Re: [PATCH v2 0/8] Run tests in parallel showing number of tests running

From: Namhyung Kim
Date: Thu Oct 17 2024 - 19:29:04 EST


On Thu, Oct 17, 2024 at 05:49:12AM -0700, Ian Rogers wrote:
> On Wed, Oct 16, 2024 at 5:28 PM Ian Rogers <irogers@xxxxxxxxxx> wrote:
> >
> > On Wed, Oct 16, 2024 at 4:49 PM Namhyung Kim <namhyung@xxxxxxxxxx> wrote:
> > >
> > > On Fri, Oct 11, 2024 at 03:03:46PM -0700, Ian Rogers wrote:
> > > > Avoid waitpid so that stdout/stderr aren't destroyed prior to wanting
> > > > to read them for display. When running on a color terminal, display
> > > > the number of running tests (1 if sequential). To avoid previous
> > > > flicker, only delete and refresh the display line when it changes. An
> > > > earlier version of this code is here:
> > > > https://lore.kernel.org/lkml/20240701044236.475098-1-irogers@xxxxxxxxxx/
> > > >
> > > > Add a signal handler for perf tests so that unexpected signals are
> > > > displayed and test clean up is possible.
> > > >
> > > > In perf test add an "exclusive" flag that causes a test to be run with
> > > > no other test. Set this flag manually for C tests and via a
> > > > "(exclusive)" in the test description for shell tests. Add the flag to
> > > > shell tests that may fail when run with other tests.
> > > >
> > > > Change the perf test loop to run in two passes. For parallel
> > > > execution, the first pass runs all tests that can be run in parallel
> > > > then the 2nd runs remaining tests sequentially. This causes the
> > > > "exclusive" tests to be run last and with test numbers moderately out
> > > > of alignment.
> > > >
> > > > Change the default to be to run tests in parallel. Running tests in
> > > > parallel brings the execution time down to less than half.
> > > >
> > > > v2: Fix inaccurate remaining counts when running specific
> > > > tests. Rename "remaining" to "active" to better reflect the
> > > > testing behavior. Move the exclusive flag to test cases and not
> > > > entire suites. Add more "(exclusive)" flags to test as
> > > > suggested-by James Clark. Remove "(exclusive)" flag from test
> > > > descriptions to keep the command line output more concise. Add
> > > > James Clark's tested-by.
> > > >
> > > > Ian Rogers (8):
> > > > tools subcmd: Add non-waitpid check_if_command_finished()
> > > > perf test: Display number of active running tests
> > > > perf test: Reduce scope of parallel variable
> > > > perf test: Avoid list test blocking on writing to stdout
> > > > perf test: Tag parallel failing shell tests with "(exclusive)"
> > > > perf test: Add a signal handler around running a test
> > > > perf test: Run parallel tests in two passes
> > > > perf test: Make parallel testing the default
> > >
> > > Nice work! It looks much better now.
> > >
> > > But I'm seeing more failures in parallel mode. Maybe we want to
> > > keep the default serial mode for a little more.
> >
> > As you say, I think we should be conservative and mark all tests that
> > need to serial/sequential/exclusive with the exclusive tag. If you
> > tell me the failing tests I can add them to 'perf test: Tag parallel
> > failing shell tests with "(exclusive)"' as I did for James Clark with
> > the ARM tests. I'd prefer we did the tagging rather than not enabling
> > parallel testing as otherwise I may never learn which tests fail for
> > people when run in parallel.
>
> With repeat testing, most often for me it was fine, I was able to get
> a flake on the probe plus vfs_getname tests like:
> ```
> $ sudo /tmp/perf/perf test vfs -v
> 91: Add vfs_getname probe to get syscall args filenames : Ok
> --- start ---
> test child forked, pid 466904
> Failed to write event: File exists
> Error: Failed to add events.
> ---- end(-1) ----
> 93: Use vfs_getname probe to get syscall args filenames : FAILED!
> --- start ---
> test child forked, pid 466906
> Error: event "vfs_getname" already exists.
> Hint: Remove existing event by 'perf probe -d'
> or force duplicates by 'perf probe -f'
> or set 'force=yes' in BPF source.
> Error: Failed to add events.
> ---- end(-1) ----
> 127: Check open filename arg using perf trace + vfs_getname : FAILED!
> ```
> So I'll make those exclusive in v2 too. If you could let me know of others.

Mine is the below (other than the existing probe test failure). I ran
them 3 times and picked one failed at least once.

92: Add vfs_getname probe to get syscall args filenames : FAILED!
94: Use vfs_getname probe to get syscall args filenames : FAILED!
112: perf stat --bpf-counters test : FAILED!
121: Test data symbol : FAILED!
128: Check open filename arg using perf trace + vfs_getname : FAILED!

Thanks,
Namhyung

>
> > > >
> > > > tools/lib/subcmd/run-command.c | 33 +++
> > > > tools/perf/tests/builtin-test.c | 274 ++++++++++++------
> > > > .../tests/shell/coresight/asm_pure_loop.sh | 2 +-
> > > > .../shell/coresight/memcpy_thread_16k_10.sh | 2 +-
> > > > .../coresight/thread_loop_check_tid_10.sh | 2 +-
> > > > .../coresight/thread_loop_check_tid_2.sh | 2 +-
> > > > .../shell/coresight/unroll_loop_thread_10.sh | 2 +-
> > > > tools/perf/tests/shell/list.sh | 5 +-
> > > > .../tests/shell/perftool-testsuite_report.sh | 2 +-
> > > > tools/perf/tests/shell/record.sh | 2 +-
> > > > tools/perf/tests/shell/record_lbr.sh | 2 +-
> > > > tools/perf/tests/shell/record_offcpu.sh | 2 +-
> > > > tools/perf/tests/shell/stat_all_pmu.sh | 2 +-
> > > > tools/perf/tests/shell/test_arm_coresight.sh | 2 +-
> > > > .../tests/shell/test_arm_coresight_disasm.sh | 2 +-
> > > > tools/perf/tests/shell/test_arm_spe.sh | 2 +-
> > > > tools/perf/tests/shell/test_intel_pt.sh | 2 +-
> > > > .../perf/tests/shell/test_stat_intel_tpebs.sh | 2 +-
> > > > tools/perf/tests/task-exit.c | 9 +-
> > > > tools/perf/tests/tests-scripts.c | 7 +-
> > > > tools/perf/tests/tests.h | 9 +
> > > > tools/perf/util/color.h | 1 +
> > > > 22 files changed, 258 insertions(+), 110 deletions(-)
> > > >
> > > > --
> > > > 2.47.0.rc1.288.g06298d1525-goog
> > > >