[
https://issues.apache.org/jira/browse/CASSANDRA-20967?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18058976#comment-18058976
]
Ariel Weisberg commented on CASSANDRA-20967:
--------------------------------------------
The new coordinated import functionality complicates this because the onus of
distributing the data is now on the database and it currently doesn't support
witnesses. If we copy the data to the witnesses it can overload them. There are
also complications with mutation tracking and backup of sstable or mutation
tracking log data. Does the backup of a witness log include the sstables from
the import? How does reconciliation for this work?
It really seems like for this to work well with witnesses we need to have the
sstables being imported stored in an external object store that is referenced
by mutation tracking so that nodes can grab that on demand.
> CEP-46: SSTable import support for witnesses
> --------------------------------------------
>
> Key: CASSANDRA-20967
> URL: https://issues.apache.org/jira/browse/CASSANDRA-20967
> Project: Apache Cassandra
> Issue Type: Sub-task
> Components: Feature/Lightweight Transactions
> Reporter: Ariel Weisberg
> Priority: Normal
>
> Mutation tracking added support for SSTable import where the mutation
> tracking log has an entry for the import action for the sstables that are
> staged on disk. This allows read repair to be done for reads that include the
> imported sstables.
> On witnesses this record needs to be aware of witnesses and validate that the
> witness does not import out of range data. This validation should occur both
> for the initial sstable important and processes that are staging the sstables
> and then also when applying the record out of the MT log.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
