When I’m looking at the usage of our tempo compactors, some of them are idling, while some of them have almost 100% of CPU usage:
I hope that the load distribution could be distributed more evenly by tweaking the compactor configuration.
If I would change max_block_bytes, or compaction_window- would it force the compactor to ingest smaller blocks, and thus would better split the work between the instances?
Our current configuration looks like:
compactor:
compaction:
# 30 days
block_retention: 720h
# 500MB max block bytes
max_block_bytes: 500_000_000
compaction_window: 1h
compaction_cycle: 30s
ring:
kvstore:
store: memberlist