Jacek Lewandowski created CASSANDRA-19363:
---------------------------------------------
Summary: Weird data loss in 3.11 flakiness during decommission
Key: CASSANDRA-19363
URL: https://issues.apache.org/jira/browse/CASSANDRA-19363
Project: Cassandra
Issue Type: Bug
Components: Consistency/Bootstrap and Decommission
Reporter: Jacek Lewandowski
While testing CASSANDRA-18824 on 3.11, we noticed one flaky result of the newly
added decommission test. It looked innocent; however, when digging into the
logs, it turned out that, for some reason, the data that were being pumped into
the cluster went to the decommissioned node instead of going to the working
node.
That is, the data were inserted into a 2-node cluster (RF=1) while, say, node2
got decommissioned. The expected behavior would be that the data land in node1
after that. However, for some reason, in this 1/1000 flaky test, the situation
was the opposite, and the data went to the decommissioned node, resulting in a
total loss.
I haven't found the reason. I don't know if it is a test failure or a
production code problem. I cannot prove that it is only a 3.11 problem. I'm
creating this ticket because if this is a real issue and exists on newer
branches, it is serious.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
