Page MenuHomePhabricator

Crash after disabling semi sync + start a master switchover process
Closed, ResolvedPublic

Description

During some of the stand by DC we've seen 3 crashes with the same pattern.
Disabling semi sync + moving the last replica made the DC master crash, so I've created a bug to MariaDB which has been confirmed: https://jira.mariadb.org/browse/MDEV-36359

The work around provided by MariaDB:

A work-around is to set rpl_semi_sync_master_wait_no_slave=1 temporarily before setting rpl_semi_sync_master_enabled=0 and keep it like that until all slaves have disconnected. This should prevent the master server from crashing.

This task is to track the resolution of the bug.

Event Timeline

Marostegui triaged this task as Medium priority.Mar 25 2025, 6:22 AM
Marostegui moved this task from Triage to Meta/Epic on the DBA board.

The fix has been merged: https://github.com/MariaDB/server/pull/3931
I am going to leave this open until this is released on the next version.

The new version with the fix has been released: T394623 https://mariadb.com/kb/en/mariadb-10-6-22-release-notes/

Replication
A primary server could crash when a semi-sync connection is stopped, if the primary previously disabled semi-sync replication while the connection was already up (and `rpl_semi_sync_master_wait_no_slave=0`). (MDEV-36359)