[
https://issues.apache.org/jira/browse/CASSANDRA-21143?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Sam Tunnicliffe updated CASSANDRA-21143:
----------------------------------------
Attachment: ci_summary.html
result_details.tar.gz
> Potential gossip deadlock during decommission
> ---------------------------------------------
>
> Key: CASSANDRA-21143
> URL: https://issues.apache.org/jira/browse/CASSANDRA-21143
> Project: Apache Cassandra
> Issue Type: Bug
> Components: Cluster/Gossip
> Reporter: Sam Tunnicliffe
> Assignee: Sam Tunnicliffe
> Priority: Normal
> Fix For: 6.x
>
> Attachments: ci_summary.html, result_details.tar.gz
>
>
> When a node is removed from the cluster via a decommission, removenode, or
> assassinate a final gossip message is broadcast with LEFT state and gossip
> expiry timestamp. The purpose of this is enable the remaining nodes to
> converge on a common expiry time, it is _not_ necessary for the correct
> removal of the leaving node.
> There is a possibility that this can coincide with the regular scheduled
> gossip task, and in a decommission this could cause a deadlock as the leaving
> node attempts to update its local gossip state map. This local update is not
> necessary, so removing that step would avoid the potential deadlock.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
