test: move --cpu-prof tests to sequential #28210
Conversation
The tests still fail after being split into multiple files, (2 out of 30 runs in roughly 48 hours) and the causes are missing target frames in the samples. This patch moves them to sequential to observe if the flakiness can be fixed when the tests are run on a system with less load. If the flake ever shows up again even after the tests are moved to sequential, we should consider make the test conditions more lenient - that is, we would only assert that there are *some* frames in the generated CPU profile but do not look for the target function there.
|
Sadly, an error occurred when I tried to trigger a build. :( |
|
hmm, the stats above may be wrong because out of those 2 failed runs, one of them is https://ci.nodejs.org/job/node-test-pull-request/23810/ which was But it is still flaky as it affects https://ci.nodejs.org/job/node-test-pull-request/23832/ |
|
Parallel (master): https://ci.nodejs.org/job/node-stress-single-test/2222/ |
FWIW, here are some failures from node-test-pull-request. It's 5 in roughly the last 24 hours: https://ci.nodejs.org/job/node-test-commit-freebsd/26873/ (test-cpu-prof-dir-worker) It's all on FreeBSD, but that's not FeeBSD per se but rather our specific config in CI which has a lot of cores so runs a lot of things in parallel |
|
Hmmm...FreeBSD 11 build failed for both stress tests. Will run again on FreeBSD 10 and FreeBSD Latest if the other platforms don't show anything interesting, since this definitely seems to affect FreeBSD a lot. |
|
@Trott Thanks for starting the stress tests. It looks like on macOS which manage to build and run the tests, there was a difference though (17 v.s. 0 out of 1000 runs). I am wondering whether running that on FreeBSD 10 alone is enough - the flake has only showed up in 11 in the CI ever since the tests were split. |
|
Unfortunately, the FreeBSD hosts are not successfully building in the stress tests for either master or this branch. I'll take a look to see if I can figure anything out, or maybe someone else on @nodejs/build will have an idea. But I think the OS X results are enough to show that it's unreliable in parallel (19 failures in 1000 runs) and seems reliable in sequential (0 failures in 1000 runs). And I strongly suspect the FreeBSD hosts would show a significantly higher rate of failure (because they usually do in these types of cases). |
The tests still fail after being split into multiple files, (2 out of 30 runs in roughly 48 hours) and the causes are missing target frames in the samples. This patch moves them to sequential to observe if the flakiness can be fixed when the tests are run on a system with less load. If the flake ever shows up again even after the tests are moved to sequential, we should consider make the test conditions more lenient - that is, we would only assert that there are *some* frames in the generated CPU profile but do not look for the target function there. PR-URL: nodejs#28210 Refs: nodejs#27611 Reviewed-By: Rich Trott <rtrott@gmail.com> Reviewed-By: Michael Dawson <michael_dawson@ca.ibm.com> Reviewed-By: James M Snell <jasnell@gmail.com>
|
Looks like this landed this in 7e5e1c2 but wasn't closed? Closing, but of course, re-open if I'm wrong somehow. /ping @joyeecheung |
The tests still fail after being split into multiple files, (2 out of 30 runs in roughly 48 hours) and the causes are missing target frames in the samples. This patch moves them to sequential to observe if the flakiness can be fixed when the tests are run on a system with less load. If the flake ever shows up again even after the tests are moved to sequential, we should consider make the test conditions more lenient - that is, we would only assert that there are *some* frames in the generated CPU profile but do not look for the target function there. PR-URL: #28210 Refs: #27611 Reviewed-By: Rich Trott <rtrott@gmail.com> Reviewed-By: Michael Dawson <michael_dawson@ca.ibm.com> Reviewed-By: James M Snell <jasnell@gmail.com>
The tests still fail after being split into multiple files, (2 out of 30 runs in roughly 48 hours) and the causes are missing target frames in the samples. This patch moves them to sequential to observe if the flakiness can be fixed when the tests are run on a system with less load. If the flake ever shows up again even after the tests are moved to sequential, we should consider make the test conditions more lenient - that is, we would only assert that there are *some* frames in the generated CPU profile but do not look for the target function there. PR-URL: #28210 Refs: #27611 Reviewed-By: Rich Trott <rtrott@gmail.com> Reviewed-By: Michael Dawson <michael_dawson@ca.ibm.com> Reviewed-By: James M Snell <jasnell@gmail.com>
The tests still fail after being split into multiple files,
(2 out of 30 runs in roughly 48 hours) and the causes are missing
target frames in the samples. This patch moves them to sequential
to observe if the flakiness can be fixed when the tests are
run on a system with less load.
If the flake ever shows up again even after the tests are moved
to sequential, we should consider make the tests more
lenient - that is, we would only assert that there are some frames
in the generated CPU profile but do not look for the target
function there.
Refs: #27611
Checklist
make -j4 test(UNIX), orvcbuild test(Windows) passes