[
https://issues.apache.org/jira/browse/CASSANDRA-15133?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16844502#comment-16844502
]
Jay Zhuang commented on CASSANDRA-15133:
----------------------------------------
Here is a purposed fix:
| [15133-trunk|https://github.com/cooldoger/cassandra/tree/15133-trunk] |
[!https://circleci.com/gh/cooldoger/cassandra/tree/15133-trunk.svg?style=svg!|https://circleci.com/gh/cooldoger/cassandra/tree/15133-trunk]
|
And also it fixes a removing {{movingEndpoint}} issue, as the following code
never works, removing item while looping the collection will throw
`ConcurrentModificationException`:
{noformat}
for (Pair<Token, InetAddressAndPort> pair : movingEndpoints)
{
if (pair.right.equals(endpoint))
{
movingEndpoints.remove(pair);
break;
}
}
{noformat}
> Node restart causes unnecessary token metadata update
> -----------------------------------------------------
>
> Key: CASSANDRA-15133
> URL: https://issues.apache.org/jira/browse/CASSANDRA-15133
> Project: Cassandra
> Issue Type: Improvement
> Components: Cluster/Gossip, Cluster/Membership
> Reporter: Jay Zhuang
> Assignee: Jay Zhuang
> Priority: Normal
>
> Restarting a node causes gossip generation update. When it propagates the
> message to the cluster, every node blindly update its local token metadata
> even it is not changed. Update token metadata is expensive for large vnode
> cluster and causes token metadata cache unnecessarily invalided.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
