811479 - Investigate the proportion of OOMs among Fennec crashes

Status: RESOLVED INCOMPLETE

(Firefox for Android Graveyard :: General, defect)

Description

[Benoit Jacob [:bjacob] (mostly away)]

FennecAndroid is massively over-represented among our crash reports: about 8% of the crash reports that we receive are FennecAndroid, which is much more than the actual proportion of FennecAndroid among our users for all I know.

The only plausible suggestion that I've heard so far, about that  aside of course for the depressing theory that FennecAndroid would be super crashy, is that the majority of these crashes are OOMs. We would just run OOM more often on Android because of the limited amount of memory on most mobile devices.

We should check that. I have an idea about how to do that, filing a separate bug as the tool would be useful on all platforms.

Comment 1

[Kartikaya Gupta (email:[e-mail])]

I think we already have telemetry metrics that record this. Look for the FENNEC_WAS_KILLED histogram in the telemetry dashboard under the Fennec app.

Comment 2

[Benoit Jacob [:bjacob] (mostly away)]

That would only tell us if we were killed by the OOM killer. It would not tell us if we crashed as a result of malloc returning null and our code mishandling this null pointer.

Comment 3

[Benoit Jacob [:bjacob] (mostly away)]

Also, it is useful to have this data in crash reports rather than in telemetry, as this is details allowing us to understand a given crash.

Comment 4

[Benoit Jacob [:bjacob] (mostly away)]

By the way here is the data:
http://people.mozilla.org/~bjacob/gfx_features_stats/#os-market-share

Comment 5

[Scoobidiver (away)]

The daily crash ratio per 100 ADUs is 0.9% on desktop and 2.5% on Android for Release versions.
The daily plugin hang ratio per 100 ADUs is 0.5% on desktop.
So Firefox for Android is almost three time crashier than desktop Firefox or twice if you count plugin hangs.

According to me, most crashes are caused by Flash for Android which requires a powerful device.
It would be interesting to know the mobile daily crash ratio with (70%?) and without (30%?) Flash in the stack trace.

Comment 6

[Benoit Jacob [:bjacob] (mostly away)]

In fact, as far as I understand, there is no need to look at the stacks to tell Flash crashes: see bug 797068 comment 37: since Flash accounts for almost 100% of plugin crashes, it should be enough to check for plugin crashes. Those have 'plugin' in the CSV files.

Comment 7

[Scoobidiver (away)]

I doubt Process Type exists on Android. Compare the crash report header of a Flash crash on Android, bp-4a4fc1ce-1c9a-48f2-94d8-9f89b2121114, and a one on Windows, bp-7da7f30e-f41a-4e7c-9eb0-b10882121114.

Comment 8

[BMO Automation]

We have completed our launch of our new Firefox on Android. The development of the new versions use GitHub for issue tracking. If the bug report still reproduces in a current version of [Firefox on Android nightly](https://play.google.com/store/apps/details?id=org.mozilla.fenix) an issue can be reported at the [Fenix GitHub project](https://github.com/mozilla-mobile/fenix/). If you want to discuss your report please use [Mozilla's chat](https://wiki.mozilla.org/Matrix#Connect_to_Matrix) server https://chat.mozilla.org and join the [#fenix](https://chat.mozilla.org/#/room/#fenix:mozilla.org) channel.