[
https://issues.apache.org/jira/browse/CASSANDRA-12510?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Kurt Greaves updated CASSANDRA-12510:
-------------------------------------
Attachment: 12501-3.x.patch
Had a crack at this. Basically just changed decommission to take the --force
argument and check keyspace RFs.
Also we were checking if this was the only node in the ring using # of tokens,
which doesn't work for vnodes.
> Decommission process should raise an error flag when nodes in a DC don't have
> any nodes to stream data to
> ---------------------------------------------------------------------------------------------------------
>
> Key: CASSANDRA-12510
> URL: https://issues.apache.org/jira/browse/CASSANDRA-12510
> Project: Cassandra
> Issue Type: Bug
> Components: Streaming and Messaging
> Environment: C* version 3.3
> Reporter: Atin Sood
> Assignee: Kurt Greaves
> Priority: Minor
> Labels: lhf
> Attachments: 12501-3.x.patch
>
>
> Steps to replicate :
> - Create a 3 node cluster in DC1 and create a keyspace test_keyspace with
> table test_table with replication strategy NetworkTopologyStrategy , DC1=3 .
> Populate some data into this table.
> - Add 5 more nodes to this cluster, but in DC2. Also do not alter the
> keyspace to add the new DC2 to replication (this is intentional and the
> reason why the bug shows up). So the desc keyspace should still list
> NetworkTopologyStrategy with DC1=3 as RF
> - As expected, this will now be a 8 node cluster with 3 nodes in DC1 and 5 in
> DC2
> - Now start decommissioning the nodes in DC1. Note that the decommission runs
> fine on all the 3 nodes, but since the new nodes are in DC2 and the RF for
> keyspace is restricted to DC1, the new 5 nodes won't get any data.
> - You will now end with the 5 node cluster which has no data from the
> decommissioned 3 nodes and hence ending up in data loss
> I do understand that this problem could have been avoided if we perform an
> alter stmt and add DC2 replication before adding the 5 nodes. But the fact
> that decommission ran fine on the 3 nodes on DC1 without complaining that
> there were no nodes to stream its data seems a little discomforting.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
