Meta: Good crash statistics

[pocmo]

Description

We want to have good crash statistics (e.g. crash rates) in order to:

• make release decisions
• measure the app quality
• identify regressions

Our current crash reporting tools are insufficient:

• Sentry only knows about crashes and not users. Therefore Sentry cannot calculate crash rates.
• Socorro is only used for native code crashes and does not know about other crashes.
• Google Play has some statistics but we can't correlate them with our other data (e.g. telemetry)

Resources

• Bug 1407557 - Fennec implementing crash ping

Target

• https://github.com/mozilla-mobile/fenix/issues/235

Tasks

• Bugzilla: Collect crash data from Android apps using glean
• (needs further breakdown)

Dependencies

• [x] Discuss with glean team.

┆Issue is synchronized with this Jira Task

[pocmo]

@Dexterp37 FYI. Good crash statistics is something that has haunted us for a while. Since we can generate custom pings with Glean I think we could just build such a crash ping.

A small technical challenge here is that the crash reporter is running in a different process (since the app process crashes) and we should avoid corrupting our data while writing from multiple processes.

[Dexterp37]

@Dexterp37 FYI. Good crash statistics is something that has haunted us for a while. Since we can generate custom pings with Glean I think we could just build such a crash ping.

This is very interesting. We are starting to have conversations around crash data using glean. I'll make sure you'll be part of these conversations as well :)

A small technical challenge here is that the crash reporter is running in a different process (since the app process crashes) and we should avoid corrupting our data while writing from multiple processes.

Yes, this sounds like an interesting technical challenge. I filed bug 1522461 on our end to track this.

[pocmo]

From the GV meeting: GV catches native code crashes; In non-GV apps with native code (e.g. Lockbox; or a WebView-based browser) we may not be able to send a crash ping since we do not catch those crashes.

[st3fan]

If we do a crash ping, it would be good to have a 'component' classifier in it, so that we can see where the crash originated. Could be a string that we fill in, like 'frontend', 'geckoview', 'glean', etc.