| Summary: | Tools/Scripts/test-webkitpy-python2 is flaky (Cannot send message to closed queue) | ||
|---|---|---|---|
| Product: | WebKit | Reporter: | Kimmo Kinnunen <kkinnunen> |
| Component: | Tools / Tests | Assignee: | Jonathan Bedard <jbedard> |
| Status: | NEW --- | ||
| Severity: | Normal | CC: | aakash_jain, ap, gsnedders, jbedard, webkit-bug-importer |
| Priority: | P2 | Keywords: | InRadar |
| Version: | Other | ||
| Hardware: | Unspecified | ||
| OS: | Unspecified | ||
|
Description
Kimmo Kinnunen
2022-08-16 05:33:35 PDT
> Blocked my commit
Note that webkitpy queue doesn't block PRs. Some other queue (e.g.: builder or layout-test queue) would have blocked the PR.
My guess is that one of the subprocesses is crashing (and older multiprocessing doesn't handle workers going away unexpectedly), causing this.
I do wonder if we should just move to running tests with -j1 on Python 2, like we already do on Windows.
Locally this is a decent perf regression, but maybe worth it for the stability:
```
% hyperfine -r1 -L children 1,8 './Tools/Scripts/test-webkitpy-python2 -j{children}'
Benchmark 1: ./Tools/Scripts/test-webkitpy-python2 -j1
Time (abs ≡): 89.726 s [User: 41.462 s, System: 15.612 s]
Benchmark 2: ./Tools/Scripts/test-webkitpy-python2 -j8
Time (abs ≡): 25.100 s [User: 43.881 s, System: 20.178 s]
Summary
./Tools/Scripts/test-webkitpy-python2 -j8 ran
3.57 times faster than ./Tools/Scripts/test-webkitpy-python2 -j1
```
This is especially true when the majority of PRs don't run the webkitpy tests at all, so the bot is largely free.
(Note however my local machine is much faster than what we're running in EWS: the -j8 time is nearly twice as fast, but a 4x multiplier would take us from 1 minute on EWS to run the Python 2 tests to 4 minutes, which is probably still acceptable.)
|