Investigate into null values in startup metrics

[rocketsroger]

Currently there are large number of startup metrics are reported as null. Investigate into how startup metrics are reported to understand the cause/solution of the issue.

Example query: https://sql.telemetry.mozilla.org/queries/79505/

┆Issue is synchronized with this Jira Task

[wlach]

@rocketsroger when you've diagnosed this issue and have a fix, could you also provide a list of affected metrics? This would be a handy thing to have in the commentary section in the glean dictionary so people looking at the data are aware of this issue / gotcha. See the commentary section here, for example: https://dictionary.protosaur.dev/apps/firefox_ios/metrics/app_opened_as_default_browser

[rocketsroger]

Glean team understand the current limitation and has opened an issue for this as well. https://bugzilla.mozilla.org/show_bug.cgi?id=1706637

[rocketsroger]

As a workaround suggestion in https://bugzilla.mozilla.org/show_bug.cgi?id=1706637. I will move any startup metrics operations to right after Glean initialization. I will create another issue to track the follow up.

[rocketsroger]

follow up issue created: #19251

[rocketsroger]

While moving the implementation I've noticed that it is using incorrect types. Found a issue tracking this #12131.

[mcomella]

I analyzed the perf impact of this and it's very large: 114ms or -8.07% in time to visual completeness on the Moto G5. This is a very large regression so I don't think we should keep this change and have recommendations of alternative implementations: #19395 (comment)

[rocketsroger]

I analyzed the perf impact of this and it's very large: 114ms or -8.07% in time to visual completeness on the Moto G5. This is a very large regression so I don't think we should keep this change and have recommendations of alternative implementations: #19395 (comment)

Agreed. I was expecting performance degradation but I didn't think it was this much. Since this is a workaround until Glean team's investigation. I will create a PR to move the metrics setup off the main thread. If moving metrics setup off the main thread causes the workaround to not work consistently then there's nothing we can do.

[rocketsroger]

Looking at the query we have about 20% null values still. This is probably the best we can do without affecting performance. Once Glean team's investigation is done I will use the follow up issue #19251 to fix the root cause. Closing this as done.