feat(perf): Fewer tags on performance score metrics #2827
Conversation
| "browser.name": "Chrome", | ||
| "environment": "fake_environment", | ||
| "release": "1.2.3", | ||
| "transaction": "gEt /api/:version/users/", | ||
| "transaction.op": "mYOp", |
There was a problem hiding this comment.
Without the tag filter, this would include platform, dist, etc.
There was a problem hiding this comment.
@edwardgou-sentry @k-fish apart from the performance score UI, could there be a more generic measurements-related part of the UI that breaks if these metrics come with only a subset of tags?
There was a problem hiding this comment.
As far as I know, Dashboards is the only other place users would be able to write a query for Performance Scores using metrics. But I'm pretty sure that if a metrics query breaks due to missing tags, it will fall back to the indexed dataset instead. I think this behaviour should be ok
| let is_performance_score = name == "score.total" | ||
| || name | ||
| .strip_prefix("score.weight.") | ||
| .or_else(|| name.strip_prefix("score.")) | ||
| .map_or(false, |suffix| measurement_names.contains(suffix)); |
There was a problem hiding this comment.
This is clumsy: It would be better to mark measurements as "performance score" somehow, but they might originate from a downstream relay so any kind of flag would have to be serialized into the event JSON.
@k-fish good point. Could we build this feature by just using the "counter" metric type (which supports floats)? |
|
@jjbayer yes, it could be count (sum) and count (count) then divide them. |
Performance score measurements are special measurements derived by relay. As such, they do not require the full set of metric tags that SDK-side measurements get.
Limiting the number of tags will reduce cardinality by ~20%, according to my analysis.
#skip-changelog