Semi-sync is not enforced on primary when it restarts and when failover doesn't happen

[pedroalb]

Hi Shlomi!
Hope you're doing well.

I came across the following behaviour:
When semi-sync is enforced at orchestrator level, orchestrator does not set the primary node with rpl_semi_sync_master_enabled=1 on primary restart (when failover doesn't happen).

Orchestrator configuration for semi-sync:
"DetectSemiSyncEnforcedQuery": "SELECT @@global.rpl_semi_sync_slave_enabled AND @@global.rpl_semi_sync_master_timeout >= 1000000;"

All of our nodes have this enforced:
MariaDB [(none)]> SELECT @@global.rpl_semi_sync_slave_enabled AND @@global.rpl_semi_sync_master_timeout >= 1000000; +-------------------------------------------------------------------------------------------+ | @@global.rpl_semi_sync_slave_enabled AND @@global.rpl_semi_sync_master_timeout >= 1000000 | +-------------------------------------------------------------------------------------------+ | 1 | +-------------------------------------------------------------------------------------------+

We deploy our clusters with rpl_semi_sync_master_enabled=0 as default and set rpl_semi_sync_master_enabled=1 at runtime in the current primary. Whenever there's a failover (manual or automatic), orchestrator takes care of setting rpl_semi_sync_master_enabled=1 at runtime in the new primary and rpl_semi_sync_master_enabled=0 to the old primary.
However, if primary is restarted and there's no failover, orchestrator does not enforce rpl_semi_sync_master_enabled=1 on the primary.

Do you think it makes sense for orchestrator to be routinely enforcing rpl_semi_sync_master_enabled=1 on the primary?

Thanks,
Pedro.

[shlomi-noach]

Ouch, completely missed this.

[shlomi-noach]

Hopefully some discussion to follow on #1360