[
https://issues.apache.org/jira/browse/CASSANDRA-15703?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17295134#comment-17295134
]
Berenguer Blasi commented on CASSANDRA-15703:
---------------------------------------------
Mmmm can anybody help with repro steps please? The matrix of combinations {{cdc
true/false * cdc yaml true/false * DC same/diff * version same/diff}} is just
too big and everything seems to work so far. I've also been reading too much
about it so I might be missing the obvious
Method:
- Start node1 and run:
{noformat}
CREATE KEYSPACE test WITH REPLICATION = {'class' : 'SimpleStrategy',
'replication_factor' : 1};
CREATE TABLE test.foo (a int, b text, PRIMARY KEY(a)) WITH cdc=true;
ALTER TABLE test.foo WITH cdc=true;
ALTER TABLE test.foo WITH cdc=false;
insert into test.foo (a, b) values (1, 'test');
{noformat}
- Start node2, wait for bootstrap to complete and node to join successfully.
3.11:
- 2 nodes on same DC works
- 2 nodes on diff DC works
4.0:
- 2 nodes on same DC works
- 2 nodes on diff DC works
Mixed:
- Create in 3.11 & sstableupgrade to 4.0. Start a second 4.0 node and it works
> When CDC is disabled bootstrapping breaks
> -----------------------------------------
>
> Key: CASSANDRA-15703
> URL: https://issues.apache.org/jira/browse/CASSANDRA-15703
> Project: Cassandra
> Issue Type: Bug
> Components: Consistency/Bootstrap and Decommission
> Reporter: T Jake Luciani
> Priority: Normal
> Fix For: 3.11.x
>
>
> Related to CASSANDRA-12697
> There is an edge case left over. If a cluster had enabled CDC on a table
> then subsequently set cdc=false, subsequent bootstraps break.
>
> This is because the cdc column is false on the existing nodes but null on the
> bootstrapping node, causing the schema sha to never match.
>
> There are a couple possible fixes:
> 1. Since 12697 was only about upgrades we can serialize the cdc column IFF
> the cluster nodes are all on the same version.
> 2. We can force cdc=false on all tables where it's null.
>
> I think #1 is probably simpler. #2 would probably cause more of the same
> problem if nodes are not all updated with the fix.
>
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
