feat: aggregate metrics back into Loki #13621
Conversation
* ignore aggregated metric streams from limits
github-actions
bot
added
the
type/docs
pkg/detection/detection.go
Outdated
| ) | ||
|
|
||
| const ( | ||
| AggregatedMetricLabel = "__aggregated_metric__" |
There was a problem hiding this comment.
I'm open to other names. Maybe __rollup__? or __service_rollup__ in case we want to add other rollups in the future?
pkg/pattern/metric/push.go
Outdated
| case <-p.quit: | ||
| cancel() | ||
| return | ||
| // TODO(twhitney): could buffer these into fewer pushes? |
There was a problem hiding this comment.
We can't push line by line !
There was a problem hiding this comment.
agreed, we need to buffer these and push once every 10s. though it's not currently line by line, but it is 1 per stream every 10s.
There was a problem hiding this comment.
we also need to increment a single int64 for each push, rather than storing all the raw samples. I can make both those changes here, or follow up in a subsequent PR?
There was a problem hiding this comment.
another option would be to use something like a buffered channel and call the API when the channel is full?
|
closing in favor of #13731 |
What this PR does / why we need it:
This feature persists aggregated metrics back into Loki as a special stream that can be queried from Explore Logs
Which issue(s) this PR fixes:
Fixes #13110