Details
-
Bug
-
Status: Closed (View Workflow)
-
Major
-
Resolution: Fixed
-
2.2.18
-
centos7 cloud container
Description
Setup dual datacenter Maxscale w/ Master-Slave node using mariadbmon with auto-failover. System is deployed correctly as can be seen from first maxadmin in mxs.issue file. Each data center is Master/Slave and the masters are replicating to each other (as external servers).
However, when I take down the master (mariadb-0) in dcA, maxscale indicates that it is failing over to mariadb-1 node and it applies promotion.sql and redirects new master (mariadb-1) to new external master. Then it looses Master status in the local datacenter.
Attached is mxs.issue file which shows maxadmin list servers in each datacenter before and after dcA-mariadb-0 is taken down.
Also attached is maxscale.log showing mariadb-1 promoted to master.
This seems to be an issue in the master detection algorithm of the monitor. The failover is indeed ran correctly, just the state of the surviving backend is set incorrectly. This is likely not an issue in 2.3.
If the cluster where the master fails contains three or more servers, the detection works. If an external master is not used, the two-server configuration works due to the "detect_standalone_master"-setting.