Closed mcomella closed 8 months ago
If it's helpful, there are several other probes that arrive in the same ping:
Thank you for filing this! It turns out that the ETL is missing timestamps during the initial stages. I may be able to turn this on within the next week, but backfilling these aggregates is a challenge (at least on release). I also filed https://github.com/mozilla/bigquery-etl/issues/1729 in bigquery-etl.
I've dug into this even deeper -- the primary reason that data within the startup-timeline
ping is missing is because the ping does not have a client id. This causes issues with aggregation because GLAM is assumes that every metric is attached to a client id, in order to perform client-level normalization before aggregating. Rows without client-ids are currently being filtered out.
One hack would be to assign client id-less pings with a default value so they are aggregated into a single row. Another would be to assign the client-id to the document id. The better solution is to assign a random client id from a fixed set. The best solution would be to enforce a client id for consistent aggregates (although there is an argument to be made about lean data practices and not collecting a client id).
@mcomella do you have context on why the startup-timeline ping doesn't include client ids?
I'm reading through some of the documentation, it looks like this ping is generally meant for automation, but it does look like it's being sent in quite a few pings.
@mcomella do you have context on why the startup-timeline ping doesn't include client ids?
@acmiyaguchi Probably because I didn't know any better 😓 I can't remember exactly but I'd speculate that I didn't think I'd need it to analyze the data in this ping but thinking about it now, it's useful to answer some questions like, "Can all the outliers be caused by an extremely small proportion of users?"
it looks like this ping is generally meant for automation,
For performance, we want to be able to extract our timing in CI and in user telemetry using the same logic, to ensure they line up so this is actually intended for both use cases (though in practice we haven't added any automation for this one). Sorry if the docs are unclear.
We discussed that this might take a little while to fix so I'll change the probe implementation to get the data faster and solve the problem with the lowest effort solution: https://github.com/mozilla-mobile/fenix/issues/17972
The GLAM page for the
startup_timeline_framework_start
probe – https://glam.telemetry.mozilla.org/fenix/probe/startup_timeline_framework_start/explore?ping_type=*&ref= – displays "404: No data found for the selected dimensions." However, this probe should have data – I wrote queries for it 3 months ago.