RestartReplicationQuick called even from Orchestrator cluster where recovery has been globally disabled

Hints for QA:

To trigger this case, we need:

1. working source node, but Orchestrator not being able to connect to it

2. lagging replica

Steps to reproduce:

1. Create 2-nodes replication chain: source (port 4000)→ replica (port 5000)

2. Allow Orchestrator to discover the topology

3. Disable recovery globally: orchestrator-client -c disable-global-recoveries

4. Start user session on source

   1. CREATE TABLE t(id INT AUTO_INCREMENT PRIMARY KEY, v DOUBLE, d DATETIME);

   2. SET binlog_format = STATEMENT;

   3. INSERT INTO t (v, d) SELECT 1, NOW() FROM information_schema.INNODB_METRICS LIMIT 10;

   4. do not close the connection

5. on the source node: set global max_connections=2;

6. restart the orchestrator. It will indicate that there is a dead master. Error log of orchestrator will show “too many connections” error returned buy source node

7. Execute UPDATE t SET v = SLEEP(15), d = now() WHERE id = 1; in user session (the one from point 4)

8. In a while, Orchestrator should detect UnreachableMasterWithLaggingReplicas (indicated in logs)

9. Orchestrator should attempt to restart replica. It will be visible in Orchestrator logs:
   2024-11-07 13:55:47 INFO stop slave io_thread on kamil-Latitude-5531:5000 as part of RestartReplicationQuick
   2024-11-07 13:57:20 INFO start slave io_thread on kamil-Latitude-5531:5000 as part of RestartReplicationQuick

The problem is point 8. When recovery is disabled globally, restarting the replica should not be attempted.

It will go to Orchestrator v3.2.6-15 (PS distribution 8.0.40)