* why does that happen at all instead finish whatever is not finished after get SIGTERM * "try to restart the slave with --slave-exec-mode=IDEMPOTENT" must be a joke when "slave-exec-mode = IDEMPOTENT" is part of my.cnf for years *because* of that failure happening randomly when stop a slave at a wrong moment 151029 9:15:44 [ERROR] Slave SQL: Fatal error: ... Slave SQL Thread stopped with incomplete event group having non-transactional changes. If the group consists solely of row-based events, you can try to restart the slave with --slave-exec-mode=IDEMPOTENT, which ignores duplicate key, key not found, and similar errors (see documentation for details). Gtid 0-1-2058483, Internal MariaDB error code: 1593