1304790 - Add a telemetry probe that measures how long it takes for a content process to start up and communicate with the parent

Status: RESOLVED FIXED

(Core :: DOM: Content Processes, defect, P2)

Description

[Mike Conley (:mconley) (:⚙️)]

This would be very useful for determining if slow content process start time is related to bug 1304393.

Comment 1

[Mike Conley (:mconley) (:⚙️)]

ni'ing as requested.

Comment 2

[Benjamin Smedberg]

Attachment: Bug 1304790 - Add a telemetry probe that measures how long it takes for a content process to start up and communicate back to the parent, data-r?francois

Review commit: https://reviewboard.mozilla.org/r/97868/diff/#index_header
See other reviews: https://reviewboard.mozilla.org/r/97868/

Comment 3

[Mike Conley (:mconley) (:⚙️)]

Comment on attachment 8817649 [details]
Bug 1304790 - Add a telemetry probe that measures how long it takes for a content process to start up and communicate back to the parent,  data-r?francois

https://reviewboard.mozilla.org/r/97868/#review98214

Excited (and maybe a little worried) to see the data here. Thanks!

::: toolkit/components/telemetry/Histograms.json:196
(Diff revision 1)
> +    "expires_in_version": "57",
> +    "bug_numbers": [1304790],
> +    "kind": "exponential",
> +    "high": 64000,
> +    "n_buckets": 100,
> +    "releaseChannelCollection": "opt-out",

I guess we'll need a test for this in a follow-up bug?

::: toolkit/components/telemetry/Histograms.json:197
(Diff revision 1)
> +    "bug_numbers": [1304790],
> +    "kind": "exponential",
> +    "high": 64000,
> +    "n_buckets": 100,
> +    "releaseChannelCollection": "opt-out",
> +    "description": "Content process launch time until the GetXPCOMProcessAttributes message is received, from  in milliseconds"

Typo?: "from  in milliseconds"

Comment 4

[Benjamin Smedberg]

Yes, I'll add a test if this metric shows value and we decide to keep it permanently.

Comment 5

[François Marier [:francois]]

Comment on attachment 8817649 [details]
Bug 1304790 - Add a telemetry probe that measures how long it takes for a content process to start up and communicate back to the parent,  data-r?francois

https://reviewboard.mozilla.org/r/97868/#review98246

This is providing user value in the form of monitoring product quality issues.

datareview+

Comment 6

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/mozilla-inbound/rev/f00c6d7631bb
Add a telemetry probe that measures how long it takes for a content process to start up and communicate back to the parent, r=mconley data-r=francois

Comment 7

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/f00c6d7631bb

Comment 9

[(no longer active)]

You will be glad to look at how this telemetry probe has been doing: https://health.graphics/quantum/track?metric=CONTENT_PROCESS_LAUNCH_TIME_MS

Something landed in February which made things an order of magnitude better.

Comment 10

[Bill McCloskey [inactive unless it's an emergency] (:billm)]

I don't think that's real. It used to measure how long it would take for the content process to start up and send its first message to the parent. Now it just measures launch time. Mike pointed this out in https://bugzilla.mozilla.org/show_bug.cgi?id=1303096#c22, but somehow that comment was never addressed (partly my fault).

We should probably base this off some other message, or maybe create a new message just for this purpose. I filed bug 1361939 for this.

Comment 11

[(no longer active)]

Thanks for the explanation, makes sense!