1807327 - setStartupMetrics perf fix is still a regression to COLD VIEW start up perf (~53ms)

Status: RESOLVED WONTFIX

(Firefox for Android :: Performance, defect)

Description

[Chris Peterson [:cpeterson]]

From github: https://github.com/mozilla-mobile/fenix/issues/20346.

We landed https://github.com/mozilla-mobile/fenix/issues/19147 to move setStartupMetrics to address a data quality issue and we saw a large regression of ~114ms on the Moto G5 https://github.com/mozilla-mobile/fenix/issues/19395#issuecomment-832873039. We landed a performance fix for this issue in https://github.com/mozilla-mobile/fenix/pull/19397#issuecomment-833965318 and I remeasured it https://github.com/mozilla-mobile/fenix/issues/19395#issuecomment-853448383 to find it still regressed performance over the original implementation: 53ms on the Moto G5.

This bug is to fix the 53ms performance regression introduced by setStartupMetrics and its performance fix. To reproduce the behavior, you can use my measurement notes from the prior comment:

To measure this without interference from other code changes, I checked out the commit before the regression 2bd933054f0152fbc4bc3d3f2f276d9c41e42e08, checked out the commit after the regression 74047cdceb16990a6a843ce2824ca5b7696fb0d3, and cherry-picked the fix onto that commit, which applied cleanly (this change is available on my branch).

┆Issue is synchronized with this Jira Task

Change performed by the Move to Bugzilla add-on.

Comment 1

[Chris Peterson [:cpeterson]]

It's too late to debug a startup time regression from July 2021. That code has changed a lot since then. We need to profile and optimize Fenix's current startup behavior.