[
https://issues.apache.org/jira/browse/CASSANDRA-2609?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Sylvain Lebresne updated CASSANDRA-2609:
----------------------------------------
Description:
Repair has no multi-DC awareness in that if you have 2 DC with 3 replica in
each, a repair of a node will transit 3 merkle tree cross-DC and potentially
initiate as many cross-DC streaming (with likely a non null intersection
between those). In theory, we could repair separately in each DC, then repair
between only two node cross-DC (that we know are up to date in their respective
DC) and finally re-repair in the separate DC (to pass the cross-DC changes to
all nodes of the DC).
It is yet unclear to me if we can make that efficient enough that it is worth
the added complexity, so this is really meant as an exploratory ticket.
was:
Repair has no multi-DC awareness in that if for instance you have 2 DC with 3
replica in each, a repair of a node will transit 3 merkle tree cross-DC and
potentially initiate as many cross-DC streaming (with likely a non null
intersection between those). In theory, we could repair separately in each DC,
then repair between only two node cross-DC (that we know are up to date in
their respective DC) and finally re-repair in the separate DC (to pass the
cross-DC changes to all nodes of the DC).
It is yet unclear to me if we can make that efficient enough that it is worth
the added complexity, so this is really met as an exploratory ticket.
> Repair multi-DC awareness
> -------------------------
>
> Key: CASSANDRA-2609
> URL: https://issues.apache.org/jira/browse/CASSANDRA-2609
> Project: Cassandra
> Issue Type: Improvement
> Components: Core
> Reporter: Sylvain Lebresne
> Priority: Minor
> Labels: ponies
>
> Repair has no multi-DC awareness in that if you have 2 DC with 3 replica in
> each, a repair of a node will transit 3 merkle tree cross-DC and potentially
> initiate as many cross-DC streaming (with likely a non null intersection
> between those). In theory, we could repair separately in each DC, then repair
> between only two node cross-DC (that we know are up to date in their
> respective DC) and finally re-repair in the separate DC (to pass the cross-DC
> changes to all nodes of the DC).
> It is yet unclear to me if we can make that efficient enough that it is worth
> the added complexity, so this is really meant as an exploratory ticket.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
