1790568 - Large change in gc_slice_was_long telemetry on 22nd July

Status: RESOLVED FIXED

(Core :: JavaScript: GC, defect, P2)

Description

[Jon Coppeard (:jonco) (PTO until 5th Jan)]

This increased from ~1% to 7% of collections. Other telemetry doesn't seem to be affected.

This coincides with some refactoring of GC telemetry in bug 1776931 but I couldn't immediately see what the problem was.

Ted, any ideas what happened here?

Comment 1

[Jon Coppeard (:jonco) (PTO until 5th Jan)]

Attachment: Screenshot 2022-09-13 at 10-38-52 gc_slice_was_long GLAM.png

Comment 2

[BugBot [:suhaib / :marco/ :calixte]]

Set release status flags based on info from the regressing bug 1776931

Comment 3

[Ted Campbell [:tcampbell]]

Oops, I see the error. While we still report 'true' in all the same cases, my patch caused us to only report 'false' when we were over-budget. This caused the metric to change from "Fraction of TimeBudget-slices that were long" to "Fraction of slice-overruns that were long (rather than minor overruns)".

Comment 4

[Ted Campbell [:tcampbell]]

Attachment: Bug 1790568 - Record GC_SLICE_WAS_LONG for all TimeBudget slices. r?jonco

This fixes a regression in Bug 1776931 that unintentionally changed the
denominator to only includes over-budget slices. This patch restores the old
definition.

Comment 5

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/410e958aef39
Record GC_SLICE_WAS_LONG for all TimeBudget slices. r=jonco

Comment 6

[Ted Campbell [:tcampbell]]

Since we missed this for a while, no need to bother with uplifts. We can always correct for the error in redash if we needed precise data on this probe in the meantime.

Comment 7

[Marian-Vasile Laza]

https://hg.mozilla.org/mozilla-central/rev/410e958aef39