1561898 - GC_MARK_ROOTS_MS and GC_MARK_GRAY_MS telemetry are misconfigured

Status: RESOLVED FIXED

(Core :: JavaScript: GC, defect, P5)

Description

[Jon Coppeard (:jonco)]

These telemetry histograms are configured as linear histograms and that means the results we get are not very precise. For example, for the former > 90% of the results fall into the first two buckets:

https://telemetry.mozilla.org/new-pipeline/dist.html#!cumulative=0&end_date=2019-06-26&include_spill=0&keys=__none__!__none__!__none__&max_channel_version=nightly%252F69&measure=GC_MARK_ROOTS_MS&min_channel_version=nightly%252F68&processType=*&product=Firefox&sanitize=1&sort_by_value=0&sort_keys=submissions&start_date=2019-05-20&table=0&trim=1&use_submission_date=0

A warning to the right reads:

Notice percentiles are estimated based on values in the histogram. Values are only guaranteed to be accurate to the nearest bucket.

Preferably these would be configured as exponential histograms, which would mean there were more buckets concentrated low in the range where most sample values are and fewer higher up where there are very few.

For comparison see:

https://telemetry.mozilla.org/new-pipeline/dist.html#!cumulative=0&end_date=2019-06-26&include_spill=0&keys=__none__!__none__!__none__&max_channel_version=nightly%252F69&measure=GC_MARK_MS&min_channel_version=nightly%252F68&processType=*&product=Firefox&sanitize=1&sort_by_value=0&sort_keys=submissions&start_date=2019-05-20&table=0&trim=1&use_submission_date=0

Comment 1

[Matthew Gaudet (he/him) [:mgaudet]]

GC_MARK_ROOTS_MS seems to have been replaced by GC_MARK_ROOTS_US, which doesn't have this issue

GC_MARK_GRAY_MS seems to have been replaced by GC_MARK_GRAY_MS_2, which does seem to still have this issue.

Comment 2

[Jon Coppeard (:jonco)]

Both GC_MARK_ROOTS_US and GC_MARK_GRAY_MS_2 are exponential so this bug is fixed.