It seems that MariaDB can receive a signal handler and start to dump a stack trace, and yet Galera cluster activity still continues in another thread. This may be exacerbated by MDEV-16084 or some other condition that causes the signal handler to get stuck and fail to terminate the process.

In the example log attached, the signal handler starts and yet wsrep activity continues for a few tens of seconds. Our automation terminated MariaDB for unrelated reasons at that point – nothing to do with the signal handler. I believe MariaDB would have continued indefinitely in some state otherwise.

We're still working on debugging the segfault, and the segfault itself is not the subject of this bug report; just the signal handler.