Open
Description
Proposal
Total number of series in memory has a direct correlation on the total memory being consumed. This is a problem when the series keep churning either because of pods either in crash loop backoff or when a user does a rollout that churned pods, did a roll back and then again rolled forward. this would 3x the total series in memory.
I would like to propose an "eager compaction of stale series" based on a flag that says "compact when total stale series in memory crosses 1M". this would greatly improve our ability to observe high churn series.