compact: add label when no labels left after deduplication

[sepich]

• I added CHANGELOG entry for this change.
• Change is not relevant to the end user.

Changes

Fixes #2550
Per #1014 (comment) i'm adding new flag --deduplication.label-if-empty=label=value to handle situation, when there are no labels left after deduplication.

Verification

Basic tests added.
Tested manually on our data, collected by 3 thanos-receive with only --label=replica="$(NAME)"

[kakkoyun]

Overall looks great. Thanks for fixing it.

However, to simplify flow rather than introducing a new CLI flag, could we just add an automated external label? Something like given_replica_label=compacted|deduped or constant_replica_label=compacted=deduped. WDYT?

[sepich]

Sure, will rewrite then to use first label name from list set in --deduplication.replica-label, and then automatically add first_label_name=deduped in this case.
I think it is better than constant_replica_label, because user probably do not have this constant in --query.replica-label on querier. Which will lead to metrics have additional label at some point in time (corresponding to compaction start)

[bwplotka]

hm.. It's brutal but think it makes sense. Thanks!

Still no unique external labels will affect people. I would also add leve.Warn log line mentioning this. WDYT? (:

Thanks! looking forward to update and then we can merge for 0.13 release.

[bwplotka]

Good idea @kakkoyun !

[sepich]

I can add level.Warn log, but for us it is usual case - so this would just spam the logs.
Could you please clarify how it "affect people", so we consider adding additional label to current replica.

[bwplotka]

I can add level.Warn log, but for us it is usual case - so this would just spam the logs.

Really? It's every compaction so it's fine. And also you should have unique label per each "stream" as it will easier to scale to more streams some day. That's why log .

[kakkoyun]

lgtm

[bwplotka]

Thanks!