[
https://issues.apache.org/jira/browse/CASSANDRA-1118?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12874278#action_12874278
]
Gary Dusbabek commented on CASSANDRA-1118:
------------------------------------------
Basically for my notes so I don't forget this...
This feature is already implemented, but I see two bugs. First, there is a
flaw in the state logic that assumes any time a node is seen for the first time
that is the canonical host for a given token. It is manifest when
NodeB(tokenX) comes online intending to replace NodeA(tokenX). when B sees A
for the first time during the normal course of gossip, B incorrectly assumes
that A is replacing it, which messes up B's view of the ring. Second,
partitions are not addressed (Node C, returning from a partition will not see
that B has replaced A) and will have an incorrect view of the ring.
> allow overriding existing token owner with a new IP
> ---------------------------------------------------
>
> Key: CASSANDRA-1118
> URL: https://issues.apache.org/jira/browse/CASSANDRA-1118
> Project: Cassandra
> Issue Type: Bug
> Components: Core
> Reporter: Jonathan Ellis
> Assignee: Gary Dusbabek
> Priority: Minor
> Fix For: 0.6.3
>
>
> We'd like to support replacing one node with another at the same IP (e.g.
> when the data is on Amazon's EBS and can easily be mounted to a new host), as
> noted in CASSANDRA-872. But in practice this is reported to not work w/o a
> cluster restart (can't find the ML thread now ... ?)
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
