xfstests: add a option to run xfstests under a cgroup
PATCH v2 1/1 xfstests: add a option to run xfstests under a cgroup
I want to add some extended statistic gathering for xfstests, but it's tricky to isolate xfstests from the rest of the host applications. The most straightforward way to do this is to run every test inside of it's own cgroup. From there we can monitor the activity of tasks in the specific cgroup using BPF.
The support for this is pretty simple, allow users to pass -C <cgroup name>. We will create the path if it doesn't already exist, and validate we can add things to cgroup.procs. If we cannot it'll be disabled, otherwise we will use this when we do _run_seq by echo'ing the bash pid into cgroup.procs, which will cause any children to run under that cgroup.
---v1->v2:
- Changed it from a local.config option to a command line option.
- Export CGROUP2_PATH for everything, utilize that path when generating our cgroup for the scripts to run in.
check | 24 +++++++++++++++++++++++- common/cgroup2 | 2 -- common/config | 1 + 3 files changed, 24 insertions(+), 3 deletions(-)
check +23 -1
@@ -72,6 +72,7 @@ check options --large-fs optimise scratch device for large filesystems -s section run only specified section from config file -S section exclude the specified section from the config file + -C cgroup_name run all the tests in the specified cgroup name testlist options -g group[,group...] include tests from these groups
@@ -101,6 +102,10 @@ 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>. +cgroup_name is just a plain name, or a path relative to the root cgroup path. +If CGROUP2_PATH does not point at where cgroup2 is mounted then adjust it +accordingly. + examples: check xfs/001 check -g quick
@@ -307,6 +312,7 @@ while [ $# -gt 0 ]; do ;; --large-fs) export LARGE_SCRATCH_DEV=yes ;; --extra-space=*) export SCRATCH_DEV_EMPTY_SPACE=${r#*=} ;; + -C) CGROUP=$2 ; shift ;; -*) usage ;; *) # not an argument, we've got tests now.
@@ -509,11 +515,24 @@ _expunge_test() OOM_SCORE_ADJ="/proc/self/oom_score_adj" test -w ${OOM_SCORE_ADJ} && echo -1000 > ${OOM_SCORE_ADJ} +# Initialize the cgroup path if it doesn't already exist +if [ ! -z "$CGROUP" ]; then + CGROUP=${CGROUP2_PATH}/${CGROUP} + mkdir -p ${CGROUP} + + # If we can't write to cgroup.procs then unset cgroup + test -w ${CGROUP}/cgroup.procs || unset CGROUP +fi + # ...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 # test framework. _run_seq() { - bash -c "test -w ${OOM_SCORE_ADJ} && echo 250 > ${OOM_SCORE_ADJ}; exec ./$seq" + _extra="test -w ${OOM_SCORE_ADJ} && echo 250 > ${OOM_SCORE_ADJ};" + if [ ! -z "$CGROUP" ]; then + _extra+="echo $$ > ${CGROUP}/cgroup.procs;" + fi + bash -c "${_extra} exec ./$seq" } _detect_kmemleak
@@ -615,6 +634,9 @@ for section in $HOST_OPTIONS_SECTIONS; do echo "MKFS_OPTIONS -- `_scratch_mkfs_options`" echo "MOUNT_OPTIONS -- `_scratch_mount_options`" fi + if [ ! -z "$CGROUP" ]; then + echo "CGROUP -- ${CGROUP}" + fi echo needwrap=true
common/cgroup2 +0 -2
@@ -1,7 +1,5 @@ # cgroup2 specific common functions -export CGROUP2_PATH="${CGROUP2_PATH:-/sys/fs/cgroup}" - _require_cgroup2() { if [ `findmnt -d backward -n -o FSTYPE -f ${CGROUP2_PATH}` != "cgroup2" ]; then
common/config +1 -0
@@ -259,6 +259,7 @@ case "$HOSTOS" in export E2FSCK_PROG=$(type -P e2fsck) export TUNE2FS_PROG=$(type -P tune2fs) export FSCK_OVERLAY_PROG=$(type -P fsck.overlay) + export CGROUP2_PATH="${CGROUP2_PATH:-/sys/fs/cgroup}" ;; esac
Signature · 1 line
2.24.1
Quoted 59 lines · Josef Bacik
I want to add some extended statistic gathering for xfstests, but it's tricky to isolate xfstests from the rest of the host applications. The most straightforward way to do this is to run every test inside of it's own cgroup. From there we can monitor the activity of tasks in the specific cgroup using BPF.
The support for this is pretty simple, allow users to pass -C <cgroup name>. We will create the path if it doesn't already exist, and validate we can add things to cgroup.procs. If we cannot it'll be disabled, otherwise we will use this when we do
_run_seqby echo'ing the bash pid into cgroup.procs, which will cause any children to run under that cgroup.Signed-off-by Josef Bacik <[email protected]>---
v1->v2:
- Changed it from a local.config option to a command line option.
- ExportCGROUP2_PATHfor everything, utilize that path when generating our cgroup for the scripts to run in.check | 24 +++++++++++++++++++++++- common/cgroup2 | 2 -- common/config | 1 + 3 files changed, 24 insertions(+), 3 deletions(-)check +6
-0@@ -72,6 +72,7 @@ check options --large-fs optimise scratch device for large filesystems -s section run only specified section from config file -S section exclude the specified section from the config file + -C cgroup_name run all the tests in the specified cgroup name testlist options -g group[,group...] include tests from these groups@@ -101,6 +102,10 @@ 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>. +cgroup_name is just a plain name, or a path relative to the root cgroup path. +If CGROUP2_PATH does not point at where cgroup2 is mounted then adjust it +accordingly. + examples: check xfs/001 check -g quick@@ -307,6 +312,7 @@ while [ $# -gt 0 ]; do ;; --large-fs) export LARGE_SCRATCH_DEV=yes ;; --extra-space=*) export SCRATCH_DEV_EMPTY_SPACE=${r#*=} ;; + -C) CGROUP=$2 ; shift ;; -*) usage ;; *) # not an argument, we've got tests now.@@ -509,11 +515,24 @@ _expunge_test() OOM_SCORE_ADJ="/proc/self/oom_score_adj" test -w ${OOM_SCORE_ADJ} && echo -1000 > ${OOM_SCORE_ADJ}
check _expunge_test() · line 526
Quoted 10 lines
check +9
-0+# Initialize the cgroup path if it doesn't already exist +if [ ! -z "$CGROUP" ]; then + CGROUP=${CGROUP2_PATH}/${CGROUP} + mkdir -p ${CGROUP} + + # If we can't write to cgroup.procs then unset cgroup + test -w ${CGROUP}/cgroup.procs || unset CGROUP +fi +
Do we need to fix up generic/563 to use this new $CGROUP value, when set? That test explicitly moves tasks to new/temporary groups, but looks like it resets back to the top-level CGROUP2_PATH group. I'm not sure how much that really matters since presumably the next test should move back into $CGROUP. Otherwise this looks reasonable enough to me.
Brian
Quoted 50 lines
check +8
-1# ...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 # test framework. _run_seq() { - bash -c "test -w ${OOM_SCORE_ADJ} && echo 250 > ${OOM_SCORE_ADJ}; exec ./$seq" + _extra="test -w ${OOM_SCORE_ADJ} && echo 250 > ${OOM_SCORE_ADJ};" + if [ ! -z "$CGROUP" ]; then + _extra+="echo $$ > ${CGROUP}/cgroup.procs;" + fi + bash -c "${_extra} exec ./$seq" } _detect_kmemleak@@ -615,6 +634,9 @@ for section in $HOST_OPTIONS_SECTIONS; do echo "MKFS_OPTIONS -- `_scratch_mkfs_options`" echo "MOUNT_OPTIONS -- `_scratch_mount_options`" fi + if [ ! -z "$CGROUP" ]; then + echo "CGROUP -- ${CGROUP}" + fi echo needwrap=truecommon/cgroup2 +0
-2@@ -1,7 +1,5 @@ # cgroup2 specific common functions -export CGROUP2_PATH="${CGROUP2_PATH:-/sys/fs/cgroup}" - _require_cgroup2() { if [ `findmnt -d backward -n -o FSTYPE -f ${CGROUP2_PATH}` != "cgroup2" ]; thencommon/config +1
-0@@ -259,6 +259,7 @@ case "$HOSTOS" in export E2FSCK_PROG=$(type -P e2fsck) export TUNE2FS_PROG=$(type -P tune2fs) export FSCK_OVERLAY_PROG=$(type -P fsck.overlay) + export CGROUP2_PATH="${CGROUP2_PATH:-/sys/fs/cgroup}" ;; esacSignature · 1 line
2.24.1
Quoted 61 lines · Brian Foster
Quoted 60 lines · Josef Bacik
I want to add some extended statistic gathering for xfstests, but it's tricky to isolate xfstests from the rest of the host applications. The most straightforward way to do this is to run every test inside of it's own cgroup. From there we can monitor the activity of tasks in the specific cgroup using BPF.
The support for this is pretty simple, allow users to pass -C <cgroup name>. We will create the path if it doesn't already exist, and validate we can add things to cgroup.procs. If we cannot it'll be disabled, otherwise we will use this when we do
_run_seqby echo'ing the bash pid into cgroup.procs, which will cause any children to run under that cgroup.Signed-off-by Josef Bacik <[email protected]>---
v1->v2:
- Changed it from a local.config option to a command line option.
- ExportCGROUP2_PATHfor everything, utilize that path when generating our cgroup for the scripts to run in.check | 24 +++++++++++++++++++++++- common/cgroup2 | 2 -- common/config | 1 + 3 files changed, 24 insertions(+), 3 deletions(-)check +6
-0@@ -72,6 +72,7 @@ check options --large-fs optimise scratch device for large filesystems -s section run only specified section from config file -S section exclude the specified section from the config file + -C cgroup_name run all the tests in the specified cgroup name testlist options -g group[,group...] include tests from these groups@@ -101,6 +102,10 @@ 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>. +cgroup_name is just a plain name, or a path relative to the root cgroup path. +If CGROUP2_PATH does not point at where cgroup2 is mounted then adjust it +accordingly. + examples: check xfs/001 check -g quick@@ -307,6 +312,7 @@ while [ $# -gt 0 ]; do ;; --large-fs) export LARGE_SCRATCH_DEV=yes ;; --extra-space=*) export SCRATCH_DEV_EMPTY_SPACE=${r#*=} ;; + -C) CGROUP=$2 ; shift ;; -*) usage ;; *) # not an argument, we've got tests now.@@ -509,11 +515,24 @@ _expunge_test() OOM_SCORE_ADJ="/proc/self/oom_score_adj" test -w ${OOM_SCORE_ADJ} && echo -1000 > ${OOM_SCORE_ADJ}
Quoted 16 lines
check _expunge_test() · line 526
Quoted 9 lines
check +9
-0+# Initialize the cgroup path if it doesn't already exist +if [ ! -z "$CGROUP" ]; then + CGROUP=${CGROUP2_PATH}/${CGROUP} + mkdir -p ${CGROUP} + + # If we can't write to cgroup.procs then unset cgroup + test -w ${CGROUP}/cgroup.procs || unset CGROUP +fi +Do we need to fix up generic/563 to use this new $CGROUP value, when set? That test explicitly moves tasks to new/temporary groups, but looks like it resets back to the top-level
CGROUP2_PATHgroup. I'm not sure how much that really matters since presumably the next test should move back into $CGROUP. Otherwise this looks reasonable enough to me.
Wtf I had an explanation for this in my commit message, I must have exited without saving.
I tried to do this actually, and it ended up spewing because you cannot change cgroup.subtree_control for a cgroup you are currently inside of. We can fix this by just making sure all controllers are enabled in our test cgroup, and then fix 562 to not bother messing subtree and just assume everything is enabled. But since it's just one test and I wasn't sure if it would create more wonkiness I just left it as is. I'm open to either solution, I just opted for the less change route. Thanks,
Josef
Quoted 78 lines · Josef Bacik
Quoted 64 lines · Brian Foster
Quoted 63 lines · Josef Bacik
I want to add some extended statistic gathering for xfstests, but it's tricky to isolate xfstests from the rest of the host applications. The most straightforward way to do this is to run every test inside of it's own cgroup. From there we can monitor the activity of tasks in the specific cgroup using BPF.
The support for this is pretty simple, allow users to pass -C <cgroup name>. We will create the path if it doesn't already exist, and validate we can add things to cgroup.procs. If we cannot it'll be disabled, otherwise we will use this when we do
_run_seqby echo'ing the bash pid into cgroup.procs, which will cause any children to run under that cgroup.Signed-off-by Josef Bacik <[email protected]>---
v1->v2:
- Changed it from a local.config option to a command line option.
- ExportCGROUP2_PATHfor everything, utilize that path when generating our cgroup for the scripts to run in.check | 24 +++++++++++++++++++++++- common/cgroup2 | 2 -- common/config | 1 + 3 files changed, 24 insertions(+), 3 deletions(-)check +13
-0@@ -72,6 +72,7 @@ check options --large-fs optimise scratch device for large filesystems -s section run only specified section from config file -S section exclude the specified section from the config file + -C cgroup_name run all the tests in the specified cgroup name testlist options -g group[,group...] include tests from these groups@@ -101,6 +102,10 @@ 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>. +cgroup_name is just a plain name, or a path relative to the root cgroup path. +If CGROUP2_PATH does not point at where cgroup2 is mounted then adjust it +accordingly. + examples: check xfs/001 check -g quick@@ -307,6 +312,7 @@ while [ $# -gt 0 ]; do ;; --large-fs) export LARGE_SCRATCH_DEV=yes ;; --extra-space=*) export SCRATCH_DEV_EMPTY_SPACE=${r#*=} ;; + -C) CGROUP=$2 ; shift ;; -*) usage ;; *) # not an argument, we've got tests now.@@ -509,11 +515,24 @@ _expunge_test() OOM_SCORE_ADJ="/proc/self/oom_score_adj" test -w ${OOM_SCORE_ADJ} && echo -1000 > ${OOM_SCORE_ADJ} +# Initialize the cgroup path if it doesn't already exist +if [ ! -z "$CGROUP" ]; then + CGROUP=${CGROUP2_PATH}/${CGROUP} + mkdir -p ${CGROUP} + + # If we can't write to cgroup.procs then unset cgroup + test -w ${CGROUP}/cgroup.procs || unset CGROUPQuoted 9 lines
check _expunge_test() · line 526
Quoted 2 lines
check +2
-0+fi +Do we need to fix up generic/563 to use this new $CGROUP value, when set? That test explicitly moves tasks to new/temporary groups, but looks like it resets back to the top-level
CGROUP2_PATHgroup. I'm not sure how much that really matters since presumably the next test should move back into $CGROUP. Otherwise this looks reasonable enough to me.Wtf I had an explanation for this in my commit message, I must have exited without saving.
Quoted 8 lines
I tried to do this actually, and it ended up spewing because you cannot change cgroup.subtree_control for a cgroup you are currently inside of. We can fix this by just making sure all controllers are enabled in our test cgroup, and then fix 562 to not bother messing subtree and just assume everything is enabled. But since it's just one test and I wasn't sure if it would create more wonkiness I just left it as is. I'm open to either solution, I just opted for the less change route. Thanks,
Oh, Ok. Eh, it seems the test should still work either way so I'm fine with leaving it as is as well until there's some better reason to do otherwise, as long as you can add that explanation back into the commit log..
Brian
Quoted 2 lines
Josef