Augustin Husson

> --- FAIL: TestQueryLog/rule_queries,_[::1]:50285,_enabled_at_start (6.58s) I don't see why this test is failing. Is it related to the changes ?

with @ahurtaud, we will load this PR in production environment to see if we have a drop of CPU usage.

I moved the PR to draft since a first world running doesn't show evidence of optimisation:  On the left, the usage of CPU when PR #10634 is deployed, then...

Soooo, Dropping the `LabelsBuilder` in favor of a map usage seems to work a bit. We gain 1 ~2 VCPUs on 8. On our side, having a logic to drop...

this PR is ready to be reviewed. At this point I don't know what we could improve more than what I found and fixed. Of course, someone with a better...

on the `Sync` method ?

I run a simple benchmark on the `Sync` method and here the result between the main branch (old time) and this PR (new time) ```txt go test -bench=Sync -count 5...

and when I'm running the same benchmark with the usage of the labelsBuilder in the method `buildTargetLabels`, here the result: ```txt go test -bench=Sync -count 5 -run=^# -benchmem | tee...

maybe the benchmark is too simple, I was more or less expected to see a difference in terme of CPU usage.

I added a suite of benchmark to see if the performances are stable when the number of targets are increasing. And it appears it is the case. `old.txt` is the...