[ https://issues.apache.org/jira/browse/CASSANDRA-19239?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17810016#comment-17810016 ]
Jacek Lewandowski commented on CASSANDRA-19239: ----------------------------------------------- If I change those settings, I'll do that only for the tests. I'm not going to touch the production configuration. Anyway, the immediate solution is to increase the heap size for tests. Those tests use {{medium}} configuration in CircleCI which means 2 vcpus and 4 GB. Increasing the heap size from 1G to 1.5G or 2GB should be fine. 1GB is pretty little for JVM dtests, especially if we run multiple nodes. Also, the diagrams above show that the cleanup is not perfect and we have a lot of leftovers after each test case. > jvm-dtests crash on java 17 > --------------------------- > > Key: CASSANDRA-19239 > URL: https://issues.apache.org/jira/browse/CASSANDRA-19239 > Project: Cassandra > Issue Type: Bug > Components: Test/dtest/java > Reporter: Jacek Lewandowski > Assignee: Jacek Lewandowski > Priority: Normal > Fix For: 5.x > > Attachments: image-2024-01-23-13-11-50-313.png, > image-2024-01-23-13-12-33-954.png > > > This is a similar problem to the one mentioned in > https://issues.apache.org/jira/browse/CASSANDRA-15981 > I'm filling it because I've noticed the same problem on JDK17, perhaps we > should also disable unloading classes with CMS for JDK17. > However, I'm in favour of moving tests to G1 instead. -- This message was sent by Atlassian Jira (v8.20.10#820010) --------------------------------------------------------------------- To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org For additional commands, e-mail: commits-h...@cassandra.apache.org