1131836 - Data output, new telemetry/FHR pipeline: Firefox browser crashes

Status: RESOLVED FIXED

(Cloud Services Graveyard :: Metrics: Pipeline, enhancement, P2)

Description

[Benjamin Smedberg]

Tracking re-implementation of engineering dashboards using the new pipeline:

This data tracks:

* the number of Firefox browser crashes over the total session length
* the number of Firefox browser crashes over the total number of sessions (% of session that ended in a crash)

The data should be separated by release channel, and then aggregated by buildid and crash date.

The graph UI will then let engineers and QA see the crash rate by date and by build-date.

Desired update frequency: daily at a minimum, but it would be lovely to have this hourly and perhaps have alerting about changes in the crash rate.

Comment 1

[Mike Trinkala [:trink]]

I thought we got a new session with each browser restart so wouldn't the first bullet item always be 0 or 1?

Comment 2

[Mike Trinkala [:trink]]

Never mind:  total crashes / uptime for everyone.  Uptime being in hours like the other reports I assume.

Comment 3

[Benjamin Smedberg]

Per-hour is fine, we can unit-factor the metric later if appropriate.

Comment 4

[Mike Trinkala [:trink]]

Please include some sample data for a crash entry.

Comment 5

[Benjamin Smedberg]

We don't have a design for the browser crash data yet. That is covered by bug 1121013.

We do have aborted sessions when bug 1133536 lands, but I don't actually docs on the exact data format. Alessio, can you include docs in your patch and maybe also a summary here?

Comment 6

[Alessio Placitelli [:Dexter]]

(In reply to Benjamin Smedberg  [:bsmedberg] from comment #5)
> We do have aborted sessions when bug 1133536 lands, but I don't actually
> docs on the exact data format. Alessio, can you include docs in your patch
> and maybe also a summary here?

"aborted-session" ping is a "main-ping" with a reason field set to "aborted-session". The documentation about the main-ping can be found at [1] (or in tree at [2]). I've added the missing documentation to bug 1133536.

[1] - http://gecko.readthedocs.org/en/latest/toolkit/components/telemetry/telemetry/main-ping.html
[2] - https://hg.mozilla.org/mozilla-central/annotate/c5b90c003be8/toolkit/components/telemetry/docs/main-ping.rst

Comment 7

[Mark Reid [:mreid]]

Sounds like we could get a reasonable estimate of sessions ending in a crash by comparing reason=shutdown with reason=aborted-session.

Comment 8

[Benjamin Smedberg]

The data from bug 1121013 is the right way to main-process crashes. It will also be valuable to be able to compare the crash rate with the aborted-session rate, since there are other reasons for an aborted session (Windows logout probably causes an aborted session, for instance).

Comment 9

[Benjamin Smedberg]

Bug 1121013 has now landed, and the in-tree docs are updated. I filed bug 11748871 to retain the new ping type for crashes. This should be unblocked now. Please let me know if you have questions about how to interpret the incoming data.

Comment 10

[Mike Trinkala [:trink]]

https://github.com/mozilla-services/data-pipeline/pull/88

Comment 11

[Mark Reid [:mreid]]

r+