A Core Dump Occurs in the Ceph Async Messenger

Symptom

During the I/O pressure test, there is a possibility that a core dump occurs in the fio or Global Cache server software. The following figure shows the core dump stack.

Cause Analysis

The Ceph Async Messenger does not protect session_stream_handlers. As a result, conflicts occur when multiple threads access session_stream_handlers.

For details, visit https://tracker.ceph.com/issues/40026.

Solution

Modify the Ceph source code based on the pull requests in the Ceph community and recompile Ceph.

Example: https://github.com/ceph/ceph/pull/30717