Brian Gallew created CASSANDRA-13560:
----------------------------------------
Summary: Improved cleanup performance
Key: CASSANDRA-13560
URL: https://issues.apache.org/jira/browse/CASSANDRA-13560
Project: Cassandra
Issue Type: Improvement
Components: Compaction
Reporter: Brian Gallew
I've been thinking about sstables. One of their properties is that they are
sorted. In the face of that property, it would seem that the cleanup
functionality *should* be very fast as all of the partitions which no longer
belong to a given node should be in either one or two contiguous blocks of
space. Perhaps this is naive, but I would think the index should clearly
indicate what needs to be retained versus what can be disposed of, and thus a
cleanup should be able to start reading with the first valid partition, stop
with the last, and skip the bulk of loading/unloading that seems to be
happening.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org
For additional commands, e-mail: commits-h...@cassandra.apache.org
