feat(2.53.5): opt-in fix for export.match from config; ensure ApplyConfig before export

[bwplotka]

Implements step 1 from go/gmp:matchstuck

NOTE: This PR also expects ApplyConfig to be triggered for export to work. Otherwise it's noop (similar to disable export = true). IMO this is required for reliable startup behaviour for all options (especially matching). Imagine flag is setting = matchAll and config is setting match only "A". We don't want risk random metrics user don't expect (e.g. broken ones) due to startup races.

In practice Prometheus ensures ApplyConfig is triggered before DB is ready, but adding safecheck for other callers.

[pintohutch]

LGTM % questions/comments

We could also use a /prombench run to ensure we don't see any regressions in steady-state performance.

[pintohutch]

/prombench

[gmp-bot]

This benchmark run compares this PR (#259) with version v2.53.4-gmp.0-gke.1 of the GMP Collector.

After a successful deployment, the benchmarking results can be viewed on the Cloud Monitoring Dashboard.

To cancel or stop the benchmark run, comment /prombench cancel on this PR.

To view the results of the profiling process you can run the following commands. After running the commands, go to localhost:31802/profiles to view your profiling results after a few minutes.

gcloud container clusters get-credentials prombench --zone us-east4-a --project gpe-test-1
kubectl config use-context gke_gpe-test-1_us-east4-a_prombench
kubectl port-forward services/parca 31802:80

[pintohutch]

/prombench cancel

[pintohutch]

I couldn't see any results there... maybe our prombench setup needs fixing...

[gmp-bot]

Cancelling the benchmark run for this PR: #259.

[bwplotka]

I'd be very surprised if it worked given we didn't touch it for ~year.

It might be only one thing though:

However, @pintohutch should we spend time fixing it now vs progressing on matcher stuckness? Is there anything you are are worried about in terms of perf in this PR? I am not adding any state and we don't filter on prombench.

[bwplotka]

Thanks for review!

As per above comment, I am not fixing prombench, but instead prioritize further parts of go/gmp:matchstuck

[pintohutch]

Great conversations - some more feedback for you (apologies for the delay)

[bwplotka]

PTAL, this should be now good to @pintohutch

I did add one more tiny change as a safeguard, please check the description note: #259 (comment)

[pintohutch]

LGTM % nit

Thanks for doing this!