The issue emerged by following fix in 10.2.10: https://jira.mariadb.org/browse/MDEV-13950
The behaviour is correct and expected. We need to fix the script that want to do stuff in /tmp, instead of the SELinux rules. We already fixed this in RHSCL MariaDB 10.1 collection, so this is a regression. The issue is reproducible only on RHEL6, since we don't execuse mysqld_safe with systemd service, as we do it in sysVinit. No later software collections are hit, since MariaDB 10.3 collection isn't shipped on RHEL6. --- For more info look at the fix in 10.1 collection: https://src.osci.redhat.com/rpms/mariadb/c/421e3cccd6b93d31edc187fb80008cdcd3bd973f?branch=rhscl-3.3-rh-mariadb101-rhel-6 --- Upstream changed location of the temporary file in the wsrep_recover_position function, which causes issues on RHEL-6, because mysqld_safe is not expected to play with files in /tmp, and SELinux started to complain. Changing the location of the temporary files to the datadir means practically reverting part of the upstream commit https://github.com/MariaDB/server/commit/bb7a70c9551c1756b1d1736ca4f6a0a965795873 Upstream issue: https://jira.mariadb.org/browse/MDEV-10767
QA note: Can be tested by service starting on RHEL 6.
Providing blocker+ to be able to push to rhscl-3.3 branch