1289671 - [meta] [Uptime] Crash report submission

Status: RESOLVED INCOMPLETE

(Firefox :: General, defect, P3)

Description

[Nicholas Nethercote [inactive]]

This is a Project Uptime meta-bug for bugs aimed at increasing the crash submission rate.

Comment 1

[:Harald Kirschner :digitarald]

njn, just to understand what the baseline for this effort is - what is the current % of crash reports submitted after prompt?

Comment 2

[Nicholas Nethercote [inactive]]

I don't know. I think bsmedberg knows these numbers -- the rates vary significantly for different kinds of crashes.

Comment 3

[Benjamin Smedberg]

If we're resource-constrained, I don't think we should focus on this

Comment 4

[:Harald Kirschner :digitarald]

While prioritization of this would be an interesting discussion, I just wanted to understand if we have any data at all: > What is the current % of crash reports submitted after prompt?

Comment 5

[Benjamin Smedberg]

Sorry wasn't done. I think we should focus primarily on client stackwalking so that we don't have to ask users in general to submit crashes. We can still prompt when we need additional data, and even be more aggressive about it because it's less frequent.

That said, the way to compute submission rates is to compare the PROCESS_CRASH_SUBMIT_ATTEMPT histogram with the relevant "crashes detected" number. So for each type:

main process: PROCESS_CRASH_SUBMIT_ATTEMPT['main-crash'] / COUNT(crash pings)
content process: PROCESS_CRASH_SUBMIT_ATTEMPT['content-crash'] / SUBPROCESS_CRASHES_WITH_DUMP['content']
npapi and gmp: PROCESS_CRASH_SUBMIT_ATTEMPT['plugin-crash'+'plugin-hang'] / SUBPROCESS_CRASHES_WITH_DUMP['plugin'+'pluginhang'+'gmplugin']

I don't have a parquet dataset for this because we currently don't aggregate the submit-attempt or submit-success numbers. Here's an email from data I analyzed a while back, though:

> Main crashes are typically submitted at a rate of around 50% (slightly more on nightly; slightly less on release).
> 
> Content crashes are typically submitted at a rate between 10-15%, but the rate is more variable. It sometimes dips down to 5% and occasionally up to 20%. Here are some hunches for why this may be more variable:
> 
>     I believe (with some evidence) that people are more likely to submit new crashes, and less likely to submit crashes that they've already submitted.
>     I also expect (with little evidence) that a crash that happens when a tab or window is closed are much less likely to be submitted than crashes that happen while a page is active.
> 
> Plugin crashes are submitted at a rate between 0.5% and 0.75%.

Comment 6

[Nicholas Nethercote [inactive]]

(In reply to Benjamin Smedberg [:bsmedberg] from comment #5)
> I think we should focus primarily on client stackwalking
> so that we don't have to ask users in general to submit crashes. We can
> still prompt when we need additional data, and even be more aggressive about
> it because it's less frequent.

FWIW, this bug covers anything that might increase the crash submission rate, including the client-side stack walking work (which is among the blocking bugs).

Comment 7

[Stephen Donner [:stephend] Not actively reading bugmail]

Mass-removing myself from cc; search for 12b9dfe4-ece3-40dc-8d23-60e179f64ac1 or any reasonable part thereof, to mass-delete these notifications (and sorry!)

Comment 8

[Andrew McCreight [:mccr8]]

This project ended a long time ago.