Ariel Weisberg created CASSANDRA-14665:
------------------------------------------

             Summary: Support monotonic reads with Transient Replication
                 Key: CASSANDRA-14665
                 URL: https://issues.apache.org/jira/browse/CASSANDRA-14665
             Project: Cassandra
          Issue Type: Sub-task
            Reporter: Ariel Weisberg
             Fix For: 4.x


When transient replication is in use blocking read repair currently doesn't 
repair transient replicas. While this results in transient replicas being 
almost free as they don't have to accept 99% of write traffic and the 
associated data it does mean that values returned by a read might not be 
replicated to a majority of nodes.

If we enabled BRR to transient replicas we would get monotonic reads, but we 
would have to BRR repaired and unrepaired data to them. This means we generally 
wouldn't want to read from them to avoid extra BRR and if nodes are down and we 
were forced to BRR them they could acquire a large amount of repaired data that 
would then have to be streamed at the next incremental repair.

If we split read results into repaired and unrepaired data it will be possible 
to only blocking read repair unrepaired data to transient replicas. Then we can 
enable BRR for transient replicas without running into performance degradation 
when nodes are down.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to