docker stop kills mariadb process

[Kai Wembacher]

We are using the latest kolla containers (MariaDB image id e4c5cfddb900) for Rocky based on Ubuntu.

When running docker stop the MariaDB container doesn't stop gracefully. Instead it is forcibly killed by docker after exceeding the stop timeout. This leads to the Galera cluster (in a two-node setup) becoming unhealthy immediately.

It seems that this behavior is related to "dumb-init --single-child" which was introduced as part of this bug: https://bugs.launchpad.net/kolla/+bug/1799642. This parameter sends the SIGTERM signal only to the direct child of dumb-init which is /bin/bash and doesn't correctly pass the signal to mysqld:

USER PID %CPU %MEM VSZ RSS TTY STAT START TIME COMMAND
mysql 1 2.0 0.0 200 4 ? Ss 14:49 0:00 dumb-init -- kolla_start
mysql 8 1.0 0.0 18640 3308 ? Ss 14:49 0:00 /bin/bash /usr/bin/mysqld_safe
mysql 187 30.0 4.2 4088432 350336 ? Sl 14:49 0:00 /usr/sbin/mysqld --basedir=/usr

We verified that in this scenario "kill -15 187" gracefully stops mysqld but "kill -15 8" does not.