Continous Compaction on updates and slowing insert rates in other column families.

[Aravindhan1995]

The total write buffer size for RocksDB instance is 256 MB, there are four column familes , In one one column family there is continous updates (ie 3 lac updates in 5 mins for same keys ) . In other column families normal inserts and deletes. There is continous compaction in RocksDB for particular column family where update is taking place (every one min). This consumes lot of CPU resources. And this also affects other column family performance (The insert and delete taking more time).

Questions

1. Why compaction happening so frquently?
2. Why it affects inserts of other column famalies?

It's high priority issue, kindly reply asap

Thank you

[anand1976]

1. Compaction frequency is related to the write rate. The exact relationship depends on the configuration and the write pattern, but I'd expect atleast the lower levels to see frequent compactions if the write rate is high.
2. Is WAL enabled? If so, all writes to any column family will be written to the WAL first, unless explicitly disabled in the write request by setting write_options.disableWAL. If one of the column families has a heavy write load, it could potentially starve other column families.