[
https://issues.apache.org/jira/browse/CASSANDRA-872?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12843858#action_12843858
]
Stu Hood commented on CASSANDRA-872:
------------------------------------
> For instance a node that is partitioned is indistinguishable from dead.
If two live nodes have the same token after a partition, I think fencing
(killing) one of them is still a reasonable solution. If both nodes think they
are responsible for the same range, you end up with a split brain situation
where both will accept reads and writes, and the operator may never notice that
they were misconfigured.
At the absolute minimum, could we have one node leave gossip and log an ERROR?
> Changing the ip address associated w/ a token should generate a warning
> -----------------------------------------------------------------------
>
> Key: CASSANDRA-872
> URL: https://issues.apache.org/jira/browse/CASSANDRA-872
> Project: Cassandra
> Issue Type: Improvement
> Components: Core
> Affects Versions: 0.6
> Reporter: Stu Hood
> Fix For: 0.7
>
>
> Currently, multiple nodes can join the ring with the same token (particularly
> when people copy the data directory between machines). The symptom is that
> the new endpoint joins the ring successfully, but both nodes can only see
> themselves in the ring (according to nodetool).
> Instead of this behaviour, we would want gossip to die rather than joining
> with a duped token.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
