Overview
RocksDB compaction in Ceph is a process aimed at reducing fragmentation in the RocksDB database and improving the performance of daemons like OSDs and MONs. Over time, as clusters handle a large number of I/O operations, RocksDB can accumulate fragmentation, leading to higher latencies.
Takeaways
- Compaction ensures that RocksDB remains performant over time.
- Compaction is applicable to both MONs and OSDs, but it is most commonly required for OSDs because they handle significant I/O workloads.
- Compaction can be triggered manually or automatically during daemon restart.
- The
osd_compact_on_startflag enables compaction during an OSD daemon restart.
Configuration
To enable RocksDB compaction during OSD daemon restarts, add the following to the ceph.conf file under the [osd] section:
[osd]
osd_compact_on_start = trueThis configuration ensures that compaction occurs each time the OSD daemon restarts, which is particularly useful for clusters that experience prolonged operation without regular maintenance.
Best Practices
- Enable compaction if your cluster shows signs of RocksDB fragmentation, such as slower OSD operations or higher latencies.
- Regularly restart OSDs with compaction enabled to maintain database health.
Version Applicability
The osd_compact_on_start flag is available starting from Ceph Pacific. Ensure that your cluster is running a compatible version before applying this configuration. 1