nodetool compact manually triggers a major compaction for selected ScyllaDB data. Unlike normal minor compactions, it can target a table or token range and pushes the selected data to the highest tier or level of its compaction strategy. section 9.3.3 section 8.4.1
Use the command carefully because it has immediate operational consequences:
- Expect significant resource consumption while the major compaction runs. section 9.3.3
- Consider it when obsolete data or many tombstones may be hurting partition-read latency. section 8.4.1
- Avoid treating it as routine maintenance: placing old data at the highest tier can make later minor compactions less effective. section 9.3.3 section 8.4.1
Checking before intervening
An operator first uses nodetool compactionstats to see which keyspaces and tables are being compacted and whether a backlog exists. If compaction activity correlates with slow reads, the operator can investigate a specific table or token range before deciding whether nodetool compact is justified. section 9.3.3 section 8.4.1
