More than one primary database in a monitor results in errors in MaxScale GUI

Hi,

setups where there are multiple primary databases in a single MaxScale monitor break some visual elements. Such a setup is, at least to my knowledge, inevitable when using the binlogrouter component.

My setup looks like this:

• maria-db-01: The primary database.
• maria-db-02: Replica, replicates from replication-server.
• maria-db-03: Replica, replicates from replication-server.
• replication-server: MaxScale internal server; in front of MaxScale's binlogrouter, replicates from maria-db-01.

A minimal setup with all configurations etc. is attached to this issue, see the readme within the archive.

When I start this setup I experience the following aspects:

• Replication works.
• The same GTID is shown across all servers.
• All MariaDB instances contain the same data.
• The tooltip over replication-server shows all properties correctly (name, replication_state, seconds_behind_master, slave_io_running, slave_sql_running).
• The tooltip over maria-db-01 shows replication-server to be healthy (green), but maria-db-02 and maria-db-03 are both orange and show "+Infinitys" replication lag.
• The tooltip over maria-db-02 and maria-db-03 contains no properties at all (even though replication is healthy and running).
• Visualizing this setup via "Visualization" -> "Clusters" -> "my-monitor" shows maria-db-01 and replication-server correctly, and the replicas maria-db-02 and maria-db-03 are put in a box titled "rejoinable nodes". Both replicas are not connected to the replication-server instance.

This seems to be a collection of purely visual bugs due to the following aspects:

• maxctrl show monitors returns correct data.
• The data returned from the API endpoint at /servers returns correct data.
• The function getSlaveServerInfo() in maxgui/src/pages/Dashboard/Servers.vue seems to filter out all slave server information where the slave's primary server is different form the monitor's primary server.
• I have not further analyzed the visualization bug, as this is likely more complicated and also less important to me, but a similar filter seems likely.

Note: This bug seems to be already documented in MXS-4254, although there was another configuration problem which resulted in the ticket's closing.