Description
Describe the bug
We have a tenant with out of order ingestion enabled. One day, the tenant stopped ingestion completely. However, its active series count still remain > 0 and it never went down even though there is no ingestion at all.
Expected behavior
When a TSDB becomes idle, Cortex Ingester will force a head compaction after a configured idle timeout. This should clean up the active series in memory.
I think the issue is that in Ingester compactHead function, Ingester uses TSDB CompactHead to perform head compaction. When out of order is enabled, there is a new function introduced called CompactOOOHead which specifically compacts the head for out of order samples and write behind log (WAL for OOO data). We should extend Ingester to also call CompactOOOHead to flush OOO data from memory.