revision based auto-compaction-retention

[clusc]

Hi,

Currently we are using the auto-compaction-retention parameter to compact etcd data every 24h. But given our spiky usage pattern(we are using etcd as distributed lock), this time based compaction policy doesn't work too well for us. Is there a revision-based auto-compaction policy that we can configure? Also, is there a metric for # of revision that we can watch on(I didn't find in the current exposed metrics)?

Thanks

[xiang90]

It makes sense. We will probably add it.

[clusc]

Thanks. Is there a workaround that we can do now to alleviate us from manually compacting every now and then? We are using etcd-operator, is this something that operator can help?

[xiang90]

@clusc Not really. It is all manual at the moment.

[xiang90]

@clusc

is there a metric for # of revision

revision is attached in every response header.

[clusc]

It doesn't for /metrics endpoint though right?

[xiang90]

@clusc No. You can add a debugging metrics for it if you wish. Open a pr, thanks.

[yudai]

I had the same idea and just crated a PR for it.
Still, compaction happens every 5 minutes, but I believe it should cover almost all environments.

[xiang90]

fixed by #8123