Closed Bug 705748 Opened 10 years ago Closed 10 years ago
Disable hang monitor (set pref to 0) until current set of reports have been figured out
We should disable the hang monitor introduced in bug 429592 by setting the timeout pref to 0 (see bug 429592 comment #136) as we have way too many reports coming in and we shouldn't annoy testers with crashes for too long. We have a good set of issues reported in this first wave and some work ahead to figure them all out and try to get them fixed. Once we are optimistic that we did that, we can enable this feature again, it's just a pref away. :)
Is there a bug on file for getting these hangs to be reported without crashing?
Benjamin, can you do the patch for this?
Assignee: nobody → benjamin
Status: NEW → RESOLVED
Closed: 10 years ago
Resolution: --- → FIXED
I think there should be a bug for reverting this change, blocked on fixing whatever bugs were found. Is there already such a bug?
(In reply to Brian Smith (:briansmith, :bsmith, use NEEDINFO?) from comment #5) > I think there should be a bug for reverting this change, blocked on fixing > whatever bugs were found. Is there already such a bug? 7 years passed-by and the pref is still set to 0. Anthony, what do you think or who could give an answer to that question?
IMHO this should probably be removed from the code. Crashing on hangs is not good, we found that out back then. We should use Telemetry to get that info (and I think we may already do at least some of that).
Hm, lets see if Andrea or Georg may know more.
hang reporter is an important component. it allowed (and allows) us to detect many bugs. I personal fixed bugs in DOM workers and fetch API for example, but many other components have been fixed because of this! We should not remove it. Plus, this code has been evolved and now it's able to detect if something blocks the event loop on main-thread. See bug 1356853, bug 1435961, bug 1435962, bug 1435966, bug 1445020. If the number of crashes is to high, we could consider to enable this component only in non-release builds.
I don't really have context on this, this is possibly something for the performance team to track going forward. This bug is very old, if any work is needed and issues need fixing, a new bug probably makes more sense.
Gabriele - any thoughts on this?
The upside of this is that it got us stacks of the hangs which made them actionable. Now that we have client-side stack walking we could take a minidump when we hang and then send the stack in a ping (w/o crashing). However, that's quite a bit of work and we don't have machinery to process raw stack-traces coming in (crash) pings yet.
3 years ago
We need to fix 1476714 before considering re-enabling hang reporting.
You need to log in before you can comment on or make changes to this bug.