Currently the Firefox hang monitor (from bug 429592) is triggering for plugin hangs because the plugin hang timeout (45s) is longer than the Firefox hang timeout (30s). I propose shortening the plugin hang timeout to 25 seconds and see what kind of data we get.
Created attachment 576998 [details] [diff] [review] Shorten plugin hang timeout to 25 secs, rev. 1
FWIW, I think this is a righteous change.
Landed https://hg.mozilla.org/mozilla-central/rev/6f079f13c06a with presumptive-r so that we can get proper feedback over the weekend without annoying users.
The plugin hang was set to 45 seconds in Firefox 3.6.6 because of on old computers some facebook games took more than 40 seconds to load and caused a number of crashes. http://www.mozilla.org/en-US/firefox/3.6.6/releasenotes/ See bug 574905
Firefox hang timeout was not implemented at that point.
Asa and I are both well aware of the history of bug 574905, which is why I wrote "and see what kind of data we get". It is likely that the hang monitor and this timeout change will be backed out of beta and release channels for a few releases yet while we monitor the crash-stats data and figure out how to work around these issues.
I fortunately caught this by reading pushlog, but could you notify the CrashKill team when tweaking those hang time parameters? As we look at numbers of hangs, there probably can be correlations with those changes, so we should know to be watching for that. Thanks.
This bug has improved things, but that's still worse than before bug 429592 landed: BuildID Crashes 20111123: 754 20111124 2939 <-- Bug 429592 20111125 2915 20111126 1809 <-- Bug 705365
Ben (or somebody else). I spoke to Asa about this. He has suggested that we back it out for now - put the timer back to 45s. He wants to do some user research targeted around assessing the right level of timeout. Also, the implication in comment #6 is that we weren't going to keep it in for beta and release. The hang detector was only on for a few days and this change just got left in.
The hang ratio is 0.06% in 10.0 and 1.4% in 11.0b1!
Is there anything QA can do to verify this fix?