1866357 - Better error telemetry

Status: NEW

(Application Services :: General, enhancement, P3)

Description

[Mark Hammond [:markh] [:mhammond]]

From github: https://github.com/mozilla/application-services/issues/4982.

Recently, we've been discussing how to improve how we use Glean to capture error telemetry. After taking a quick survey of our current error telemetry, I think our current code is actually close to what we want, we just need a few tweaks.

I think we should take the logins android error handling as a starting point:

• In metrics.yaml we define several metrics:
  • Total write query count
  • Total read query count
  • Read query error counts (this is a labeled counter, so we can create a count for each error type)
  • Write query error counts (also a labeled counter)
• In DatabaseLoginsStorage.kt we increment those counters
• Finally we graph the errors on our logins dashboard

I think we can get pretty good error telemetry with a few tweaks:

• Better metrics.
  • I don't think the read/write distinction is that useful, what if we replace that with just a total query count?
  • We use labeled counters to track error types, but we don't use that in the graph. What if we:
    • Combined read_query_error_count and write_query_error_count into a single errors_by_type labeled counter.
    • Visualize that on the dashboard as errors per day, grouped by type, like we do with sync errors
    • Improve the error type detection. Right now, almost all errors are grouped under the __other__ type. Getting access to Glean from Rust would be great, since then we could have this code in Rust.
  • Add an errors_by_function labeled counter and visualize that in a similar way. This could help us track down which function was generating errors. It seems like an improvement on the read/write distinction to me.
• Use this system for other components as well.
  • I think this means a bunch of similar metrics.yaml files per-component
  • Maybe the upcoming struct metric could reduce the duplication?
  • Create a global errors_by_component labeled counter. This would provide a nice overview for our main dashboard.
• Create a shared metric system. We should track the same metrics on iOS (and desktop once we're there). Getting access to Glean from Rust would be a huge help here too.

┆Issue is synchronized with this Jira Task

Change performed by the Move to Bugzilla add-on.