cont "[dumped core]" + fi + (_te_adjust_oom_score 250; _save_coredump "$i") + tc_status="fail" + done + + if [ -f $seqres.notrun ]; then + stop=`_wallclock` + if [ "$_te_emit_timestamps" == "true" ]; then + _te_timestamp + echo " [not run]" + echo -n " $seqnum -- " + else + echo -n "[not run] " + fi + cat $seqres.notrun + tc_status="notrun" + _te_stash_test_status "$seqnum" "$tc_status" + + # Unmount the scratch fs so that we can wipe the scratch + # dev state prior to the next test run. + _scratch_unmount 2> /dev/null + return + fi + + if [ $sts -ne 0 ]; then + _dump_err_cont "[failed, exit status $sts]" + _test_unmount 2> /dev/null + _scratch_unmount 2> /dev/null + rm -f ${RESULT_DIR}/require_test* + rm -f ${RESULT_DIR}/require_scratch* + # Even though we failed, there may be something interesting in + # dmesg which can help debugging. + _check_dmesg + tc_status="fail" + else + # The test apparently passed, so check for corruption + # and log messages that shouldn't be there. Run the + # checking tools from a subshell with adjusted OOM + # score so that the OOM killer will target them instead + # of the check script itself. + (_te_adjust_oom_score 250; _te_check_filesystems) || tc_status="fail" + _check_dmesg || tc_status="fail" + + # Save any coredumps from the post-test fs checks + for i in core core.*; do + test -f "$i" || continue + if ((cores++ == 0)); then + _dump_err_cont "[dumped core]" + fi + (_te_adjust_oom_score 250; _save_coredump "$i") + tc_status="fail" + done + fi + + # Reload the module after each test to check for leaks or + # other problems. + if [ -n "${TEST_FS_MODULE_RELOAD}" ]; then + _test_unmount 2> /dev/null + _scratch_unmount 2> /dev/null + modprobe -r fs-$FSTYP + modprobe fs-$FSTYP + fi + + # Scan for memory leaks after every test so that associating + # a leak to a particular test will be as accurate as possible. + _check_kmemleak || tc_status="fail" + + # test ends after all checks are done. + _te_timestamp + stop=`_wallclock` + + if [ ! -f $seq.out ]; then + _dump_err "no qualified output" + tc_status="fail" + _te_stash_test_status "$seqnum" "$tc_status" + return; + fi + + # coreutils 8.16+ changed quote formats in error messages + # from `foo' to 'foo'. Filter old versions to match the new + # version. + sed -i "s/\`/\'/g" $tmp.out + if diff $seq.out $tmp.out >/dev/null 2>&1 ; then + if [ "$tc_status" != "fail" ]; then + echo "$seqnum `expr $stop - $start`" >>$tmp.time + echo -n " `expr $stop - $start`s" + fi + echo "" + else + _dump_err "- output mismatch (see $seqres.out.bad)" + mv $tmp.out $seqres.out.bad + $diff $seq.out $seqres.out.bad | { + if test "$DIFF_LENGTH" -le 0; then + cat + else + head -n "$DIFF_LENGTH" + echo "..." + echo "(Run '$diff $here/$seq.out $seqres.out.bad'" \ + " to see the entire diff)" + fi; } | sed -e 's/^\(.\)/ \1/' + tc_status="fail" + fi + if [ -f $seqres.hints ]; then + if [ "$tc_status" == "fail" ]; then + echo + cat $seqres.hints + else + rm -f $seqres.hints + fi + fi + _te_stash_test_status "$seqnum" "$tc_status" +}
RFC PATCH 20/28 check-parallel: run tests directly without using check
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
Now that we've factored the code that runs individual tests out of check, we can start using it in check-parallel instead of calling check to do so.
I originally thought that:
We can do this by implementing a _run_seq() callback that runs the individual test in it's own private namespace, using the generic test execution and status tracking to record what happens with each test that is run by a given runner.
This allows the runner to destage a test at a time from the global list and execute it in a batched setup, all without requiring us to run check for each test.
But, no, no we can't do that. We have to mount the test filesystem inside the new private mount namespace context, otherwise all the other runner contexts see it and we reintroduce shared mount namespace problems...
Hence we need a helper script that sets up the test device and does all that admin stuff before it starts grabbing tests from the test list. This means that multiple tests will run inside the runner's private namespace; the helper is really now a very minimal version of check....
This requires that check-parallel also sets up the required environment for all tests to run in. We can do this now simply by sourcing common/config as this now sets up and exports the entire environment that tests rely on.
Note: this change is mainly for RFC purposes right now - it still needs more cleanup and probably some splitting up before it is ready to go. I'm including it more as a informational "this is why various factoring and rearrangement has been done" patch than a full review request...
--- check | 29 ++---------- check-parallel | 75 ++++++++++++------------------ common/config | 6 +++ common/rc | 2 + common/test_exec | 29 +++++++++++- common/test_list | 6 +++ tests/xfs/271 | 2 - tools/run_test.sh | 116 ++++++++++++++++++++++++++++++++++++++++++++++ 8 files changed, 191 insertions(+), 74 deletions(-) create mode 100755 tools/run_test.sh
check +5 -24
@@ -25,15 +25,6 @@ _err_msg="" # start the initialisation work now iam=check.$$ -# mkfs.xfs uses the presence of both of these variables to enable formerly -# supported tiny filesystem configurations that fstests use for fuzz testing -# in a controlled environment -export MSGVERB="text:action" -export QA_CHECK_FS=${QA_CHECK_FS:=true} - -# number of diff lines from a failed test, 0 for whole output -export DIFF_LENGTH=${DIFF_LENGTH:=10} - rm -f $tmp.list $tmp.tmp $tmp.grep $here/$iam.out $tmp.report.* $tmp.arglist . ./common/exit
@@ -188,6 +179,10 @@ fi # by default don't output timestamps _te_emit_timestamps=${TIMESTAMP:=} +# number of diff lines from a failed test, 0 for whole output +_te_diff_length=${DIFF_LENGTH:=10} + + # If the test config specified a soak test duration, see if there are any # unit suffixes that need converting to an integer seconds count. if [ -n "$SOAK_DURATION" ]; then
@@ -245,21 +240,7 @@ _wrapup() fi needwrap=false elif $needwrap; then - if [ -f $check.time -a -f $tmp.time ]; then - cat $check.time $tmp.time \ - | $AWK_PROG ' - { t[$1] = $2 } - END { - if (NR > 0) { - for (i in t) print i " " t[i] - } - }' \ - | sort -n >$tmp.out - mv $tmp.out $check.time - if $OPTIONS_HAVE_SECTIONS; then - cp $check.time ${REPORT_DIR}/check.time - fi - fi + _te_time_report set +x _global_log ""
check-parallel +29 -46
140 lines not on this page · raw message
common/config +6 -0
@@ -70,6 +70,12 @@ export LOAD_FACTOR=${LOAD_FACTOR:=1} export SOAK_DURATION=${SOAK_DURATION:=} export DEBUGFS_MNT=${DEBUGFS_MNT:="/sys/kernel/debug"} +# mkfs.xfs uses the presence of both of these variables to enable formerly +# supported tiny filesystem configurations that fstests use for fuzz testing +# in a controlled environment +export MSGVERB="text:action" +export QA_CHECK_FS=${QA_CHECK_FS:=true} + # some constants for overlayfs setup export OVL_UPPER="ovl-upper" export OVL_LOWER="ovl-lower"
common/rc +2 -0
@@ -2936,6 +2936,8 @@ _require_xfs_io_command() _notrun "O_TMPFILE is not supported" ;; "fsmap") + df -h >> $seqres.full 2>&1 + df -i >> $seqres.full 2>&1 testio=`$XFS_IO_PROG -f -c "fsmap" $testfile 2>&1` echo $testio | grep -q "Inappropriate ioctl" && \ _notrun "xfs_io $command support is missing"
common/test_exec +27 -2
@@ -21,6 +21,12 @@ _te_notrun=() _te_loop_status=() _te_emit_timestamps="" _te_dry_run="" +_te_diff_length=10 + +# This is needed because callers don't always source common/report and +# if $do_report is undefined or true we'll try to call _make_testcase_report() +# without it being defined. +do_report=false _te_wipe_counters() {
@@ -43,6 +49,25 @@ _te_timestamp() fi } +_te_time_report() +{ + if [ -f $check.time -a -f $tmp.time ]; then + cat $check.time $tmp.time \ + | $AWK_PROG ' + { t[$1] = $2 } + END { + if (NR > 0) { + for (i in t) print i " " t[i] + } + }' \ + | sort -n >$tmp.out + mv $tmp.out $check.time + if $OPTIONS_HAVE_SECTIONS; then + cp $check.time ${REPORT_DIR}/check.time + fi + fi +} + _te_check_filesystems() { local ret=0
@@ -330,10 +355,10 @@ _te_run_test() _dump_err "- output mismatch (see $seqres.out.bad)" mv $tmp.out $seqres.out.bad $diff $seq.out $seqres.out.bad | { - if test "$DIFF_LENGTH" -le 0; then + if test "$_te_diff_length" -le 0; then cat else - head -n "$DIFF_LENGTH" + head -n "$_te_diff_length" echo "..." echo "(Run '$diff $here/$seq.out $seqres.out.bad'" \ " to see the entire diff)"
common/test_list +6 -0
@@ -160,6 +160,11 @@ _tl_prepare_test_list() trim_test_list $list done + # Remove expunged tests + for f in "${_tl_exclude_tests[@]}"; do + trim_test_list tests/$f + done + # sort the list of tests into numeric order unless we're running tests # in the exact order specified if ! $_tl_exact_order; then
@@ -194,6 +199,7 @@ _tl_expunge_test() return 1 } + _tl_setup_exclude_tests() { local list="$1"
tests/xfs/271 +0 -2
@@ -22,8 +22,6 @@ _cleanup() _require_xfs_scratch_rmapbt _require_xfs_io_command "fsmap" -rm -f "$seqres.full" - echo "Format and mount" _scratch_mkfs > "$seqres.full" 2>&1 _scratch_mount
tools/run_test.sh new file +116 -0
116 lines not on this page · raw message
Signature · 1 line
2.45.2
Re: [PATCH 20/28] [RFC] check-parallel: run tests directly without using check
Quoted 13 lines · Dave Chinner
From: Dave Chinner <[email protected]>
Now that we've factored the code that runs individual tests out of check, we can start using it in check-parallel instead of calling check to do so.
I originally thought that:
We can do this by implementing a
_run_seq()callback that runs the individual test in it's own private namespace, using the generic test execution and status tracking to record what happens with each test that is run by a given runner.
Quoted 8 lines
This allows the runner to destage a test at a time from the global list and execute it in a batched setup, all without requiring us to run check for each test.
But, no, no we can't do that. We have to mount the test filesystem inside the new private mount namespace context, otherwise all the other runner contexts see it and we reintroduce shared mount namespace problems...
And earlier before this patch, we were using tools/run_privatens to direcly invoke check in a private namespace and then let check do the rest of the work. Now, since check-parallel is moving away from check, it is the responsibility of the check-parallel to setup a private namespace, then setup the devices in the private namespace, then run the test and finally get the results. Am I correct?
Quoted 393 lines
Hence we need a helper script that sets up the test device and does all that admin stuff before it starts grabbing tests from the test list. This means that multiple tests will run inside the runner's private namespace; the helper is really now a very minimal version of check....
This requires that check-parallel also sets up the required environment for all tests to run in. We can do this now simply by sourcing common/config as this now sets up and exports the entire environment that tests rely on.
Note: this change is mainly for RFC purposes right now - it still needs more cleanup and probably some splitting up before it is ready to go. I'm including it more as a informational "this is why various factoring and rearrangement has been done" patch than a full review request...
Signed-off-by Dave Chinner <[email protected]>--- check | 29 ++---------- check-parallel | 75 ++++++++++++------------------ common/config | 6 +++ common/rc | 2 + common/test_exec | 29 +++++++++++- common/test_list | 6 +++ tests/xfs/271 | 2 - tools/run_test.sh | 116 ++++++++++++++++++++++++++++++++++++++++++++++ 8 files changed, 191 insertions(+), 74 deletions(-) create mode 100755 tools/run_test.shcheck +5
-24@@ -25,15 +25,6 @@ _err_msg="" # start the initialisation work now iam=check.$$ -# mkfs.xfs uses the presence of both of these variables to enable formerly -# supported tiny filesystem configurations that fstests use for fuzz testing -# in a controlled environment -export MSGVERB="text:action" -export QA_CHECK_FS=${QA_CHECK_FS:=true} - -# number of diff lines from a failed test, 0 for whole output -export DIFF_LENGTH=${DIFF_LENGTH:=10} - rm -f $tmp.list $tmp.tmp $tmp.grep $here/$iam.out $tmp.report.* $tmp.arglist . ./common/exit@@ -188,6 +179,10 @@ fi # by default don't output timestamps _te_emit_timestamps=${TIMESTAMP:=} +# number of diff lines from a failed test, 0 for whole output +_te_diff_length=${DIFF_LENGTH:=10} + + # If the test config specified a soak test duration, see if there are any # unit suffixes that need converting to an integer seconds count. if [ -n "$SOAK_DURATION" ]; then@@ -245,21 +240,7 @@ _wrapup() fi needwrap=false elif $needwrap; then - if [ -f $check.time -a -f $tmp.time ]; then - cat $check.time $tmp.time \ - | $AWK_PROG ' - { t[$1] = $2 } - END { - if (NR > 0) { - for (i in t) print i " " t[i] - } - }' \ - | sort -n >$tmp.out - mv $tmp.out $check.time - if $OPTIONS_HAVE_SECTIONS; then - cp $check.time ${REPORT_DIR}/check.time - fi - fi + _te_time_report set +x _global_log ""check-parallel +29
-46@@ -18,13 +18,13 @@ run_section="" exclude_section="" iam="check-parallel" -tmp=/tmp/check-parallel.$$ -test_list="$tmp.test_list" +export here=`pwd` . ./common/exit . ./common/test_names . ./common/test_list . ./common/config-sections +. ./common/config usage() {@@ -155,6 +155,18 @@ elif [ -d "$basedir/runner-0/" ]; then prev_results=`ls -tr $basedir/runner-0/ | grep results | tail -1` fi +# We keep the test list in the base directory because it needs to be common +# across all test runners and their private namespaces. Because runners execute +# in private /tmp/ instances, we can't keep it there, so use the shared +# $basedir that hosts all the runners as the location of the test list. +test_list="$basedir/test_list" + +# Similarly, set our tmp dir to be under the shared basedir so that all the +# temp files that stuff like test execution use for expunge lists work +# appropriately. +export tmp="$basedir/tmp/check-parallel.$$" +mkdir -p $basedir/tmp + # grab all previously run tests and order them from highest runtime to lowest # We are going to try to run the longer tests first, hopefully so we can avoid # massive thundering herds trying to run lots of really short tests in parallel@@ -219,22 +231,6 @@ setup_test_list() echo $_tl_tests |sed -e 's/ /\n/g' | tac > $test_list } -# Grab the next test to be run from the tail of the file. -# Returns an empty string if there is no tests remaining to run. -# File operations are run under flock so concurrent gets are serialised against -# each other. -get_next_test() -{ - local test_file="$test_list.$1" - local test= - - flock 99 - test=$(tail -1 $test_file) - sed -i "\,$test,d" $test_file - flock -u 99 - echo $test -} - _create_loop_device() {@@ -260,32 +256,6 @@ _destroy_loop_device() losetup -d $dev || _fail "Cannot destroy loop device $dev" } -run_tests() -{ - local section="$1" - local logfile="$2" - - exec 99<>$tmp.test_list_lock - - local test_to_run=$(get_next_test $section) - - # Run the tests in it's own mount namespace, as per the comment below - # that precedes making the basedir a private mount. - # - # Similarly, we need to run check in it's own PID namespace so that - # operations like pkill only affect the runner instance, not globally - # kill processes from other check instances. - while [ -n "$test_to_run" ]; do - echo -n " $test_to_run " - unset FSTESTS_ISOL - if ! _tl_expunge_test $test_to_run; then - tools/run_privatens ./check -s $section $test_to_run >> $logfile 2>&1 - fi - - test_to_run=$(get_next_test $section) - done -} - runner_go() {@@ -301,6 +271,8 @@ runner_go() local _results=$me/results-$2 local section=$3 + export tmp=/tmp/check-parallel-runner-$id + mkdir -p $me xfs_io -f -c "truncate $TEST_DEV_SIZE" $_test@@ -323,6 +295,7 @@ runner_go() export LOGWRITES_DEV=$(_create_loop_device $_logwrites) export RESULT_BASE=$_results + export REPORT_DIR="$RESULT_BASE/$section" mkdir -p $TEST_DIR mkdir -p $SCRATCH_MNT@@ -338,7 +311,7 @@ runner_go() # export DUMP_CORRUPT_FS=1 - run_tests $section $_results/log + tools/run_test.sh $basedir $test_list.$section >> $_results/log 2>&1 wait sleep 1@@ -351,6 +324,7 @@ runner_go() _destroy_loop_device $SCRATCH_RTDEV _destroy_loop_device $SCRATCH_LOGDEV _destroy_loop_device $LOGWRITES_DEV + rm -f $tmp.* grep -q Failures: $_results/log if [ $? -eq 0 ]; then@@ -366,6 +340,7 @@ run_section() local now="$2" local results="$basedir/*/results-$now" local i + local threads=$runners echo $run_section |grep -qw $section || return echo $exclude_section |grep -qw $section && return@@ -391,7 +366,15 @@ run_section() fi cp $test_list $test_list.$section - for ((i = 0; i < $runners; i++)); do + # only run as many runners are there are tests to run + i=$(cat $test_list.$section | wc -l) + if [ "$i" -lt "$runners" ]; then + threads=$i + fi + + echo Test list $test_list.$section contains $i tests + echo Running $threads tests concurrently + for ((i = 0; i < $threads; i++)); do runner_go $i $now $section & done waitcommon/config +6
-0@@ -70,6 +70,12 @@ export LOAD_FACTOR=${LOAD_FACTOR:=1} export SOAK_DURATION=${SOAK_DURATION:=} export DEBUGFS_MNT=${DEBUGFS_MNT:="/sys/kernel/debug"} +# mkfs.xfs uses the presence of both of these variables to enable formerly +# supported tiny filesystem configurations that fstests use for fuzz testing +# in a controlled environment +export MSGVERB="text:action" +export QA_CHECK_FS=${QA_CHECK_FS:=true} + # some constants for overlayfs setup export OVL_UPPER="ovl-upper" export OVL_LOWER="ovl-lower"common/rc +2
-0@@ -2936,6 +2936,8 @@ _require_xfs_io_command() _notrun "O_TMPFILE is not supported" ;; "fsmap") + df -h >> $seqres.full 2>&1 + df -i >> $seqres.full 2>&1 testio=`$XFS_IO_PROG -f -c "fsmap" $testfile 2>&1` echo $testio | grep -q "Inappropriate ioctl" && \ _notrun "xfs_io $command support is missing"common/test_exec +27
-2@@ -21,6 +21,12 @@ _te_notrun=() _te_loop_status=() _te_emit_timestamps="" _te_dry_run="" +_te_diff_length=10 + +# This is needed because callers don't always source common/report and +# if $do_report is undefined or true we'll try to call _make_testcase_report() +# without it being defined. +do_report=false _te_wipe_counters() {@@ -43,6 +49,25 @@ _te_timestamp() fi } +_te_time_report() +{ + if [ -f $check.time -a -f $tmp.time ]; then + cat $check.time $tmp.time \ + | $AWK_PROG ' + { t[$1] = $2 } + END { + if (NR > 0) { + for (i in t) print i " " t[i] + } + }' \ + | sort -n >$tmp.out + mv $tmp.out $check.time + if $OPTIONS_HAVE_SECTIONS; then + cp $check.time ${REPORT_DIR}/check.time + fi + fi +} + _te_check_filesystems() { local ret=0@@ -330,10 +355,10 @@ _te_run_test() _dump_err "- output mismatch (see $seqres.out.bad)" mv $tmp.out $seqres.out.bad $diff $seq.out $seqres.out.bad | { - if test "$DIFF_LENGTH" -le 0; then + if test "$_te_diff_length" -le 0; then cat else - head -n "$DIFF_LENGTH" + head -n "$_te_diff_length" echo "..." echo "(Run '$diff $here/$seq.out $seqres.out.bad'" \ " to see the entire diff)"common/test_list +6
-0@@ -160,6 +160,11 @@ _tl_prepare_test_list() trim_test_list $list done + # Remove expunged tests + for f in "${_tl_exclude_tests[@]}"; do + trim_test_list tests/$f + done + # sort the list of tests into numeric order unless we're running tests # in the exact order specified if ! $_tl_exact_order; then@@ -194,6 +199,7 @@ _tl_expunge_test() return 1 } + _tl_setup_exclude_tests() { local list="$1"tests/xfs/271 +0
-2@@ -22,8 +22,6 @@ _cleanup() _require_xfs_scratch_rmapbt _require_xfs_io_command "fsmap" -rm -f "$seqres.full" - echo "Format and mount" _scratch_mkfs > "$seqres.full" 2>&1 _scratch_mounttools/run_test.sh new file +25
-0@@ -0,0 +1,116 @@ +#!/bin/bash +# SPDX-License-Identifier: GPL-2.0 +# Copyright (c) 2025 Red Hat, Inc. All Rights Reserved. +# +# Test run helper for check-parallel +# +# check-parallel sets up all the devices and environments needs to run tests, +# but we can't set up for a test until we are executing in a private pid/mount +# namespace. +# +# This means we cannot simply run the test itself in a private name space; they +# require things like the test device to already be mounted, and we require a +# private mount namespace before we start mounting devices. +# +# Hence we need a helper that first creates a private namespace, then +# does all the setup work for tests to run, then iterates tests until there +# are no more tests to run. +# +# The tests to run are held in a shared file in $basedir so that we set up +# a private /tmp for the namespace and no lose access to the test list. + +# re-execute in a private namespace as a first step +if [ -z "${FSTESTS_ISOL}" ]; then + if [ -z "$1" ] || [ "$1" = "--help" ]; then + echo "Usage: $0 basedir test_list "
tools/run_test.sh line 35
Quoted 10 lines
tools/run_test.sh new file +10
-0+ exit 1 + fi + + if [ ! -d "$1" ]; then + echo "invalid basedir ($1) specified" + exit 1 + fi + + FSTESTS_ISOL=privatens exec "$(dirname "$0")/../src/nsexec" -z -m -p "$0" "$@" + exit $?
So when the first time tools/run_test.sh gets invoked for a particular runner, it sets FSTESTS_ISOL, then does an exec with $0 i.e, the current script name. So tools/run_test.sh again gets invoked (as a part of exec) but this time FSTESTS_ISOL is set, so this if condition isn't executed and hence we move to the rest of the script. Simarly from second time, FSTESTS_ISOL is still set, so we don't enter the if condition and move to the rest of the script. This is repeated for all the runners. Is that correct?
Quoted 39 lines
tools/run_test.sh new file +39
-0+fi + +# Everything past this point runs in a private namespace. +# +# We set up private mounts for /proc and /tmp so they aren't visible outside +# this mount namespace and it's children. +for path in /proc /tmp; do + mountpoint "$path" >/dev/null && \ + mount --make-private "$path" +done +mount -t proc proc /proc +mount -t tmpfs tmpfs /tmp + +echo $PWD $FSTYP + +. ./common/exit +. ./common/test_names +. ./common/test_list +. ./common/test_exec +. ./common/rc + +basedir="$1" +test_file="$2" +_te_diff_length=${DIFF_LENGTH:=10} +export tmp=/tmp/run-helper.$$ + +# XXX: should be a _te_diff variable +diff='diff -u' + +# Grab the next test to be run from the tail of the file. +# Returns an empty string if there is no tests remaining to run. +# File operations are run under flock so concurrent gets are serialised against +# each other. +get_next_test() +{ + local test= + + flock 99 + test=$(tail -1 $test_file)
tools/run_test.sh line 84
Quoted 10 lines
tools/run_test.sh new file +10
-0+ sed -i "\,$test,d" $test_file + flock -u 99 + echo $test +} + +_run_seq() +{ + ./$seq + return $? +}
Why can't we use the _run_seq() defined in "check"? --NR
Quoted 32 lines
tools/run_test.sh new file +32
-0+ +exec 99<>$test_file.lock + +# XXX - refactor this back to a test_exec variable +check="$RESULT_BASE/check" +touch $check.time + +init_rc + +test_to_run=$(get_next_test) +while [ -n "$test_to_run" ]; do + _te_run_test "tests/$test_to_run" + + test_to_run=$(get_next_test) +done + +_te_time_report + +if ((${#_te_try[*]} > 0)); then + echo "Ran: ${_te_try[*]}" +fi + +if ((${#_te_notrun[*]} > 0)); then + echo "Not run: ${_te_notrun[*]}" +fi + +if ((${#_te_bad[*]} > 0)); then + echo "Failures: ${_te_bad[*]}" + echo "Failed ${#_te_bad[*]} of ${#_te_try[*]} tests" +else + echo "Passed all ${#_te_try[*]} tests" +fi
PATCH 21/28 generic/531: limit max files per CPU
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
Currently g/531 runs t_open_files on every CPU, and with default kernel settings that means 50,000 files per CPU are tested. On 64p machines this means the test tries to create and unlink over 3 million files. This takes a long time:
Ten slowest tests - runtime in seconds:
generic/531 534
.....
Yet generic/531 is included in the 'quick' test group. It is anything but "quick" on large CPU count systems.
Further, small filesystems like are typically used for fstests do not have the inherent concurrency to scale out this workload effectively. Even using the mkfs.xfs concurrency options requires using >250GB scratch devices on 64p machines because it won't make AGs smaller than 4GB. Hence to get 64-way concurrency in the filesystem, we need huge devices to be set up, and that's not really practical for check-parallel.
Hence limit the total number of files this test will create to a sane number, and distribute them over all the CPUs so that the test runtime does not blow out on big systems. LOAD_FACTOR can still be used to increase runtime of the test by increasing the total number of files created.
Limiting the total number of files created brings g/531 system back into the "quick" test range on a 64p system:
generic/531 5s
--- tests/generic/531 | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)
tests/generic/531 +4 -4
@@ -29,13 +29,13 @@ _scratch_mount # Try to load up all the CPUs, two threads per CPU. nr_cpus=$(( $(getconf _NPROCESSORS_ONLN) * 2 )) -# Set ULIMIT_NOFILE to min(file-max / $nr_cpus / 2, 50000 files per LOAD_FACTOR) +# Set ULIMIT_NOFILE to min(file-max / 2, 100000) / $nr_cpus files per LOAD_FACTOR) # so that this test doesn't take forever or OOM the box -max_files=$((50000 * LOAD_FACTOR)) -max_allowable_files=$(( $(cat /proc/sys/fs/file-max) / $nr_cpus / 2 )) +max_files=$((100000 * LOAD_FACTOR)) +max_allowable_files=$(( $(cat /proc/sys/fs/file-max) / 2 )) test $max_allowable_files -gt 0 && test $max_files -gt $max_allowable_files && \ max_files=$max_allowable_files -ulimit -n $max_files +ulimit -n $((max_files / nr_cpus)) # Open a lot of unlinked files echo create >> $seqres.full
Signature · 1 line
2.45.2
Re: [PATCH 21/28] generic/531: limit max files per CPU
Quoted 28 lines · Dave Chinner
From: Dave Chinner <[email protected]>
Currently g/531 runs
t_open_fileson every CPU, and with default kernel settings that means 50,000 files per CPU are tested. On 64p machines this means the test tries to create and unlink over 3 million files. This takes a long time:Ten slowest tests - runtime in seconds:
generic/531 534
.....Yet generic/531 is included in the 'quick' test group. It is anything but "quick" on large CPU count systems.
Further, small filesystems like are typically used for fstests do not have the inherent concurrency to scale out this workload effectively. Even using the mkfs.xfs concurrency options requires using >250GB scratch devices on 64p machines because it won't make AGs smaller than 4GB. Hence to get 64-way concurrency in the filesystem, we need huge devices to be set up, and that's not really practical for check-parallel.
Hence limit the total number of files this test will create to a sane number, and distribute them over all the CPUs so that the test runtime does not blow out on big systems.
LOAD_FACTORcan still be used to increase runtime of the test by increasing the total number of files created.
Quoted 4 lines
Limiting the total number of files created brings g/531 system back into the "quick" test range on a 64p system:
generic/531 5s
Right, on my system with 16p, I too noticed the reduced test time. generic/531 24s ... 199s (24s with this change, 199s on for-next)
--NR
Quoted 28 lines
Signed-off-by Dave Chinner <[email protected]>--- tests/generic/531 | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)tests/generic/531 +4
-4@@ -29,13 +29,13 @@ _scratch_mount # Try to load up all the CPUs, two threads per CPU. nr_cpus=$(( $(getconf _NPROCESSORS_ONLN) * 2 )) -# Set ULIMIT_NOFILE to min(file-max / $nr_cpus / 2, 50000 files per LOAD_FACTOR) +# Set ULIMIT_NOFILE to min(file-max / 2, 100000) / $nr_cpus files per LOAD_FACTOR) # so that this test doesn't take forever or OOM the box -max_files=$((50000 * LOAD_FACTOR)) -max_allowable_files=$(( $(cat /proc/sys/fs/file-max) / $nr_cpus / 2 )) +max_files=$((100000 * LOAD_FACTOR)) +max_allowable_files=$(( $(cat /proc/sys/fs/file-max) / 2 )) test $max_allowable_files -gt 0 && test $max_files -gt $max_allowable_files && \ max_files=$max_allowable_files -ulimit -n $max_files +ulimit -n $((max_files / nr_cpus)) # Open a lot of unlinked files echo create >> $seqres.full
PATCH 22/28 fsync-tester.c: use syncfs() rather than sync()
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
generic/311 runs each fsync-tester unit test 4 times, and there are 20 separate tests. At least 9 of those unit tests run sync() to flush dirty data periodically.
When running check-parallel, sync() can take a -long- time to run as there can be dozens of filesystems that need to be synced, not to mention sync getting hung up behind all the mount and unmounts that are also being run. This results in:
Ten slowest tests - runtime in seconds:
generic/311 419
This test running for a really long time.
Convert the sync() calls to syncfs() so that they only try to sync the filesystem under test and not the entire system. This avoids interactions and delays with other tests and mount/unmount operations, hence allowing both the test and the overall check-parallel operation to run faster:
generic/311 166s
--- src/fsync-tester.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
src/fsync-tester.c +1 -1
@@ -131,7 +131,7 @@ static int test_three(int *max_blocks, int prealloc, int rand_fsync,/* Force a transaction commit in between just for fun */ if (blocks == sync_block && (do_sync || drop_caches)) {if (do_sync) -sync (); +syncfs (test_fd);else sync_file_range(test_fd, 0, 0, SYNC_FILE_RANGE_WRITE|
Signature · 1 line
2.45.2
Re: [PATCH 22/28] fsync-tester.c: use syncfs() rather than sync()
Quoted 34 lines · Dave Chinner
From: Dave Chinner <[email protected]>
generic/311 runs each fsync-tester unit test 4 times, and there are 20 separate tests. At least 9 of those unit tests run
sync()to flush dirty data periodically.When running check-parallel,
sync()can take a -long- time to run as there can be dozens of filesystems that need to be synced, not to mention sync getting hung up behind all the mount and unmounts that are also being run. This results in:Ten slowest tests - runtime in seconds:
generic/311 419This test running for a really long time.
Convert the
sync()calls tosyncfs()so that they only try to sync the filesystem under test and not the entire system. This avoids interactions and delays with other tests and mount/unmount operations, hence allowing both the test and the overall check-parallel operation to run faster:generic/311 166s
Signed-off-by Dave Chinner <[email protected]>--- src/fsync-tester.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)src/fsync-tester.c +0
-0
src/fsync-tester.c static int test_three(int *max_blocks, int · line 136
Quoted 7 lines
src/fsync-tester.c +1
-1@@ -131,7 +131,7 @@ static int test_three(int *max_blocks, int prealloc, int rand_fsync,/* Force a transaction commit in between just for fun */ if (blocks == sync_block && (do_sync || drop_caches)) {if (do_sync) -sync (); +syncfs (test_fd);
Feel free to add
Quoted 3 lines
src/fsync-tester.c +0
-0else src/fsync-tester.c +0
-0sync_file_range(test_fd, 0, 0, SYNC_FILE_RANGE_WRITE|
PATCH 23/28 open-by-handle.c: use syncfs() rather than sync()
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
generic/467 runs open_by_handle at least 15 times. Each execution runs sync() at least once, sometimes three times.
When running check-parallel, sync() can take a -long- time to run as there can be dozens of filesystems that need to be synced, not to mention sync getting hung up behind all the mount and unmounts that are also being run. This results in:
Ten slowest tests - runtime in seconds:
generic/467 442
.....
This test running for a really long time.
Convert the sync() calls to syncfs() so that they only try to sync the filesystem under test and not the entire system. This avoids interactions and delays with other tests and mount/unmount operations, hence allowing both the test and the overall check-parallel operation to run faster:
generic/467 6s
--- src/open_by_handle.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
src/open_by_handle.c +3 -3
@@ -430,7 +430,7 @@ int main(int argc, char **argv) }/* sync to get the new inodes to hit the disk */ -sync (); +syncfs (mount_fd);/* * encode the file handles or read them from file (-i) and maybe store
@@ -563,10 +563,10 @@ int main(int argc, char **argv) }/* sync to get log forced for unlink transactions to hit the disk */ -sync (); +syncfs (mount_fd);/* sync once more FTW */ -sync (); +syncfs (mount_fd);/* * now drop the caches so that unlinked inodes are reclaimed and
Signature · 1 line
2.45.2
Re: [PATCH 23/28] open-by-handle.c: use syncfs() rather than sync()
Quoted 48 lines · Dave Chinner
From: Dave Chinner <[email protected]>
generic/467 runs
open_by_handleat least 15 times. Each execution runssync()at least once, sometimes three times.When running check-parallel,
sync()can take a -long- time to run as there can be dozens of filesystems that need to be synced, not to mention sync getting hung up behind all the mount and unmounts that are also being run. This results in:Ten slowest tests - runtime in seconds:
generic/467 442
.....This test running for a really long time.
Convert the
sync()calls tosyncfs()so that they only try to sync the filesystem under test and not the entire system. This avoids interactions and delays with other tests and mount/unmount operations, hence allowing both the test and the overall check-parallel operation to run faster:generic/467 6s
Signed-off-by Dave Chinner <[email protected]>--- src/open_by_handle.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)src/open_by_handle.c +1
-1@@ -430,7 +430,7 @@ int main(int argc, char **argv) }/* sync to get the new inodes to hit the disk */ -sync (); +syncfs (mount_fd);/* * encode the file handles or read them from file (-i) and maybe store
src/open_by_handle.c +0
-0@@ -563,10 +563,10 @@ int main(int argc, char **argv) }/* sync to get log forced for unlink transactions to hit the disk */
src/open_by_handle.c int main(int argc, char **argv) · line 569
Quoted 6 lines
src/open_by_handle.c +2
-2-sync (); +syncfs (mount_fd);/* sync once more FTW */ -sync (); +syncfs (mount_fd);
I don't see mount_fd being closed at the end. Maybe all the file descriptors are closed after the program finishes execution? Other than this:
Feel free to add
Quoted 4 lines
src/open_by_handle.c +0
-0/* * now drop the caches so that unlinked inodes are reclaimed and
Re: [PATCH 23/28] open-by-handle.c: use syncfs() rather than sync()
Quoted 52 lines · Nirjhar Roy (IBM)
Quoted 51 lines · Dave Chinner
From: Dave Chinner <[email protected]>
generic/467 runs
open_by_handleat least 15 times. Each execution runssync()at least once, sometimes three times.When running check-parallel,
sync()can take a -long- time to run as there can be dozens of filesystems that need to be synced, not to mention sync getting hung up behind all the mount and unmounts that are also being run. This results in:Ten slowest tests - runtime in seconds:
generic/467 442
.....This test running for a really long time.
Convert the
sync()calls tosyncfs()so that they only try to sync the filesystem under test and not the entire system. This avoids interactions and delays with other tests and mount/unmount operations, hence allowing both the test and the overall check-parallel operation to run faster:generic/467 6s
Signed-off-by Dave Chinner <[email protected]>--- src/open_by_handle.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)src/open_by_handle.c +1
-1@@ -430,7 +430,7 @@ int main(int argc, char **argv) }/* sync to get the new inodes to hit the disk */ -sync (); +syncfs (mount_fd);/* * encode the file handles or read them from file (-i) and maybe store
src/open_by_handle.c +0
-0@@ -563,10 +563,10 @@ int main(int argc, char **argv) }/* sync to get log forced for unlink transactions to hit the disk */
src/open_by_handle.c +1
-1-sync (); +syncfs (mount_fd);
Quoted 5 lines
src/open_by_handle.c int main(int argc, char **argv) · line 569
Quoted 3 lines
src/open_by_handle.c +1
-1/* sync once more FTW */ -sync (); +syncfs (mount_fd);I don't see
mount_fdbeing closed at the end. Maybe all the file descriptors are closed after the program finishes execution?
Yes, they are closed on exit. Again, simple test code that isn't running out of open files, so there's not a lot of reason for this patch series to care about such existing issues in the code right now.
-Dave.
Signature · 2 lines
Dave Chinner [email protected]
Re: [PATCH 23/28] open-by-handle.c: use syncfs() rather than sync()
Quoted 58 lines · Dave Chinner
Quoted 52 lines · Nirjhar Roy (IBM)
Quoted 51 lines · Dave Chinner
From: Dave Chinner <[email protected]>
generic/467 runs
open_by_handleat least 15 times. Each execution runssync()at least once, sometimes three times.When running check-parallel,
sync()can take a -long- time to run as there can be dozens of filesystems that need to be synced, not to mention sync getting hung up behind all the mount and unmounts that are also being run. This results in:Ten slowest tests - runtime in seconds:
generic/467 442
.....This test running for a really long time.
Convert the
sync()calls tosyncfs()so that they only try to sync the filesystem under test and not the entire system. This avoids interactions and delays with other tests and mount/unmount operations, hence allowing both the test and the overall check-parallel operation to run faster:generic/467 6s
Signed-off-by Dave Chinner <[email protected]>--- src/open_by_handle.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)src/open_by_handle.c +1
-1@@ -430,7 +430,7 @@ int main(int argc, char **argv) }/* sync to get the new inodes to hit the disk */ -sync (); +syncfs (mount_fd);/* * encode the file handles or read them from file (-i) and maybe store
src/open_by_handle.c +0
-0@@ -563,10 +563,10 @@ int main(int argc, char **argv) }/* sync to get log forced for unlink transactions to hit the disk */
src/open_by_handle.c +1
-1-sync (); +syncfs (mount_fd);Quoted 5 lines
src/open_by_handle.c
Quoted 3 lines
/* sync once more FTW */
src/open_by_handle.c +1
-1-sync (); +syncfs (mount_fd);I don't see
mount_fdbeing closed at the end. Maybe all the file descriptors are closed after the program finishes execution?
Quoted 4 lines
Yes, they are closed on exit. Again, simple test code that isn't running out of open files, so there's not a lot of reason for this patch series to care about such existing issues in the code right now.
Yeah, makes sense.
--NR
Quoted 2 lines
-Dave.
Signature · 3 lines
Nirjhar Roy Linux Kernel Developer IBM, Bangalore
PATCH 24/28 open-by-handle.c: use syncfs() rather than sync()
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
xfs/183 runs bulkstat_unlink_test to create 100 inodes, unlink on and bulkstat them. It takes a ridiculously long time to run under check-parallel because it runs sync() multiple times per iteration:
Ten slowest tests - runtime in seconds:
.....
xfs/183 328
When running check-parallel, sync() can take a -long- time to run as there can be dozens of filesystems that need to be synced, not to mention sync getting hung up behind all the mount and unmounts that are also being run.
Convert the sync() calls to syncfs() so that they only try to sync the filesystem under test and not the entire system. This avoids interactions and delays with other tests and mount/unmount operations, hence allowing both the test and the overall check-parallel operation to run faster:
xfs/183 4s
--- src/bulkstat_unlink_test.c | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-)
src/bulkstat_unlink_test.c +6 -6
@@ -88,7 +88,7 @@ main(int argc, char *argv[]) }if (chknb) {/* Get the original number of inodes (lazy) */ -sync (); +syncfs (fd[nfiles]);if (xfsctl (dirname, fd[nfiles],XFS_IOC_FSBULKSTAT , &a) !=0 ) {printf ("Warning (%s:%d), xfsctl(XFS_IOC_FSBULKSTAT) FAILED.\n" , __FILE__, __LINE__); }
@@ -118,7 +118,7 @@ main(int argc, char *argv[])*The files are still opened (but unlink()ed) , * we should have more inodes than before */ -sync (); +syncfs (fd[nfiles]); last_inode =0 ;if (xfsctl (dirname, fd[nfiles],XFS_IOC_FSBULKSTAT , &a) !=0 ) {printf ("Warning (%s:%d), xfsctl(XFS_IOC_FSBULKSTAT) FAILED.\n" , __FILE__, __LINE__);
@@ -139,7 +139,7 @@ main(int argc, char *argv[])* The files are now closed, we should be back to our, * previous inode count */ -sync (); +syncfs (fd[nfiles]); last_inode =0 ;if (xfsctl (dirname, fd[nfiles],XFS_IOC_FSBULKSTAT , &a) !=0 ) {printf ("Warning (%s:%d), xfsctl(XFS_IOC_FSBULKSTAT) FAILED.\n" , __FILE__, __LINE__);
@@ -150,7 +150,7 @@ main(int argc, char *argv[]) } } -sync (); +syncfs (fd[nfiles]); last_inode =0 ;for (;;) {if ((e =xfsctl (dirname, fd[nfiles],XFS_IOC_FSBULKSTAT , &a)) <0 ) {
@@ -173,11 +173,11 @@ main(int argc, char *argv[]) } } -close (fd[nfiles]);sprintf (fname,"rm -rf %s\n" , dirname);system (fname); -sync (); +syncfs (fd[nfiles]); +close (fd[nfiles]);sleep (2 );printf ("passed\n" ); }
Signature · 1 line
2.45.2
Re: [PATCH 24/28] open-by-handle.c: use syncfs() rather than sync()
Quoted 37 lines · Dave Chinner
From: Dave Chinner <[email protected]>
xfs/183 runs
bulkstat_unlink_testto create 100 inodes, unlink on and bulkstat them. It takes a ridiculously long time to run under check-parallel because it runssync()multiple times per iteration:Ten slowest tests - runtime in seconds:
.....
xfs/183 328When running check-parallel,
sync()can take a -long- time to run as there can be dozens of filesystems that need to be synced, not to mention sync getting hung up behind all the mount and unmounts that are also being run.Convert the
sync()calls tosyncfs()so that they only try to sync the filesystem under test and not the entire system. This avoids interactions and delays with other tests and mount/unmount operations, hence allowing both the test and the overall check-parallel operation to run faster:xfs/183 4s
Signed-off-by Dave Chinner <[email protected]>--- src/bulkstat_unlink_test.c | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-)src/bulkstat_unlink_test.c +0
-0@@ -88,7 +88,7 @@ main(int argc, char *argv[]) }if (chknb) {/* Get the original number of inodes (lazy) */
src/bulkstat_unlink_test.c
Quoted 2 lines
src/bulkstat_unlink_test.c +1
-1-sync (); +syncfs (fd[nfiles]);
This looks good to me.
Minor: Is it safe to use int fd[nfiles + 1]; Isn't it better to use malloc()?
Other than this:
Feel free to add
Quoted 52 lines
if (xfsctl(dirname, fd[nfiles],
XFS_IOC_FSBULKSTAT, &a) != 0) {
printf("Warning (%s:%d),xfsctl(XFS_IOC_FSBULKSTAT)FAILED.\n",__FILE__,__LINE__);src/bulkstat_unlink_test.c +1
-1}@@ -118,7 +118,7 @@ main(int argc, char *argv[])*The files are still opened (but unlink()ed) , * we should have more inodes than before */ -sync (); +syncfs (fd[nfiles]); last_inode =0 ;if (xfsctl (dirname, fd[nfiles],XFS_IOC_FSBULKSTAT , &a) !=0 ) {printf("Warning (%s:%d),
xfsctl(XFS_IOC_FSBULKSTAT)FAILED.\n",__FILE__,__LINE__);src/bulkstat_unlink_test.c +1
-1@@ -139,7 +139,7 @@ main(int argc, char *argv[])* The files are now closed, we should be back to our, * previous inode count */ -sync (); +syncfs (fd[nfiles]); last_inode =0 ; if (xfsctl(dirname, fd[nfiles],
XFS_IOC_FSBULKSTAT, &a) != 0) {
printf("Warning (%s:%d),xfsctl(XFS_IOC_FSBULKSTAT)FAILED.\n",__FILE__,__LINE__);src/bulkstat_unlink_test.c +1
-1@@ -150,7 +150,7 @@ main(int argc, char *argv[]) } } -sync (); +syncfs (fd[nfiles]); last_inode =0 ; for (;;) { if ((e = xfsctl(dirname, fd[nfiles],
XFS_IOC_FSBULKSTAT, &a)) < 0) {src/bulkstat_unlink_test.c +2
-2@@ -173,11 +173,11 @@ main(int argc, char *argv[]) } } -close (fd[nfiles]);sprintf (fname,"rm -rf %s\n" , dirname);system (fname); -sync (); +syncfs (fd[nfiles]); +close (fd[nfiles]);sleep (2 );printf ("passed\n" ); }
Re: [PATCH 24/28] open-by-handle.c: use syncfs() rather than sync()
Quoted 36 lines · Nirjhar Roy (IBM)
Quoted 35 lines · Dave Chinner
From: Dave Chinner <[email protected]>
xfs/183 runs
bulkstat_unlink_testto create 100 inodes, unlink on and bulkstat them. It takes a ridiculously long time to run under check-parallel because it runssync()multiple times per iteration:Ten slowest tests - runtime in seconds:
.....
xfs/183 328When running check-parallel,
sync()can take a -long- time to run as there can be dozens of filesystems that need to be synced, not to mention sync getting hung up behind all the mount and unmounts that are also being run.Convert the
sync()calls tosyncfs()so that they only try to sync the filesystem under test and not the entire system. This avoids interactions and delays with other tests and mount/unmount operations, hence allowing both the test and the overall check-parallel operation to run faster:xfs/183 4s
Signed-off-by Dave Chinner <[email protected]>--- src/bulkstat_unlink_test.c | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-)src/bulkstat_unlink_test.c +0
-0@@ -88,7 +88,7 @@ main(int argc, char *argv[]) }
Quoted 6 lines
src/bulkstat_unlink_test.c
Quoted 4 lines
src/bulkstat_unlink_test.c +0
-0if (chknb) {/* Get the original number of inodes (lazy) */
src/bulkstat_unlink_test.c +1
-1-sync (); +syncfs (fd[nfiles]);This looks good to me.
Minor: Is it safe to use int fd[nfiles + 1];
Yes, it is. Variable size array declarations like this have been supported by C and C compilers for a long time.
Quoted 2 lines
Isn't it better to use
malloc()?
See my previous comments about stack usage in userspace test code. :)
-Dave.
Signature · 2 lines
Dave Chinner [email protected]
Re: [PATCH 24/28] open-by-handle.c: use syncfs() rather than sync()
Quoted 43 lines · Dave Chinner
Quoted 36 lines · Nirjhar Roy (IBM)
Quoted 35 lines · Dave Chinner
From: Dave Chinner <[email protected]>
xfs/183 runs
bulkstat_unlink_testto create 100 inodes, unlink on and bulkstat them. It takes a ridiculously long time to run under check-parallel because it runssync()multiple times per iteration:Ten slowest tests - runtime in seconds:
.....
xfs/183 328When running check-parallel,
sync()can take a -long- time to run as there can be dozens of filesystems that need to be synced, not to mention sync getting hung up behind all the mount and unmounts that are also being run.Convert the
sync()calls tosyncfs()so that they only try to sync the filesystem under test and not the entire system. This avoids interactions and delays with other tests and mount/unmount operations, hence allowing both the test and the overall check-parallel operation to run faster:xfs/183 4s
Signed-off-by Dave Chinner <[email protected]>--- src/bulkstat_unlink_test.c | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-)src/bulkstat_unlink_test.c +0
-0@@ -88,7 +88,7 @@ main(int argc, char *argv[]) }Quoted 6 lines
src/bulkstat_unlink_test.c
Quoted 4 lines
src/bulkstat_unlink_test.c +0
-0if (chknb) {/* Get the original number of inodes (lazy) */
src/bulkstat_unlink_test.c +1
-1-sync (); +syncfs (fd[nfiles]);This looks good to me.
Minor: Is it safe to use int fd[nfiles + 1];
Quoted 6 lines
Yes, it is. Variable size array declarations like this have been supported by C and C compilers for a long time.
Quoted 2 lines
Isn't it better to use
malloc()?See my previous comments about stack usage in userspace test code. :)
Yes. I mostly meant in terms of stack overflow issues if "nfiles" is large, but I checked C stack space in the userspace is around 8M so that should be fine. Thanks.
--NR
Quoted 3 lines
-Dave.
Signature · 3 lines
Nirjhar Roy Linux Kernel Developer IBM, Bangalore
PATCH 25/28 bulkstat_unlink_test_modified.c: remove unused test code
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
The built binary is not used by any test. Remove the dead code.
--- .gitignore | 1 - src/Makefile | 2 +- src/bulkstat_unlink_test_modified.c | 193 ---------------------------- 3 files changed, 1 insertion(+), 195 deletions(-) delete mode 100644 src/bulkstat_unlink_test_modified.c
.gitignore +0 -1
@@ -65,7 +65,6 @@ tags /src/btrfs_encoded_write /src/bulkstat_null_ocount /src/bulkstat_unlink_test -/src/bulkstat_unlink_test_modified /src/checkpoint_journal /src/chprojid_fail /src/cloner
src/Makefile +1 -1
@@ -26,7 +26,7 @@ TARGETS = dirstress fill fill2 getpagesize holes lstat64 \ LINUX_TARGETS = xfsctl bstat t_mtab getdevicesize preallo_rw_pattern_reader \ preallo_rw_pattern_writer ftrunc trunc fs_perms testx looptest \ locktest unwritten_mmap bulkstat_unlink_test deduperace \ - bulkstat_unlink_test_modified t_dir_offset t_futimens t_immutable \ + t_dir_offset t_futimens t_immutable \ stale_handle pwrite_mmap_blocked t_dir_offset2 seek_sanity_test \ seek_copy_test t_readdir_1 t_readdir_2 fsync-tester nsexec cloner \ renameat2 t_getcwd e4compact test-nextquota punch-alternating \
src/bulkstat_unlink_test_modified.c deleted +0 -193
193 lines not on this page · raw message
Signature · 1 line
2.45.2
Re: [PATCH 25/28] bulkstat_unlink_test_modified.c: remove unused test code
Quoted 3 lines · Dave Chinner
From: Dave Chinner <[email protected]>
The built binary is not used by any test. Remove the dead code.
Yes.
Feel free to add
Quoted 253 lines
Signed-off-by Dave Chinner <[email protected]>--- .gitignore | 1 - src/Makefile | 2 +- src/bulkstat_unlink_test_modified.c | 193 -------------------------- -- 3 files changed, 1 insertion(+), 195 deletions(-) delete mode 100644 src/bulkstat_unlink_test_modified.c.gitignore +0
-1@@ -65,7 +65,6 @@ tags /src/btrfs_encoded_write /src/bulkstat_null_ocount /src/bulkstat_unlink_test -/src/bulkstat_unlink_test_modified /src/checkpoint_journal /src/chprojid_fail /src/clonersrc/Makefile +0
-1@@ -26,7 +26,7 @@ TARGETS = dirstress fill fill2 getpagesize holes lstat64 \ LINUX_TARGETS = xfsctl bstat t_mtab getdevicesize preallo_rw_pattern_reader \ preallo_rw_pattern_writer ftrunc trunc fs_perms testx looptest \ locktest unwritten_mmap bulkstat_unlink_test deduperace \ - bulkstat_unlink_test_modified t_dir_offset t_futimens
t_immutable\src/Makefile +1
-0+ t_dir_offset t_futimens t_immutable \ stale_handle pwrite_mmap_blocked t_dir_offset2 seek_sanity_test\
seek_copy_testt_readdir_1t_readdir_2fsync-tester nsexec
cloner \src/Makefile +0
-0renameat2 t_getcwd e4compact test-nextquota punch-alternating \src/bulkstat_unlink_test_modified.c deleted +0
-2@@ -1,193 +0,0 @@ -/* -* $Id: bulkstat_unlink_test_modified.c,v 1.1 2007/10/03 16:23:57 mohamedb.longdrop.melbourne.sgi.com Exp $
- * Test bulkstat doesn't returned unlinked inodes.
- * Mark Goodwin <[email protected]> Fri Jul 20 09:13:57 EST 2007
- *
- * This is a modified version ofbulkstat_unlink_test.cto reproduce a specific
- * problem see pv 969192
- */src/bulkstat_unlink_test_modified.c deleted +0
-29-#include <stdlib.h> -#include <sys/types.h> -#include <sys/stat.h> -#include <fcntl.h> -#include <xfs/xfs.h> -#include <unistd.h> -#include <string.h> - -int -main (int argc,char *argv[]) -{ -int e; -int fd =0 ; -int i; -int j; -int k; -int nfiles; -int stride; -struct stat sbuf; -ino_t *inodelist; -__u32 *genlist; -struct xfs_fsop_bulkreq a; -struct xfs_bstat *ret; -int iterations; -char fname[MAXPATHLEN ]; -char *dirname; - -if (argc !=5 ) { - fprintf(stderr, "Usage: %s iterations nfiles stride dir\n",argv[0]);
- fprintf(stderr, "Create dir with nfiles, unlink each stride'th
file, sync, bulkstat\n");src/bulkstat_unlink_test_modified.c deleted +0
-15-exit (1 ); - } - - iterations =atoi (argv[1 ]); - nfiles =atoi (argv[2 ]); - stride =atoi (argv[3 ]); - dirname = argv[4 ]; -if (!nfiles || !iterations) { -fprintf (stderr,"Iterations and nfiles showld be non zero.\n" ); -exit (1 ); - } - - inodelist = (ino_t *)malloc (nfiles *sizeof (ino_t)); - genlist = (__u32 *)malloc (nfiles *sizeof (__u32)); - ret = (struct xfs_bstat *)malloc(nfiles * sizeof(struct
xfs_bstat));src/bulkstat_unlink_test_modified.c deleted +0
-26- -for (k=0 ; k < iterations; k++) { -xfs_ino_t last_inode =0 ; -int count =0 ; -int testFiles =0 ; - -printf ("Iteration %d ... \n" , k); - -memset (inodelist,0 , nfiles *sizeof (ino_t)); -memset (genlist,0 , nfiles *sizeof (__u32)); -memset (ret,0 , nfiles *sizeof (struct xfs_bstat )); -memset (&a,0 ,sizeof (struct xfs_fsop_bulkreq )); - a.lastip = (__u64 *)&last_inode; - a.icount = nfiles; - a.ubuffer = ret; - a.ocount = &count; - -if (mkdir (dirname,0755 ) <0 ) { -perror (dirname); -exit (1 ); - } - -/* create nfiles and store their inode numbers in inodelist */ -for (i=0 ; i < nfiles; i++) { -sprintf (fname,"%s/file%06d" , dirname, i); - if ((fd = open(fname, O_RDWR | O_CREAT | O_TRUNC, 0644)) <0) {
src/bulkstat_unlink_test_modified.c deleted +0
-23-perror (fname); -exit (1 ); - } -write (fd, fname,sizeof (fname)); -if (fstat (fd, &sbuf) <0 ) { -perror (fname); -exit (1 ); - } - inodelist[i] = sbuf.st_ino; -close (fd); - } - -sync (); - -/* collect bs_gen for the nfiles files */ -if ((fd =open (dirname,O_RDONLY )) <0 ) { -perror (dirname); -exit (1 ); - } - - testFiles =0 ; -for (;;) { -if ((e =xfsctl (dirname, fd,XFS_IOC_FSBULKSTAT , &a)) <0 ){
src/bulkstat_unlink_test_modified.c deleted +0
-33-perror ("XFS_IOC_FSBULKSTAT1:" ); -exit (1 ); - } - -if (count ==0 ) -break ; - -for (i=0 ; i < count; i++) { -for (j=0 ; j < nfiles; j += stride) { -if (ret[i].bs_ino == inodelist[j]) { - genlist[j] = ret[i].bs_gen; - testFiles++; - } - } - } - } -close (fd); - -printf ("testFiles %d ... \n" , testFiles); - -/* remove some of the first set of files */ -for (i=0 ; i < nfiles; i += stride) { -sprintf (fname,"%s/file%06d" , dirname, i); -if (unlink (fname) <0 ) { -perror (fname); -exit (1 ); - } - } - -/* create a new set of files (replacing the unlinked ones) */ -for (i=0 ; i < nfiles; i += stride) { -sprintf (fname,"%s/file%06d" , dirname, i); - if ((fd = open(fname, O_RDWR | O_CREAT | O_TRUNC, 0644)) <0) {
src/bulkstat_unlink_test_modified.c deleted +0
-17-perror (fname); -exit (1 ); - } -write (fd, fname,sizeof (fname)); -close (fd); - } - -sync (); - last_inode =0 ; count =0 ; - -if ((fd =open (dirname,O_RDONLY )) <0 ) { -perror (dirname); -exit (1 ); - } - -for (;;) { -if ((e =xfsctl (dirname, fd,XFS_IOC_FSBULKSTAT , &a)) <0 ){
src/bulkstat_unlink_test_modified.c deleted +0
-13-perror ("XFS_IOC_FSBULKSTAT:" ); -exit (1 ); - } - -if (count ==0 ) -break ; - -for (i=0 ; i < count; i++) { -for (j=0 ; j < nfiles; j += stride) { -if ((ret[i].bs_ino == inodelist[j]) && - (ret[i].bs_gen == genlist[j])) { -/* oops, the same inode with old gen number */ - printf("Unlinked inode %llu with generation %d"
src/bulkstat_unlink_test_modified.c deleted +0
-8-"returned by bulkstat\n" , - (unsigned long long )inodelist[j], - genlist[j]); -exit (1 ); - } -if (ret[i].bs_ino == inodelist[j] && - ret[i].bs_gen != genlist[j] +1 ) { -/* oops, the new gen number is not 1 bigger than the old */
src/bulkstat_unlink_test_modified.c deleted +0
-1- printf("Inode with old generation %d, newgeneration %d\n",
src/bulkstat_unlink_test_modified.c deleted +0
-19- genlist[j], ret[i].bs_gen); -exit (1 ); - } - } - } - } - -close (fd); - -sprintf (fname,"rm -rf %s\n" , dirname); -system (fname); - -sync (); -sleep (2 ); -printf ("passed\n" ); - } - -exit (0 ); -}
PATCH 26/28 stale-handle.c: use syncfs() rather than sync()
In reply to Dave Chinnerxfs/238 runs stale_handle to create 1000 inodes, grab their file handles, unlink them and then try to open them from the stored file handles. It takes a ridiculously long time to run under check-parallel because it runs sync() multiple times:
xfs/238 144s
When running check-parallel, sync() can take a -long- time to run as there can be dozens of filesystems that need to be synced, not to mention sync getting hung up behind all the mount and unmounts that are also being run.
Convert the sync() calls to syncfs() so that they only try to sync the filesystem under test and not the entire system. This avoids interactions and delays with other tests and mount/unmount operations, hence allowing both the test and the overall check-parallel operation to run faster:
xfs/238 5s
--- src/stale_handle.c | 15 +++++++++++---- 1 file changed, 11 insertions(+), 4 deletions(-)
src/stale_handle.c +11 -4
@@ -21,7 +21,7 @@int main (int argc,char **argv) {int i; -int fd; +int fd, test_dir_fd;int ret;int failed =0 ;char fname[MAXPATHLEN ];
@@ -44,6 +44,12 @@ int main(int argc, char **argv)return EXIT_FAILURE ; } + test_dir_fd =open (test_dir,O_RDONLY |O_DIRECTORY ); +if (test_dir_fd <0 ) { +perror (test_dir); +return EXIT_FAILURE ; + } + ret =path_to_fshandle (test_dir, (void **)fshandle, &fshlen);if (ret <0 ) {perror ("path_to_fshandle" );
@@ -66,7 +72,7 @@ int main(int argc, char **argv) }/* sync to get the new inodes to hit the disk */ -sync (); +syncfs (test_dir_fd);/* create the handles */ for (i=0 ; i <NUMFILES ; i++) {
@@ -89,10 +95,10 @@ int main(int argc, char **argv) }/* sync to get log forced for unlink transactions to hit the disk */ -sync (); +syncfs (test_dir_fd);/* sync once more FTW */ -sync (); +syncfs (test_dir_fd);/* * now drop the caches so that unlinked inodes are reclaimed and
@@ -120,6 +126,7 @@ int main(int argc, char **argv)free_handle (handle[i], hlen[i]); failed++; } +close (test_dir_fd);if (failed)return EXIT_FAILURE ;return EXIT_SUCCESS ;
Signature · 1 line
2.45.2
Re: [PATCH 26/28] stale-handle.c: use syncfs() rather than sync()
Quoted 12 lines · Dave Chinner
xfs/238 runs
stale_handleto create 1000 inodes, grab their file handles, unlink them and then try to open them from the stored file handles. It takes a ridiculously long time to run under check-parallel because it runssync()multiple times:xfs/238 144s
When running check-parallel,
sync()can take a -long- time to run as there can be dozens of filesystems that need to be synced, not to mention sync getting hung up behind all the mount and unmounts that are also being run.
Quoted 5 lines
Convert the
sync()calls tosyncfs()so that they only try to sync the filesystem under test and not the entire system. This avoids interactions and delays with other tests and mount/unmount operations, hence allowing both the test and the overall check-parallel operation to run faster:
Yes, filesystem wise sync is makes more sense. I remember similar changes with your previous check-parralel patch series as well.
Quoted 13 lines
xfs/238 5s
Signed-off-by Dave Chinner <[email protected]>--- src/stale_handle.c | 15 +++++++++++---- 1 file changed, 11 insertions(+), 4 deletions(-)src/stale_handle.c +0
-0
src/stale_handle.c line 27
Quoted 8 lines
src/stale_handle.c +1
-1@@ -21,7 +21,7 @@int main (int argc,char **argv) {int i; -int fd; +int fd, test_dir_fd;int ret;int failed =0 ;char fname[MAXPATHLEN ];
Not related to this change:
A lot of local variables are being used in this test:
char fname[MAXPATHLEN];
void *handle[NUMFILES];size_t hlen[NUMFILES];
char fshandle[256];
Is this fine? Shouldn't we limit the usage of so much local stack?
src/stale_handle.c int main(int argc, char **argv) · line 52
Quoted 10 lines
src/stale_handle.c +6
-0@@ -44,6 +44,12 @@ int main(int argc, char **argv)return EXIT_FAILURE ; } + test_dir_fd =open (test_dir,O_RDONLY |O_DIRECTORY ); +if (test_dir_fd <0 ) { +perror (test_dir); +return EXIT_FAILURE ; + } +
Now that we are already opening the test_dir to get an fd of the filessytem being tested, should we remove the stat call(line 42) just before this open call and instead use this open call to check the presence/absence of test_dir?
Other than this, the rest of the changes look good to me (some minor comments below but not related to this change).
Feel free to add
ret = path_to_fshandle(test_dir, (void **)fshandle, &fshlen);
src/stale_handle.c int main(int argc, char **argv) · line 78
Quoted 11 lines
src/stale_handle.c +1
-1if (ret <0 ) {perror ("path_to_fshandle" );@@ -66,7 +72,7 @@ int main(int argc, char **argv) }/* sync to get the new inodes to hit the disk */ -sync (); +syncfs (test_dir_fd);/* create the handles */ for (i=0 ; i <NUMFILES ; i++) {
Not related to this change and minor:
Shouldn't we replace usage of sprintf with snprintf()s? --NR
Quoted 23 lines
src/stale_handle.c +0
-0@@ -89,10 +95,10 @@ int main(int argc, char **argv) }/* sync to get log forced for unlink transactions to hit the disk */
src/stale_handle.c +2
-2-sync (); +syncfs (test_dir_fd);/* sync once more FTW */ -sync (); +syncfs (test_dir_fd);/* * now drop the caches so that unlinked inodes are reclaimed and
src/stale_handle.c +1
-0@@ -120,6 +126,7 @@ int main(int argc, char **argv)free_handle (handle[i], hlen[i]); failed++; } +close (test_dir_fd);if (failed)return EXIT_FAILURE ;return EXIT_SUCCESS ;
Re: [PATCH 26/28] stale-handle.c: use syncfs() rather than sync()
Quoted 40 lines · Nirjhar Roy (IBM)
Quoted 12 lines · Dave Chinner
xfs/238 runs
stale_handleto create 1000 inodes, grab their file handles, unlink them and then try to open them from the stored file handles. It takes a ridiculously long time to run under check-parallel because it runssync()multiple times:xfs/238 144s
When running check-parallel,
sync()can take a -long- time to run as there can be dozens of filesystems that need to be synced, not to mention sync getting hung up behind all the mount and unmounts that are also being run.Quoted 5 lines
Convert the
sync()calls tosyncfs()so that they only try to sync the filesystem under test and not the entire system. This avoids interactions and delays with other tests and mount/unmount operations, hence allowing both the test and the overall check-parallel operation to run faster:Yes, filesystem wise sync is makes more sense. I remember similar changes with your previous check-parralel patch series as well.
Quoted 20 lines
xfs/238 5s
Signed-off-by Dave Chinner <[email protected]>--- src/stale_handle.c | 15 +++++++++++---- 1 file changed, 11 insertions(+), 4 deletions(-)src/stale_handle.c +1
-1@@ -21,7 +21,7 @@int main (int argc,char **argv) {int i; -int fd; +int fd, test_dir_fd;int ret;int failed =0 ;
Quoted 8 lines
src/stale_handle.c line 27
Quoted 1 line
src/stale_handle.c +0
-0char fname[MAXPATHLEN ];Not related to this change:
A lot of local variables are being used in this test:
char fname[MAXPATHLEN];
void *handle[NUMFILES];size_thlen[NUMFILES];
char fshandle[256];
Is this fine? Shouldn't we limit the usage of so much local stack?
It's userspace and it's not a big/complex program. Using stack like this instead of allocating memory is simple, fast, and easy to maintain. There's no reason to complicate the code unnecessarily.
Quoted 4 lines
Quoted 4 lines
src/stale_handle.c +0
-0@@ -44,6 +44,12 @@ int main(int argc, char **argv)return EXIT_FAILURE ; }
Quoted 10 lines
src/stale_handle.c int main(int argc, char **argv) · line 52
Quoted 6 lines
src/stale_handle.c +6
-0+ test_dir_fd =open (test_dir,O_RDONLY |O_DIRECTORY ); +if (test_dir_fd <0 ) { +perror (test_dir); +return EXIT_FAILURE ; + } +Now that we are already opening the
test_dirto get an fd of the filessytem being tested, should we remove the stat call(line 42) just before this open call and instead use this open call to check the presence/absence oftest_dir?
We can, it doesn't matter either way...
Quoted 5 lines
Other than this, the rest of the changes look good to me (some minor comments below but not related to this change).
Feel free to add
Reviewed-by Nirjhar Roy (IBM) <[email protected]>
Thanks!
-Dave.
Signature · 2 lines
Dave Chinner [email protected]
PATCH 27/28 scaleread: remove dead test code
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
scaleread.{c,sh} is a one-off test case for a "will-it-scale" page cache read workload from NASA back in 2003. This has not been directly exercised by fstests since it was added 20 years ago. The scaleread.c source code isn't even built by src/Makefile, so it is definitely stale, dead code. Remove it.
--- src/Makefile | 2 +- src/scaleread.c | 224 ----------------------------------------------- src/scaleread.sh | 64 -------------- 3 files changed, 1 insertion(+), 289 deletions(-) delete mode 100644 src/scaleread.c delete mode 100644 src/scaleread.sh
src/Makefile +1 -1
@@ -37,7 +37,7 @@ LINUX_TARGETS = xfsctl bstat t_mtab getdevicesize preallo_rw_pattern_reader \ detached_mounts_propagation ext4_resize t_readdir_3 splice2pipe \ uuid_ioctl t_snapshot_deleted_subvolume fiemap-fault min_dio_alignment -EXTRA_EXECS = dmerror fill2attr fill2fs fill2fs_check scaleread.sh \ +EXTRA_EXECS = dmerror fill2attr fill2fs fill2fs_check \ btrfs_crc32c_forged_name.py popdir.pl popattr.py \ soak_duration.awk parse-dev-tree.awk parse-extent-tree.awk
src/scaleread.c deleted +0 -224
224 lines not on this page · raw message
src/scaleread.sh deleted +0 -64
64 lines not on this page · raw message
Signature · 1 line
2.45.2
Re: [PATCH 27/28] scaleread: remove dead test code
Quoted 7 lines · Dave Chinner
From: Dave Chinner <[email protected]>
scaleread.{c,sh} is a one-off test case for a "will-it-scale" page cache read workload from NASA back in 2003. This has not been directly exercised by fstests since it was added 20 years ago. The
scaleread.csource code isn't even built by src/Makefile, so it is definitely stale, dead code. Remove it.
I did a grep for "scaleread" and yes, it seems that the file scaleread.sh only gets installed/copied into $(PKG_LIB_DIR)/src, however it doesn't get invoked and called from any call site. Also, the binary "scaleread" is only referenced/invoked from scaleread.sh but again, neither scaleread is compiled in the Makefile nor scaleread.sh gets invoked from anywhere. So this change makes sense to me.
Feel free to add
Quoted 333 lines
Signed-off-by Dave Chinner <[email protected]>--- src/Makefile | 2 +- src/scaleread.c | 224 ------------------------------------------- ---- src/scaleread.sh | 64 -------------- 3 files changed, 1 insertion(+), 289 deletions(-) delete mode 100644 src/scaleread.c delete mode 100644 src/scaleread.shsrc/Makefile +0
-0@@ -37,7 +37,7 @@ LINUX_TARGETS = xfsctl bstat t_mtab getdevicesize preallo_rw_pattern_reader \ detached_mounts_propagation ext4_resize t_readdir_3 splice2pipe \ uuid_ioctl t_snapshot_deleted_subvolume fiemap-fault
min_dio_alignmentsrc/Makefile +1
-1-EXTRA_EXECS = dmerror fill2attr fill2fs fill2fs_check scaleread.sh \ +EXTRA_EXECS = dmerror fill2attr fill2fs fill2fs_check \ btrfs_crc32c_forged_name.py popdir.pl popattr.py \ soak_duration.awk parse-dev-tree.awk parse-extent-tree.awk
src/scaleread.c deleted +0
-132@@ -1,224 +0,0 @@ -// SPDX-License-Identifier: GPL-2.0 -/* -* Copyright (c) 2003-2004 Silicon Graphics, Inc. -* All Rights Reserved. -*/ -/* -* Test scaling of multiple processes opening/reading -* a number of small files simultaneously. -* - create <f> files -* - fork <n> processes -* - wait for all processes ready -* - start all proceses at the same time -* - each processes opens , read, closes each file -* - option to resync each process at each file -* -* test [-c cpus] [-b bytes] [-f files] [-v] [-s] [-S] -* OR -* test -i [-b bytes] [-f files] -*/ -#include <unistd.h> -#include <string.h> -#include <stdio.h> -#include <stdlib.h> -#include <sys/types.h> -#include <sys/stat.h> -#include <sys/wait.h> -#include <fcntl.h> -#include <stdlib.h> -#include <sys/ipc.h> -#include <sys/shm.h> - -void do_initfiles (void ); -void slave (int ); - -#define VPRINT(x...) do { if(verbose) fprintf(x);} while(0) -#define perrorx (s) do {perror(s); exit(1);} while (0) - -long bytes=8192 ; -int cpus=1 ; -int init=0 ; -int strided=0 ; -int files=1 ; -int blksize=512 ; -int syncstep=0 ; -int verbose=0 ; - -typedef struct { -volatile long go; -long fill[15 ]; -volatile long rdy[512 ]; -}share_t ; - -share_t *sharep; - - -int -runon (int cpu) -{ -#ifdef sys_sched_setaffinity -unsigned long mask[8 ]; - -if (cpu <0 || cpu >=512 ) -return -1 ; -memset (mask,0 ,sizeof (mask)); - mask[cpu/64 ] |=1UL <<(cpu&63 ); - -if (syscall (sys_sched_setaffinity,0 ,sizeof (mask), mask)) -return -1 ; -#endif -return 0 ; -} - -long -scaled_atol (char *p) -{ -long val; -char *pe; - - val =strtol (p, &pe,0 ); -if (*pe =='K' || *pe =='k' ) - val *=1024L ; -else if (*pe =='M' || *pe =='m' ) - val *=1024L *1024L ; -else if (*pe =='G' || *pe =='g' ) - val *=1024L *1024L *1024L ; -else if (*pe =='p' || *pe =='P' ) - val *=getpagesize (); -return val; -} - - -int -main(int argc, char** argv) { - int shmid; - static char optstr[] = "c:b:f:sSivH"; - int notdone, stat, i, j, c, er=0; - - opterr=1; - while ((c = getopt(argc, argv, optstr)) != EOF) - switch (c) { - case 'c': - cpus = atoi(optarg); - break; - case 'b': - bytes = scaled_atol(optarg); - break; - case 'f': - files = atoi(optarg); - break; - case 'i': - init++; - break; - case 's': - syncstep++; - break; - case 'S': - strided++; - break; - case 'v': - verbose++; - break; - case '?': - er = 1; - break; - } - if (er) { - printf("usage: %s %s\n", argv[0], optstr); - exit(1); - } - - - if ((shmid = shmget(IPC_PRIVATE, sizeof (share_t),IPC_CREAT|SHM_R|SHM_W)) == -1)
src/scaleread.c deleted +0
-54-perrorx ("shmget failed" ); - sharep = (share_t *)shmat (shmid, (void *)0 ,SHM_R |SHM_W ); -memset (sharep,-1 ,sizeof (share_t)); - -if (init) { -do_initfiles (); -exit (0 ); - } -for (i=0 ; i<cpus; i++) { -if (fork () ==0 ) -slave (i); - } - -for (i=0 ; i<files; i++) { -VPRINT (stderr,"%d:" , i); - notdone = cpus; -do { -for (j=0 ; j<cpus; j++) { -if (sharep->rdy[j] == i) { - sharep->rdy[j] =-1 ; -VPRINT (stderr," %d" , j); - notdone--; - } - } - }while (notdone); -VPRINT (stderr,"\n" ); - sharep->go = i; -if (!syncstep) -break ; - } -VPRINT (stderr,"\n" ); - -while (wait (&stat)>0 ) -VPRINT (stderr,"." ); -VPRINT (stderr,"\n" ); - -exit (0 ); -} - -void -slave (int id) -{ -int i, fd, byte; -char *buf, filename[32 ]; - -runon (id+1 ); - buf =malloc (blksize); -bzero (buf, blksize); -for (i=0 ; i<files; i++) { -if (!i || syncstep) { - sharep->rdy[id] = i; -while (sharep->go != i); - } - sprintf(filename, "/tmp/tst.%d", (strided ? ((i + id) %files) : i));
src/scaleread.c deleted +0
-38-if ((fd =open (filename,O_RDONLY )) <0 ) { -perrorx (filename); - } - -for (byte=0 ; byte<bytes; byte+=blksize) { -if (read (fd, buf, blksize) != blksize) -perrorx ("read of file failed" ); - } -close (fd); - } -exit (0 ); -} - -void -do_initfiles (void ) -{ -int i, fd, byte; -char *buf, filename[32 ]; - - buf =malloc (blksize); -bzero (buf, blksize); - -for (i=0 ; i<files; i++) { -sprintf (filename,"/tmp/tst.%d" , i); -unlink (filename); -if ((fd =open (filename,O_RDWR |O_CREAT ,0644 )) <0 ) -perrorx (filename); - -for (byte=0 ; byte<bytes; byte+=blksize) { -if (write (fd, buf, blksize) != blksize) -perrorx ("write of file failed" ); - } -close (fd); - } -sync (); -} - -src/scaleread.sh deleted +0
-3@@ -1,64 +0,0 @@ -#!/bin/sh -# -# Copyright (c) 2003-2004 Silicon Graphics, Inc. All RightsReserved.
src/scaleread.sh deleted +0
-57-# - -help() { -cat <<END -Measure scaling of multiple cpus readin the same set of files. -(NASA testcase). - Usage: $0 [-b <bytes>] [-f <files>] [-s] [-B] [-v] cpus ... - or - $0 -i [-b <bytes>] [-f <files>] - - -b file size in bytes - -f number of files - -s keep processes synchronized when reading files - -B use bcfree to free buffer cache pages before each run -END -exit 1 -} - -err () { - echo "ERROR - $*" - exit 1 -} - -BYTES=8192 -FILES=10 -SYNC="" -VERBOSE="" -STRIDED="" -BCFREE=0 -INIT=0 -OPTS="f:b:vsiSBH" -while getopts "$OPTS" c ; do - case $c in - H) help;; - f) FILES=${OPTARG};; - b) BYTES=${OPTARG};; - i) INIT=1;; - B) BCFREE=1;; - S) STRIDED="-S";; - s) SYNC="-s";; - v) VERBOSE="-v";; - \?) help;; - esac - -done -shift `expr $OPTIND - 1` - -if [ $INIT -gt 0 ] ; then - echo "Initializing $BYTES bytes, $FILES files" - ./scaleread $VERBOSE -i -b $BYTES -f $FILES - sync -else - [ $# -gt 0 ] || help - echo "Testing $BYTES bytes, $FILES files" - for CPUS in $* ; do - [ $BCFREE -eq 0 ] || bcfree -a - /usr/bin/time -f "$CPUS: %e wall, %S sys, %Uuser" ./scaleread \
src/scaleread.sh deleted +0
-1- $SYNC $STRIDED $VERBOSE -b $BYTES -f $FILES -c$CPUS
src/scaleread.sh deleted +0
-3- done -fi -
PATCH 28/28 xfs/259: no need to call sync
In reply to Dave Chinnerxfs/259 runs sync every time through it's loop.
It takes a ridiculously long time to run under
check-parallel:
xfs/259 461s
When running check-parallel, sync can take a -long- time to run as there can be dozens of filesystems that need to be synced, not to mention sync getting hung up behind all the mount and unmounts that are also being run.
sync is used at the end of the loop before destroying the loop device, but the contents of the loop device is completely discarded at the start of the next loop. i.e. the image file is unlinked and recreated. Hence the sync call does nothing useful and only slows down the test. Removing it runs the test much faster:
xfs/259 23s
--- tests/xfs/259 | 1 - 1 file changed, 1 deletion(-)
tests/xfs/259 +0 -1
@@ -52,7 +52,6 @@ for del in $sizes_to_check; do $MKFS_XFS_PROG -l size=32m -b size=$bs $loop_dev | _filter_mkfs \ >/dev/null 2> $tmp.mkfs || echo "mkfs failed!" . $tmp.mkfs - sync _destroy_loop_device $loop_dev unset loop_dev done
Signature · 1 line
2.45.2
Re: [PATCH 28/28] xfs/259: no need to call sync
Quoted 11 lines · Dave Chinner
xfs/259 runs sync every time through it's loop.
It takes a ridiculously long time to run under
check-parallel:xfs/259 461s
When running check-parallel, sync can take a -long- time to run as there can be dozens of filesystems that need to be synced, not to mention sync getting hung up behind all the mount and unmounts that are also being run.
Quoted 7 lines
sync is used at the end of the loop before destroying the loop device, but the contents of the loop device is completely discarded at the start of the next loop. i.e. the image file is unlinked and recreated. Hence the sync call does nothing useful and only slows down the test. Removing it runs the test much faster:
xfs/259 23s
Yes. The sync doesn't do anything useful since the loop device is destroyed and the underlying loop device image file is also re-created at the beginning of the loop using dd.
This looks good to me.
Feel free to add
Quoted 19 lines
Signed-off-by Dave Chinner <[email protected]>--- tests/xfs/259 | 1 - 1 file changed, 1 deletion(-)tests/xfs/259 +0
-1@@ -52,7 +52,6 @@ for del in $sizes_to_check; do $MKFS_XFS_PROG -l size=32m -b size=$bs $loop_dev | _filter_mkfs \ >/dev/null 2> $tmp.mkfs || echo "mkfs failed!" . $tmp.mkfs - sync _destroy_loop_device $loop_dev unset loop_devdone