1213865 - Limit the depth of the "nativeStack" reported by Telemetry in the thread hangs

Status: RESOLVED INVALID

(Toolkit :: Telemetry, defect, P4)

Description

[Alessio Placitelli [:Dexter]]

Some thread hangs reported in the threadHangStats section of Telemetry shutdown pings include a "nativeStack" element, along with the "stack", with a depth of 35+ calls.

We should investigate if it's worth reducing the depth of the reported nativeStack to reduce the ping size.

Comment 1

[Alessio Placitelli [:Dexter]]

Vladan, do you think it makes sense to limit the depth of the "nativeStack", as we are doing with the "stack" entry?

It's worth noting that the "nativeStack" entry doesn't come with every hang in threadHangStats and, as such, doesn't have a huge impact on the ping size.

Comment 2

[Georg Fritzsche [:gfritzsche]]

I think we should have a stack depth limit either way (it shouldn't be unbounded).
The question is probably whether it should be "pretty high" (capturing 99-100% of the populations data) or whether we need to cut it down more agressively.

Comment 3

[Vladan Djeric (:vladan)]

(In reply to Alessio Placitelli [:Dexter] from comment #1)
> Vladan, do you think it makes sense to limit the depth of the "nativeStack",
> as we are doing with the "stack" entry?

I think the native stack is currently only captured during permanent thread hangs (see bug 1034138).
We might need to have different max depths for native & BHR stacks, e.g. 95th percentile of each

Jim: We're discussing imposing maximum lengths on BHR stacks in this bug (because of Telemetry ping size). Is there a reason to treat native BHR stacks differently from pseudo-stack? For example, I think native stacks will generally be deeper but currently they're only for permanent hangs, so it seems ok to allow them to be longer. What do you think?

Comment 4

[(inactive) Jim Chen [:jchen] [:darchons]]

Sorry for the late response. Having a shorter limit for pseudostack and longer limit for native stack sounds reasonable. FWIW, native stacks are actually turned off right now because of work needed on the Breakpad backend (bug 1196381).

Comment 5

[Chris H-C :chutten]

bug 1380081 removed thread hangs from the ping, so no bug here :D