Closed Bug 1104317 Opened 6 years ago Closed 6 years ago
Signatures for shutdown crashes
The shutdown crashes added by 1038342 are coming from a watchdog thread which is uninteresting for purposes of crash bucketing and analysis. It would be more helpful to see the main thread was doing. If the signature matches: [@ mozilla::`anonymous namespace''::RunWatchdog(void*)] [@ mozilla::(anonymous namespace)::RunWatchdog(void*)] (or maybe just some regex for RunWatchdog) Then instead let's show the stack for thread 0, ignoring anything on the regular ignore-list as well as anything containing ProcessNextEvent. E.g. bp-569807e3-431a-4f7b-a800-0bbc62141124 would display as "shutdownhang | mozilla::layers::CompositorParent::ShutDown()" This won't work well for busy-hangs like bp-b776b61d-a73a-40b1-81dd-581212141124. Taking the approach above we'd get tons of random signatures. bsmedberg are you ok with that or do you want to include some kind of cleverness like search for a frame containing the word Shutdown?
Let's ignore the busy-hang case for now: if we want to we can go back and instrument that case by measuring and annotating CPU usage. In bug 1103833 I suggested that rather than using RunWatchdog as the marker, we could use an explicit annotation. But in the short term, Run Watchdog is probably good enough.
Lonnen can you find an owner for this? It's needed in order to diagnose one of our biggest nightly crashes.
can "ProcessNextEvent" be added to the general ignore list or must it be a special case for this signature variant only? adding that frame signature to the general ignore list makes this trivial to implement. Having it as a special case makes the implementation a bit more complicated.
QA Contact: lars
Am I remembering correctly, that there are two lists, an "ignore" list and an "append" list? I think it would be reasonable to include /ProcessNextEvent/ on the general append list. (I wouldn't want to strip it altogether though.)
Yes, that is correct, there are both "ignore" and "append" lists (see: http://socorro.readthedocs.org/en/v8/signaturegeneration.html). I have proceeded with adding ProcessNextEvent to the "append" list. To see if my modifications produce what you expect, please compare these*: from production, a crash with the target signature: https://crash-stats.mozilla.com/report/index/1d69a631-7ece-4570-8c72-bffa12150120 that same crash in staging, reprocessed with a new signature rule for "shutdown hangs": https://crash-stats.allizom.org/report/index/1d69a631-7ece-4570-8c72-bffa12150120 Please verify that the signature in staging is correct. On your approval, I will submit this PR and if you act quickly, I bet we can get this into production this week. * the orignial example cited in Comment #0 could not be used because its symbols have expired, reprocessing now results in a signature that wouldn't trigger the shutdown hang rule.
Assignee: nobody → lars
QA Contact: lars
Commit pushed to master at https://github.com/mozilla/socorro https://github.com/mozilla/socorro/commit/a9a3cd3ff837b7e44d48499d29c72215588bcccd Merge pull request #2588 from twobraids/runwatchdog Fixes Bug 1104317 - adds SignatureRunWatchDog rule to processor
Status: NEW → RESOLVED
Closed: 6 years ago
Resolution: --- → FIXED
"shutdownhang..." signatures are now streaming out of the Socorro processor. There are about 4 per minute.
You need to log in before you can comment on or make changes to this bug.