You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@cassandra.apache.org by "Cyril Scetbon (JIRA)" <ji...@apache.org> on 2014/03/13 17:11:51 UTC
[jira] [Commented] (CASSANDRA-6852) can't repair -pr part of data
when not replicating data everywhere (multiDCs)
[ https://issues.apache.org/jira/browse/CASSANDRA-6852?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13933466#comment-13933466 ]
Cyril Scetbon commented on CASSANDRA-6852:
------------------------------------------
I don't think the original JIRA can really help me :(
I understand that you chose to not repair -pr on a node not storing data for one keyspace. However here is my problem or show me I'm wrong: http://pastebin.com/VrDnWvAu
So the incriminated node owns 10.1% of the tokens which seem to be responsible for rows in keyspace ks1. However as ks1 is not replicated to that node, how can I repair it without a FULL REPAIR on all nodes ?
At the beginning I thought like [~jjordan] when I read his [comment|https://issues.apache.org/jira/browse/CASSANDRA-5424?focusedCommentId=13673260&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-13673260]. However I can see that tokens on the incriminated node (not replicating data for ks1) is referring rows ... I suppose it's what we call primary ranges and if not why do I find ks1's data for those tokens ?
> can't repair -pr part of data when not replicating data everywhere (multiDCs)
> -----------------------------------------------------------------------------
>
> Key: CASSANDRA-6852
> URL: https://issues.apache.org/jira/browse/CASSANDRA-6852
> Project: Cassandra
> Issue Type: Bug
> Components: Core
> Reporter: Cyril Scetbon
> Labels: multi-dcs, ranges, repair
>
> Our environment is as follows :
> - 3 DCS : dc1,dc2 and dc3
> - replicate all keyspaces to dc1 and dc2
> - replicate a few keyspaces to dc3 as we have less hardware and use it for computing statistics
> We use repair -pr everywhere regularly. FYI, a full repair takes almost 20 hours per node. The matter is that we can't use "repair -pr" anymore for tokens stored on dc3 concerning keyspaces not replicated. We should have a way to repair those ranges without doing a FULL REPAIR everywhere
--
This message was sent by Atlassian JIRA
(v6.2#6252)