[ 
https://issues.apache.org/jira/browse/CASSANDRA-21477?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Sam Tunnicliffe updated CASSANDRA-21477:
----------------------------------------
    Change Category: Operability
         Complexity: Normal
        Component/s: Cluster/Schema
                     Transactional Cluster Metadata
      Fix Version/s: 6.0-alpha2
           Assignee: Sam Tunnicliffe
             Status: Open  (was: Triage Needed)

> Defer the creation of system_cluster_metadata until CMS initialisation
> ----------------------------------------------------------------------
>
>                 Key: CASSANDRA-21477
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-21477
>             Project: Apache Cassandra
>          Issue Type: Improvement
>          Components: Cluster/Schema, Transactional Cluster Metadata
>            Reporter: Sam Tunnicliffe
>            Assignee: Sam Tunnicliffe
>            Priority: Normal
>             Fix For: 6.0-alpha2
>
>         Attachments: ci_summary.html, result_details.tar.gz
>
>          Time Spent: 10m
>  Remaining Estimate: 0h
>
> During the major upgrade to 6.0, each node independently creates a 
> placeholder {{KeyspaceMetadata}} instance for the new 
> {{system_cluster_metadata}} keyspace. 
> The replication strategy for this placeholder is derived by selecting the 
> datacenters listed in {{system.peers_v2}}.
> The keyspace itself is unused until all peers are running the new version and 
> the CMS is initialised, at which point it is re-created and the with the 
> correct replication which properly reflects the initial CMS membership. 
> Where there are inconsistencies in the peers_v2 table across the cluster, 
> nodes can construct these pre-init {{KeyspaceMetadata}} instances slightly 
> differently, which in turn causes a schema mismatch that prevents the CMS 
> initialisation from running without the operator excluding some subset of 
> nodes. 
> The upgrade is still able to complete, but this adds toil to the process and 
> is quite confusing behaviour, so we should improve it.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to