perf-lib: remove old result files before running tests
PATCH 1/1 perf-lib: remove old result files before running tests
The perf tests write files recording the results of tests. These results are later aggregated by 'aggregate.perl'. If the tests are run multiple times, those results are overwritten by the new results. This works just fine as long as there are only perf tests measuring the times, whose results are stored in "$base".times files.
However 22bec79d1a ("t/perf: add infrastructure for measuring sizes", 2018-08-17) introduced a new type of test for measuring the size of something. The results of this are written to "$base".size files.
"$base" is essentially made up of the basename of the script plus the test number. So if test numbers shift because a new test was introduced earlier in the script we might end up with both a ".times" and a ".size" file for the same test. In the aggregation script the ".times" file is preferred over the ".size" file, so some size tests might end with performance numbers from a previous run of the test.
This is mainly relevant when writing perf tests that check both performance and sizes, and can get quite confusing during developement.
---
This came out of something different that I'm working on, but makes most sense as a standalone patch, rather than part of that series, so I'm sending this out separately.
t/perf/perf-lib.sh | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-)
t/perf/perf-lib.sh +4 -3
@@ -178,10 +178,11 @@ test_wrapper_ () { export test_prereq if ! test_skip "$@" then - base=$(basename "$0" .sh) - echo "$test_count" >>"$perf_results_dir"/$base.subtests - echo "$1" >"$perf_results_dir"/$base.$test_count.descr base="$perf_results_dir"/"$PERF_RESULTS_PREFIX$(basename "$0" .sh)"."$test_count" + rm -f "$base".* + no_prefix_base="$perf_results_dir"/$(basename "$0" .sh) + echo "$test_count" >>$no_prefix_base.subtests + echo "$1" >$no_prefix_base.$test_count.descr "$test_wrapper_func_" "$@" fi
Signature · 1 line
2.24.0.155.gd9f6f3b619
t/perf/perf-lib.sh test_wrapper_ () { · line 182
Quoted 6 lines · Thomas Gummerer
t/perf/perf-lib.sh +0
-1@@ -178,10 +178,11 @@ test_wrapper_ () { export test_prereq if ! test_skip "$@" then - base=$(basename "$0" .sh)
So we used to use $base to hold the number and the filename here
t/perf/perf-lib.sh test_wrapper_ () { · line 181
Quoted 3 lines
t/perf/perf-lib.sh +0
-2- echo "$test_count" >>"$perf_results_dir"/$base.subtests - echo "$1" >"$perf_results_dir"/$base.$test_count.descr base="$perf_results_dir"/"$PERF_RESULTS_PREFIX$(basename "$0" .sh)"."$test_count"
and then redefined it to be the results-prefix specific one.
t/perf/perf-lib.sh test_wrapper_ () { · line 182
Quoted 1 line
t/perf/perf-lib.sh +1
-0+ rm -f "$base".*
you now remove those results-prefix specific one for the $test_count (I guess that is as specific you can go) before writing the count and the description.
So this "rm -f" is a no-op when perf-results-prefix is not empty?
Quoted 5 lines
t/perf/perf-lib.sh +3
-0+ no_prefix_base="$perf_results_dir"/$(basename "$0" .sh) + echo "$test_count" >>$no_prefix_base.subtests + echo "$1" >$no_prefix_base.$test_count.descr "$test_wrapper_func_" "$@" fi
Quoted 18 lines · Junio C Hamano
t/perf/perf-lib.sh test_wrapper_ () { · line 182
Quoted 6 lines · Thomas Gummerer
t/perf/perf-lib.sh +0
-1@@ -178,10 +178,11 @@ test_wrapper_ () { export test_prereq if ! test_skip "$@" then - base=$(basename "$0" .sh)So we used to use $base to hold the number and the filename here
t/perf/perf-lib.sh test_wrapper_ () { · line 181
Quoted 3 lines
t/perf/perf-lib.sh +0
-2- echo "$test_count" >>"$perf_results_dir"/$base.subtests - echo "$1" >"$perf_results_dir"/$base.$test_count.descr base="$perf_results_dir"/"$PERF_RESULTS_PREFIX$(basename "$0" .sh)"."$test_count"and then redefined it to be the results-prefix specific one.
Quoted 7 lines
t/perf/perf-lib.sh test_wrapper_ () { · line 182
Quoted 1 line
t/perf/perf-lib.sh +1
-0+ rm -f "$base".*you now remove those results-prefix specific one for the $test_count (I guess that is as specific you can go) before writing the count and the description.
So this "rm -f" is a no-op when perf-results-prefix is not empty?
No, not quite. It is a no-op the first time a particular test runs for a specific prefix. The prefix is usually set by the 't/perf/run' script, and indicates the revision that is tested.
So if we were running for example
./run deadbeef... p0001-rev-list.sh
we'd have a file 'test-results/build_deadbeef....p0001-rev-list.1.times' in the t/perf directory.
Now we add a 'test_size' test before the first 'test_perf' test, and run the tests again. After this run we'd still have that original file from the test results from the previous run, as well as a 'test-results/build_deadbeef....p0001-rev-list.1.size'.
This duplicate file at the matching "$base" is what we want to avoid. The "rm -f" above would remove 'test-results/build_deadbeef....p0001-rev-list.1.times' so we now only have the '.size' file left, and 'aggregate.perl' gives us the right result.
Quoted 5 lines
Quoted 5 lines
t/perf/perf-lib.sh +3
-0+ no_prefix_base="$perf_results_dir"/$(basename "$0" .sh) + echo "$test_count" >>$no_prefix_base.subtests + echo "$1" >$no_prefix_base.$test_count.descr "$test_wrapper_func_" "$@" fi
Quoted 17 lines · Thomas Gummerer
The perf tests write files recording the results of tests. These results are later aggregated by 'aggregate.perl'. If the tests are run multiple times, those results are overwritten by the new results. This works just fine as long as there are only perf tests measuring the times, whose results are stored in "$base".times files.
However
22bec79d1a("t/perf: add infrastructure for measuring sizes", 2018-08-17) introduced a new type of test for measuring the size of something. The results of this are written to "$base".size files."$base" is essentially made up of the basename of the script plus the test number. So if test numbers shift because a new test was introduced earlier in the script we might end up with both a ".times" and a ".size" file for the same test. In the aggregation script the ".times" file is preferred over the ".size" file, so some size tests might end with performance numbers from a previous run of the test.
Quoted 3 lines
This is mainly relevant when writing perf tests that check both performance and sizes, and can get quite confusing during developement.
The problem description makes sense to me.
Quoted 7 lines
t/perf/perf-lib.sh +0
-0@@ -178,10 +178,11 @@ test_wrapper_ () { export test_prereq if ! test_skip "$@"
t/perf/perf-lib.sh test_wrapper_ () { · line 186
Quoted 10 lines
t/perf/perf-lib.sh +4
-3then - base=$(basename "$0" .sh) - echo "$test_count" >>"$perf_results_dir"/$base.subtests - echo "$1" >"$perf_results_dir"/$base.$test_count.descr base="$perf_results_dir"/"$PERF_RESULTS_PREFIX$(basename "$0" .sh)"."$test_count" + rm -f "$base".* + no_prefix_base="$perf_results_dir"/$(basename "$0" .sh) + echo "$test_count" >>$no_prefix_base.subtests + echo "$1" >$no_prefix_base.$test_count.descr "$test_wrapper_func_" "$@"
I had a little trouble following the patch because of a few things:
- the reordering of earlier lines. We don't care about subtests at all, and that line could stay the same. But we do have to reorder the "descr" one because of the broad wildcard in the "rm". That could be narrowed, but I guess you wanted to future-proof it against new types.
- the $no_prefix_base variable differs from $base not just in lacking the prefix, but also in lacking $test_count at the end.
So I think it's doing the right thing overall, though there is one bug:
$no_prefix_base is not quoted, but could contain spaces due to $perf_results_dir. I think that would cause bash to complain if there's a space in your path.
But I wonder if it would be simpler to just always use the same file for the test result, overwriting it each time, and let the reader figure out the type. The aggregate script's get_times() already uses a regex to distinguish the two. That's enough for the two types we have, and we could later add a header line if it becomes necessary.
Something like the patch below. That removes any confusion about cruft files being left behind, or which file should be preferred, etc.
The diff would be even smaller if we just kept calling it "times", but that's probably unnecessarily confusing.
-Peff
---
t/perf/aggregate.perl +3 -9
@@ -219,13 +219,7 @@ sub print_default_results { for my $i (0..$#dirs) { my $d = $dirs[$i]; my $base = "$resultsdir/$prefixes{$d}$t"; - $times{$prefixes{$d}.$t} = []; - foreach my $type (qw(times size)) { - if (-e "$base.$type") { - $times{$prefixes{$d}.$t} = [get_times("$base.$type")]; - last; - } - } + $times{$prefixes{$d}.$t} = [get_times("$base.result")]; my ($r,$u,$s) = @{$times{$prefixes{$d}.$t}}; my $w = length format_times($r,$u,$s,$firstr); $colwidth[$i] = $w if $w > $colwidth[$i];
@@ -267,7 +261,7 @@ sub print_sorted_results { my ($prevr, $prevu, $prevs, $prevrev); for my $i (0..$#dirs) { my $d = $dirs[$i]; - my ($r, $u, $s) = get_times("$resultsdir/$prefixes{$d}$t.times"); + my ($r, $u, $s) = get_times("$resultsdir/$prefixes{$d}$t.result"); if ($i > 0 and defined $r and defined $prevr and $prevr > 0) { my $percent = 100.0 * ($r - $prevr) / $prevr; push @evolutions, { "percent" => $percent,
@@ -327,7 +321,7 @@ sub print_codespeed_results { my $commitid = $prefixes{$d}; $commitid =~ s/^build_//; $commitid =~ s/\.$//; - my ($result_value, $u, $s) = get_times("$resultsdir/$prefixes{$d}$t.times"); + my ($result_value, $u, $s) = get_times("$resultsdir/$prefixes{$d}$t.result"); my %vals = ( "commitid" => $commitid,
t/perf/perf-lib.sh +2 -2
@@ -214,7 +214,7 @@ test_perf_ () { else test_ok_ "$1" fi - "$TEST_DIRECTORY"/perf/min_time.perl test_time.* >"$base".times + "$TEST_DIRECTORY"/perf/min_time.perl test_time.* >"$base".result } test_perf () {
@@ -223,7 +223,7 @@ test_perf () { test_size_ () { say >&3 "running: $2" - if test_eval_ "$2" 3>"$base".size; then + if test_eval_ "$2" 3>"$base".result; then test_ok_ "$1" else test_failure_ "$@"
Quoted 8 lines · Jeff King
But I wonder if it would be simpler to just always use the same file for the test result, overwriting it each time, and let the reader figure out the type. The aggregate script's
get_times()already uses a regex to distinguish the two. That's enough for the two types we have, and we could later add a header line if it becomes necessary.Something like the patch below. That removes any confusion about cruft files being left behind, or which file should be preferred, etc.
Yeah, I like what you have below much more than what I had, thanks!
Quoted 2 lines
The diff would be even smaller if we just kept calling it "times", but that's probably unnecessarily confusing.
Yup, I prefer a more readable end result to a small diff :)
Quoted 62 lines
(quoted diff) +0
-2-Peff ---t/perf/aggregate.perl +3
-9@@ -219,13 +219,7 @@ sub print_default_results { for my $i (0..$#dirs) { my $d = $dirs[$i]; my $base = "$resultsdir/$prefixes{$d}$t"; - $times{$prefixes{$d}.$t} = []; - foreach my $type (qw(times size)) { - if (-e "$base.$type") { - $times{$prefixes{$d}.$t} = [get_times("$base.$type")]; - last; - } - } + $times{$prefixes{$d}.$t} = [get_times("$base.result")]; my ($r,$u,$s) = @{$times{$prefixes{$d}.$t}}; my $w = length format_times($r,$u,$s,$firstr); $colwidth[$i] = $w if $w > $colwidth[$i];@@ -267,7 +261,7 @@ sub print_sorted_results { my ($prevr, $prevu, $prevs, $prevrev); for my $i (0..$#dirs) { my $d = $dirs[$i]; - my ($r, $u, $s) = get_times("$resultsdir/$prefixes{$d}$t.times"); + my ($r, $u, $s) = get_times("$resultsdir/$prefixes{$d}$t.result"); if ($i > 0 and defined $r and defined $prevr and $prevr > 0) { my $percent = 100.0 * ($r - $prevr) / $prevr; push @evolutions, { "percent" => $percent,@@ -327,7 +321,7 @@ sub print_codespeed_results { my $commitid = $prefixes{$d}; $commitid =~ s/^build_//; $commitid =~ s/\.$//; - my ($result_value, $u, $s) = get_times("$resultsdir/$prefixes{$d}$t.times"); + my ($result_value, $u, $s) = get_times("$resultsdir/$prefixes{$d}$t.result"); my %vals = ( "commitid" => $commitid,t/perf/perf-lib.sh +2
-2@@ -214,7 +214,7 @@ test_perf_ () { else test_ok_ "$1" fi - "$TEST_DIRECTORY"/perf/min_time.perl test_time.* >"$base".times + "$TEST_DIRECTORY"/perf/min_time.perl test_time.* >"$base".result } test_perf () {@@ -223,7 +223,7 @@ test_perf () { test_size_ () { say >&3 "running: $2" - if test_eval_ "$2" 3>"$base".size; then + if test_eval_ "$2" 3>"$base".result; then test_ok_ "$1" else test_failure_ "$@"
Quoted 10 lines · Thomas Gummerer
Quoted 8 lines · Jeff King
But I wonder if it would be simpler to just always use the same file for the test result, overwriting it each time, and let the reader figure out the type. The aggregate script's
get_times()already uses a regex to distinguish the two. That's enough for the two types we have, and we could later add a header line if it becomes necessary.Something like the patch below. That removes any confusion about cruft files being left behind, or which file should be preferred, etc.
Quoted 6 lines
Yeah, I like what you have below much more than what I had, thanks!
Quoted 2 lines
The diff would be even smaller if we just kept calling it "times", but that's probably unnecessarily confusing.
Yup, I prefer a more readable end result to a small diff :)
OK, here it is wrapped up in a commit message. I wasn't sure how to attribute it, since you really did most of the interesting work and I just swooped in with an alternative. I'm happy to make you the author, but I didn't want you to get blamed for my bugs. ;)
Junio, this would replace tg/perf-remove-stale-result.
-- >8 --
Subject: [PATCH] perf-lib: use a single filename for all measurement types
The perf tests write files recording the results of tests. These results are later aggregated by 'aggregate.perl'. If the tests are run multiple times, those results are overwritten by the new results. This works just fine as long as there are only perf tests measuring the times, whose results are stored in "$base".times files.
However 22bec79d1a ("t/perf: add infrastructure for measuring sizes", 2018-08-17) introduced a new type of test for measuring the size of something. The results of this are written to "$base".size files.
"$base" is essentially made up of the basename of the script plus the test number. So if test numbers shift because a new test was introduced earlier in the script we might end up with both a ".times" and a ".size" file for the same test. In the aggregation script the ".times" file is preferred over the ".size" file, so some size tests might end with performance numbers from a previous run of the test.
This is mainly relevant when writing perf tests that check both performance and sizes, and can get quite confusing during developement.
We could fix this by doing a more thorough job of cleaning out old ".times" and ".size" files before running each test. However, an even easier solution is to just use the same filename for both types of measurement, meaning we'll always overwrite the previous result. We don't even need to change the file format to distinguish the two; aggregate.perl already decides which is which based on a regex of the content (this may become ambiguous if we add new types in the future, but we could easily add a header field to the file at that point).
Based on an initial patch from Thomas Gummerer, who discovered the problem and did all of the analysis (which I stole for the commit message above):
https://public-inbox.org/git/[email protected]/
--- t/perf/aggregate.perl | 12 +++--------- t/perf/perf-lib.sh | 4 ++-- 2 files changed, 5 insertions(+), 11 deletions(-)
t/perf/aggregate.perl +3 -9
@@ -219,13 +219,7 @@ sub print_default_results { for my $i (0..$#dirs) { my $d = $dirs[$i]; my $base = "$resultsdir/$prefixes{$d}$t"; - $times{$prefixes{$d}.$t} = []; - foreach my $type (qw(times size)) { - if (-e "$base.$type") { - $times{$prefixes{$d}.$t} = [get_times("$base.$type")]; - last; - } - } + $times{$prefixes{$d}.$t} = [get_times("$base.result")]; my ($r,$u,$s) = @{$times{$prefixes{$d}.$t}}; my $w = length format_times($r,$u,$s,$firstr); $colwidth[$i] = $w if $w > $colwidth[$i];
@@ -267,7 +261,7 @@ sub print_sorted_results { my ($prevr, $prevu, $prevs, $prevrev); for my $i (0..$#dirs) { my $d = $dirs[$i]; - my ($r, $u, $s) = get_times("$resultsdir/$prefixes{$d}$t.times"); + my ($r, $u, $s) = get_times("$resultsdir/$prefixes{$d}$t.result"); if ($i > 0 and defined $r and defined $prevr and $prevr > 0) { my $percent = 100.0 * ($r - $prevr) / $prevr; push @evolutions, { "percent" => $percent,
@@ -327,7 +321,7 @@ sub print_codespeed_results { my $commitid = $prefixes{$d}; $commitid =~ s/^build_//; $commitid =~ s/\.$//; - my ($result_value, $u, $s) = get_times("$resultsdir/$prefixes{$d}$t.times"); + my ($result_value, $u, $s) = get_times("$resultsdir/$prefixes{$d}$t.result"); my %vals = ( "commitid" => $commitid,
t/perf/perf-lib.sh +2 -2
@@ -214,7 +214,7 @@ test_perf_ () { else test_ok_ "$1" fi - "$TEST_DIRECTORY"/perf/min_time.perl test_time.* >"$base".times + "$TEST_DIRECTORY"/perf/min_time.perl test_time.* >"$base".result } test_perf () {
@@ -223,7 +223,7 @@ test_perf () { test_size_ () { say >&3 "running: $2" - if test_eval_ "$2" 3>"$base".size; then + if test_eval_ "$2" 3>"$base".result; then test_ok_ "$1" else test_failure_ "$@"
Signature · 1 line
2.24.0.716.g722aff65ed
Quoted 17 lines · Jeff King
Quoted 15 lines · Thomas Gummerer
Quoted 8 lines · Jeff King
But I wonder if it would be simpler to just always use the same file for the test result, overwriting it each time, and let the reader figure out the type. The aggregate script's
get_times()already uses a regex to distinguish the two. That's enough for the two types we have, and we could later add a header line if it becomes necessary.Something like the patch below. That removes any confusion about cruft files being left behind, or which file should be preferred, etc.
Yeah, I like what you have below much more than what I had, thanks!
Quoted 2 lines
The diff would be even smaller if we just kept calling it "times", but that's probably unnecessarily confusing.
Quoted 6 lines
Quoted 1 line
Yup, I prefer a more readable end result to a small diff :)
OK, here it is wrapped up in a commit message. I wasn't sure how to attribute it, since you really did most of the interesting work and I just swooped in with an alternative. I'm happy to make you the author, but I didn't want you to get blamed for my bugs. ;)
Thanks for tying this up. I wasn't sure how to proceed either, so I'm glad you submitted the patch. And I'm happy to have you as author with the helped-by attribution as you are the one that actually wrote the code :)
I eyeballed the patch again and applied and tested it for good measure and it all looks good to me!
Quoted 110 lines
Junio, this would replace tg/perf-remove-stale-result.
-- >8 --
Subject: [PATCH] perf-lib: use a single filename for all measurement typesThe perf tests write files recording the results of tests. These results are later aggregated by 'aggregate.perl'. If the tests are run multiple times, those results are overwritten by the new results. This works just fine as long as there are only perf tests measuring the times, whose results are stored in "$base".times files.
However
22bec79d1a("t/perf: add infrastructure for measuring sizes", 2018-08-17) introduced a new type of test for measuring the size of something. The results of this are written to "$base".size files."$base" is essentially made up of the basename of the script plus the test number. So if test numbers shift because a new test was introduced earlier in the script we might end up with both a ".times" and a ".size" file for the same test. In the aggregation script the ".times" file is preferred over the ".size" file, so some size tests might end with performance numbers from a previous run of the test.
This is mainly relevant when writing perf tests that check both performance and sizes, and can get quite confusing during developement.
We could fix this by doing a more thorough job of cleaning out old ".times" and ".size" files before running each test. However, an even easier solution is to just use the same filename for both types of measurement, meaning we'll always overwrite the previous result. We don't even need to change the file format to distinguish the two; aggregate.perl already decides which is which based on a regex of the content (this may become ambiguous if we add new types in the future, but we could easily add a header field to the file at that point).
Based on an initial patch from Thomas Gummerer, who discovered the problem and did all of the analysis (which I stole for the commit message above):
https://public-inbox.org/git/[email protected]/
Helped-by Thomas Gummerer <[email protected]>Signed-off-by Jeff King <[email protected]>--- t/perf/aggregate.perl | 12 +++--------- t/perf/perf-lib.sh | 4 ++-- 2 files changed, 5 insertions(+), 11 deletions(-)t/perf/aggregate.perl +3
-9@@ -219,13 +219,7 @@ sub print_default_results { for my $i (0..$#dirs) { my $d = $dirs[$i]; my $base = "$resultsdir/$prefixes{$d}$t"; - $times{$prefixes{$d}.$t} = []; - foreach my $type (qw(times size)) { - if (-e "$base.$type") { - $times{$prefixes{$d}.$t} = [get_times("$base.$type")]; - last; - } - } + $times{$prefixes{$d}.$t} = [get_times("$base.result")]; my ($r,$u,$s) = @{$times{$prefixes{$d}.$t}}; my $w = length format_times($r,$u,$s,$firstr); $colwidth[$i] = $w if $w > $colwidth[$i];@@ -267,7 +261,7 @@ sub print_sorted_results { my ($prevr, $prevu, $prevs, $prevrev); for my $i (0..$#dirs) { my $d = $dirs[$i]; - my ($r, $u, $s) = get_times("$resultsdir/$prefixes{$d}$t.times"); + my ($r, $u, $s) = get_times("$resultsdir/$prefixes{$d}$t.result"); if ($i > 0 and defined $r and defined $prevr and $prevr > 0) { my $percent = 100.0 * ($r - $prevr) / $prevr; push @evolutions, { "percent" => $percent,@@ -327,7 +321,7 @@ sub print_codespeed_results { my $commitid = $prefixes{$d}; $commitid =~ s/^build_//; $commitid =~ s/\.$//; - my ($result_value, $u, $s) = get_times("$resultsdir/$prefixes{$d}$t.times"); + my ($result_value, $u, $s) = get_times("$resultsdir/$prefixes{$d}$t.result"); my %vals = ( "commitid" => $commitid,t/perf/perf-lib.sh +2
-2@@ -214,7 +214,7 @@ test_perf_ () { else test_ok_ "$1" fi - "$TEST_DIRECTORY"/perf/min_time.perl test_time.* >"$base".times + "$TEST_DIRECTORY"/perf/min_time.perl test_time.* >"$base".result } test_perf () {@@ -223,7 +223,7 @@ test_perf () { test_size_ () { say >&3 "running: $2" - if test_eval_ "$2" 3>"$base".size; then + if test_eval_ "$2" 3>"$base".result; then test_ok_ "$1" else test_failure_ "$@"Signature · 1 line
2.24.0.716.g722aff65ed