Marcus Eriksson created CASSANDRA-8858:
------------------------------------------
Summary: Avoid not doing anticompaction on compacted away sstables
Key: CASSANDRA-8858
URL: https://issues.apache.org/jira/browse/CASSANDRA-8858
Project: Cassandra
Issue Type: Bug
Reporter: Marcus Eriksson
Fix For: 3.0
Currently, if an sstable is compacted away during repair, we will not
anticompact it, meaning we will do too much work when we run the next repair.
There are a few ways to solve this:
1. track where the compacted sstables end up (ie, we compact and sstable 1,2
that are being repaired into sstable 3, we can anticompact sstable 3 once
repair is done). Note that this would force us to not compact newly flushed
sstables with the ones that existed when we started repair.
2. don't do compactions at all among the sstables we repair (essentially just
mark the as compacting when we start validating and keep them that way
throughout the repair)
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
