Tom van der Woerdt created CASSANDRA-15836:
----------------------------------------------
Summary: nodetool garbagecollect could leave LCS level intact
Key: CASSANDRA-15836
URL: https://issues.apache.org/jira/browse/CASSANDRA-15836
Project: Cassandra
Issue Type: Improvement
Reporter: Tom van der Woerdt
The nodetool command `garbagecollect` will run a single-sstable compaction for
every sstable in a cf, while using other sstables on the side to allow for
dropping tombstoned data. However, in doing so, it resets all LCS levels back
to 0, causing a significant write amplification.
Fundamentally there's no reason why LCS levels have to be changed here, since
these are single-sstable compactions. When the expected reduction in data set
size is small (say, 10%) it may be preferable for sstables to remain in place
instead of having to go through the entire compaction process again.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org
For additional commands, e-mail: commits-h...@cassandra.apache.org
