Tsm compaction and influxd memory usage

How does periodic tsm compaction (cache, full, level 1,2,3) affect influxd memory usage?

Is there a way to tune parameters in influxdb.conf to reduce memory usage by internal influxd housekeeping or make it more predictable (for example, in terms of scheduling)? Inline comments in default influxdb.conf are too brief.

I’ve seen several out-of-memory crushes for influxd v1.2.0 and v1.2.4 recently and they seem to correlate with tsm compaction related lines in log.

There has been a lot of work on master to improve efficiency and resource scheduling during compaction events. You can see some of that in @jason’s recent PR: https://github.com/influxdata/influxdb/pull/8384

1 Like