Right now, we collect a single backtrace from BHR when we get a hang, at the 128ms mark. This means that we don't necessarily know what prefix of each hang report is interesting, and what parts just happened to be on the stack at the time. Perhaps we should collect, say, 4 stacks (100ms, 110ms, 120ms, 130ms perhaps? Frequency and when up to debate), and take either the common prefix of the stacks, or if some of the stacks don't contain useful native information (say because they got stuck unwinding, or tried to unwind through code without debug information) then throw them out. An alternative to doing this in firefox would be for us to simply collect more stacks and just submit them to telemetry, and handle extracting prefixes in hangs.html. This has the advantage of being easier to iterate on, but means that we have to transmit more data.
I'm not sure if this has immediate benefits which I should be spending time on right now, considering the scale of the changes required, so I'm going to unassign myself until I determine it will be useful for me to work on this.
Assignee: michael → nobody
You need to log in before you can comment on or make changes to this bug.