1376589 - Add Crashes (M+C-S) to Experiments Aggregates

Status: RESOLVED FIXED

(Data Platform and Tools :: General, enhancement, P2)

Description

[Frank Bertsch [:frank]]

This will be a scalar value. We will read directly from the raw data (no interim dataset yet).

Main Crashes: Any crash pings with processType = main
Content Crashes: Any crash pines with processType = content

We will ignore all Shutdown crashes.

See the crash aggregates view for exactly how to parse these fields.

Comment 1

[François Marier [:francois]]

As discussed in person, we would be interested in seeing the rate of shutdown crashes too, ideally, separate from the other ones.

Comment 2

[François Marier [:francois]]

Do we have any way to query and compare the crash rates between the two conditions of an opt-out experiment?

We are going to need this for bug 1377267.

Comment 3

[Frank Bertsch [:frank]]

Immediately, you can do analysis in Spark on each branch, and count the crashes. Crash pings give main process crashes, and all other process crashes are recorded in the Main ping.

Next week we will have crashes available in error_aggregates. This will be queryable in re:dash, and you can compare across branches very easily.

Within two weeks it will be showing in the experiments viewer.

Comment 4

[François Marier [:francois]]

Great!

Comment 5

[François Marier [:francois]]

Is this now available from the experiments viewer?

Comment 6

[Frank Bertsch [:frank]]

(In reply to François Marier [:francois] from comment #5)
> Is this now available from the experiments viewer?

No, we're running into issues deploying the changes to error_aggregates. You can follow along in bug 1388351.

Comment 7

[Frank Bertsch [:frank]]

I've given up on spark-streaming for now. The new implementation will be to add experiment annotations to CrashSummary, and read the crashes from there for the experiments viewer.