[MDEV-26133] Galera graceful shutdown without removing node from joined cluster Created: 2021-07-13  Updated: 2021-07-13

Status: Open
Project: MariaDB Server
Component/s: Docker, Galera, Server
Fix Version/s: None

Type: Task Priority: Minor
Reporter: Tymoteusz Blazejczyk Assignee: Daniel Black
Resolution: Unresolved Votes: 0
Labels: None


 Description   

It is possible to have a graceful shutdown on running MariaDB/Galera node without removing it from joined cluster?

Currently graceful shutdown removes node from joined cluster. This behavior is unwanted in Docker Compose/Swarm environments. There is a need for additional graceful shutdown signal (maybe USR1) to gracefully shutdown MariadDB daemon and Galera without removing node from cluster.

Simple use case:

  • docker-compose up --scale db=3
  • docker-compose stop
  • All nodes received graceful shutdown signal in the same time from the stop command. In that case, Galera cannot handle this properly and the safe_to_bootstrap variable is not set to 1 in the grastate.dat file on any node. All nodes will fail after next docker-compose up/start invocation.


 Comments   
Comment by Daniel Black [ 2021-07-13 ]

something in wsrep_notify_cmd to record the last member of the primary. I think I was right, bootstrap was just the first problem, recovery is the harder one.

Generated at Thu Feb 08 09:43:00 UTC 2024 using Jira 8.20.16#820016-sha1:9d11dbea5f4be3d4cc21f03a88dd11d8c8687422.