sync_integration_test tests likely leak state between test runs. |
||
Issue descriptionFor context, see: https://bugs.chromium.org/p/chromium/issues/detail?id=919945 The test SingleClientWalletSyncTest.DownloadAccountStorage_Card [TSAN build] fails very frequently when run standalone. However, it fails very rarely when run as part of the whole test suite. This suggests that we are leaking state between test runs [reusing Chrome instance, perhaps?]. We should fix this [assuming it doesn't increase the total test run time by a crazy amount] to improve consistency of test results.
,
Jan 10
,
Jan 11
I am not super convinced by the argumentation for the state leakage. I can imagine this relates to the way TSAN runs / checks for thread violations. The TSAN issue hopefully should get fixed before the weekend. Do we have any way to test/debug it when the TSAN issue is solved?
,
Jan 11
Do we reuse the browser between tests in sync_integration_test? That could be a source of state leakage. It looks like the test in this bug is part of a test fixture -- those normally also keep some state between tests. Here's another example of a non-TSAN sync_integration_test test which occasionally flakes when run in a batch, but regularly fails when run standalone: https://bugs.chromium.org/p/chromium/issues/detail?id=908771#c4 > Do we have any way to test/debug it when the TSAN issue is solved? I don't know of a way to confirm that state *isn't* being leaked. I can just point out several examples in sync_integration_tests that suggest that state is being leaked. |
||
►
Sign in to add a comment |
||
Comment 1 by treib@chromium.org
, Jan 9Components: Services>Sync