Botond Dénes

@avikivity should we backport this?

Backported to 5.4, 5.2 is not vulnerable.

> * [test_login_message_after_half_of_the_cluster_is_down](https://jenkins.scylladb.com//job/scylla-master/job/scylla-ci/6700/testReport/junit/%28root%29/test_password_login_message/Tests___Unit_Tests___test_login_message_after_half_of_the_cluster_is_down) [🔍](https://github.com/scylladb/scylladb/issues?q=is:issue+is:open+test_login_message_after_half_of_the_cluster_is_down) > > * [auth_cluster.test_password_login_message.debug.3](https://jenkins.scylladb.com//job/scylla-master/job/scylla-ci/6700/testReport/junit/%28root%29/non-boost%20tests/Tests___Unit_Tests___auth_cluster_test_password_login_message_debug_3) [🔍](https://github.com/scylladb/scylladb/issues?q=is:issue+is:open+auth_cluster.test_password_login_message.debug.3) https://github.com/scylladb/scylladb/issues/16821

> < t:2024-02-16 23:54:50,312 f:cluster.py l:2659 c:sdcm.cluster p:DEBUG > what(): Could not setup Async I/O: Resource temporarily unavailable. The required nr_event (1) exceeds the limit of request capacity in /proc/sys/fs/aio-max-nr...

Thanks @tchaikov for getting to the bottom of this.

> hmm, if that's the case, we would have to update scylla-nodetool to try aio and fallback to epoll. we should not do this in seastar. I don't think it...

PR here: https://github.com/scylladb/scylladb/pull/17438

> I don't understand - why this started happening now? Due to the move to the native implementation? Yes, as we implement new commands and push fixes into scylla-nodetool, the...

@mykaul after https://github.com/scylladb/scylladb/pull/17168 gets in, dtests will also start to use the native nodetool. Said PR currently serves as CI on fixing any remaining problems that surface when running dtest...

I think we can consider this fixed by https://github.com/scylladb/scylladb/pull/17438.