We should investigate why |RecordShutdownEndTimeStamp| is not guarded by |CanRecordExtended|: it only checks that a previous call to |GetShutdownTimeFileName| [2] was made. This made bug 1148763 hard to track down. We should investigate if this behaviour is fine, and simply make it clearer that we are crashing because |RecordShutdownEndTimeStamp| was not called, or if we should consider guarding |RecordShutdownEndTimeStamp| with |CanRecordExtended|. [1] - https://hg.mozilla.org/mozilla-central/annotate/18a8ea7c2c62/toolkit/components/telemetry/Telemetry.cpp#l3568 [2] - https://hg.mozilla.org/mozilla-central/annotate/18a8ea7c2c62/toolkit/components/telemetry/Telemetry.cpp#l1684

This should fix the mochitest issues for bug 1137252 as well.

Comment on attachment 8592927 [details] [diff] [review] bug1149980.patch Review of attachment 8592927 [details] [diff] [review]: ----------------------------------------------------------------- ::: toolkit/components/telemetry/Telemetry.cpp @@ +3585,5 @@ > > + if (gRecordedShutdownStartTime.IsNull()) { > + // If |CanRecordExtended()| is true before |AsyncFetchTelemetryData| is called and > + // then disabled before shutdown, |RecordShutdownStartTimeStamp| will bail out and > + // we will end up with an invalid |gRecordedShutdownStartTime| here. This can happen technically, the start time can end up being null, not invalid

Thanks Vladan, I've changed the comment accordingly.

This definitely fixes the intermittent failures from bug 1154154: Linux debug bc3 retriggers: https://treeherder.mozilla.org/#/jobs?repo=try&revision=516dc607dfd0 Full try push stacking on bug 1137252: https://treeherder.mozilla.org/#/jobs?repo=try&revision=114653822987