[
https://issues.apache.org/jira/browse/CASSANDRA-19103?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17791038#comment-17791038
]
Stefan Miklosovic commented on CASSANDRA-19103:
-----------------------------------------------
I made the in-jvm dtest stable and not flaky but I have broken
test_counter_leader_with_partial_view in Python dtests. We might rewrite it to
in-jvm dtest instead as I find the debugging in Python one rather cumbersome.
> Filter replicas for counter mutation leader to be JOINED replicas to decouple
> from RPC_READY state
> --------------------------------------------------------------------------------------------------
>
> Key: CASSANDRA-19103
> URL: https://issues.apache.org/jira/browse/CASSANDRA-19103
> Project: Cassandra
> Issue Type: Improvement
> Components: Transactional Cluster Metadata
> Reporter: Stefan Miklosovic
> Assignee: Stefan Miklosovic
> Priority: Normal
> Fix For: 5.x
>
> Time Spent: 1h 20m
> Remaining Estimate: 0h
>
> In CASSANDRA-18935, what I did was that if transport is turned off while node
> is up, it will set RPC_READY to "false". Before 18935 was in, if you turned
> off transport while node was up and transport was on, RPC_READY was not
> changed to false but it stayed to be "true" for ever.
> Clearly, this was wrong, so I fixed that, but when I did that, the code which
> decides who will be the leader of counter mutation was broken, because that
> node filtered out all nodes for which RPC_READY is false, which never
> happened before, because nobody has set it to "false" on turned off transport.
> So, we decided to revert this regression in 5.0-rc1 and I was waiting for
> CEP-21 to be in because I saw it is rewritten there with TODO to filter
> JOINED nodes.
> Hence, we can decouple this logic and fix RPC_READY status finally.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
