[
https://issues.apache.org/jira/browse/CASSANDRA-13643?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16072528#comment-16072528
]
Sylvain Lebresne commented on CASSANDRA-13643:
----------------------------------------------
As an aside, running from memory but I'm reasonably confident this is a
regression from 3.0 in that we use to do the transformation of 'expired cell'
to 'equivalent tombstone' in the deserialization code, with the purge code
running afterwards, so that if the generated tombstone is directly purgeable,
it would be purged right away. So I suppose the argument can be made that this
is a bug and hence putting it in 3.0/3.11. I don't have a huge preference
either way.
> converting expired ttl cells to tombstones causing unnecessary digest
> mismatches
> --------------------------------------------------------------------------------
>
> Key: CASSANDRA-13643
> URL: https://issues.apache.org/jira/browse/CASSANDRA-13643
> Project: Cassandra
> Issue Type: Bug
> Reporter: Blake Eggleston
> Assignee: Blake Eggleston
> Priority: Minor
>
> In
> [{{AbstractCell#purge}}|https://github.com/apache/cassandra/blob/26e025804c6777a0d124dbc257747cba85b18f37/src/java/org/apache/cassandra/db/rows/AbstractCell.java#L77]
> , we convert expired ttl'd cells to tombstones, and set the the local
> deletion time to the cell's expiration time, less the ttl time. Depending on
> the timing of the purge, this can cause purge to generate tombstones that are
> otherwise purgeable. If compaction for a row with ttls isn't at the same
> state between replicas, this will then cause digest mismatches between
> logically identical rows, leading to unnecessary repair streaming and read
> repairs.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
