Recent improvements in crash reporting

234 views
Skip to first unread message

Gabriele Svelto

unread,
Sep 26, 2024, 7:41:14 AMSep 26
to stability, dev-platform, crash-reporting-wg
[cross-posting to dev-platform and crash-reporting-wg]

A few important improvements in crash reporting have landed recently and
it's worth highlighting them since they'll have a significant impact.

First of all we've updated the minidump-writer crate, picking up several
months worth of work. Generating minidumps on Linux and Android now
takes hundreds of microseconds to a few milliseconds at most, a 25-30x
speedup which will be particularly welcome on slow and loaded machines
(low-end phones in particular). Additionally the logic that retrieves
information about modules does not rely on the files being available on
disk anymore. This will greatly reduce the number of cases where we're
left with truncated stack traces or missing symbols. Last but not least
some common failures when generating minidumps on Android are now
non-critical, so we expect more crashes to be captured successfully.

The second big improvement is that the new Rust-based minidump-analyzer
has just landed. This is a tool that runs on the user machine after a
crash to populate our telemetry with stack traces. The previous version
of the tool, based on Breakpad, produced decent traces on 64-bit Windows
but poor ones on all other platforms. The new tool produces quality
stack traces on all platforms we support (and some we don't). These will
soon be fed into our new telemetry dashboards, providing a clearer
picture of Firefox stability and a better way to identify the most
significant issues.

I'm taking this occasion to thank Alex Franchuk for the large amount of
work poured into reaching these milestones and Jake Shadle for his very
appreciated contributions.

Gabriele Svelto

William Kahn-Greene

unread,
Sep 26, 2024, 10:59:09 AMSep 26
to Gabriele Svelto, stability, dev-platform, crash-reporting-wg
The second big improvement affects crash pings and comparing signatures between crash pings and crash reports. Lots of groups are doing things with crash ping data. I regularly see comments regarding analysis comparing crash ping signatures to crash report signatures leading to conclusions about visibility of certain kinds of crashes.

Is anyone maintaining a "things to know about crash ping data" doc somewhere? This big change should be documented in it.

--
You received this message because you are subscribed to the Google Groups "crash-reporting-wg" group.
To unsubscribe from this group and stop receiving emails from it, send an email to crash-reporting...@mozilla.com.
To view this discussion on the web visit https://groups.google.com/a/mozilla.com/d/msgid/crash-reporting-wg/0f6df3c9-774d-4b36-a556-67958a5fbee3%40mozilla.com.
For more options, visit https://groups.google.com/a/mozilla.com/d/optout.
Reply all
Reply to author
Forward
0 new messages