Domain Configuration - Losing Primary Datasource and Wildfly

[bohremsdaq]

I have a 3 node Cluster which is running in domain mode. The Primary node being the Host controller. The system is running in a vSphere environment so has HA protection and machines will automatically restart when power is returned to a host. The problem is that the Primary Database and WIldfly are on a single host (and need to remain there) so losing the Host loses the Primary Datasource and Wildfly. The Secondary/Tertiary Wildlfy Move seamlessly to the secondary Datatsource after a configured connectTimeout. When the primary datasource and wildfly reconnect they choose the primary datasource in the list (this is a multimaster so primary/secondary and tertiary database are all write capable). The objective is to try and keep all wildfly servers talking to the same database (even though the are all write masters) to reduce any potential issues. Is there a method of getting the primary host to "see" which datasource is in use by the secondary/tertiary and use that as it's datasource instead of the first in the list

[rhusar]

I don't think there would be such functionality out of box. Sounds like you can setup an external process to monitor the status and then submit management commands to reconfigure the datasource and reload. Or maybe you can implement this using the clustering API (see https://github.com/wildfly/wildfly/blob/10.0.0.Final/clustering/api/src/main/java/org/wildfly/clustering/dispatcher/CommandDispatcher.java) to figure out what the datasource setting should be and react to it.

[mchoma]

Which database do you use? Couldn't you utilize some HA solution of your DB? For example PostgreSQL:

• PostgreSQL: Documentation: 9.4: High Availability, Load Balancing, and Replication
• and setting datasource Connecting to the Database

[bohremsdaq]

Thanks,

I shall have a look at the API.

[bohremsdaq]

Thanks Martin,

I am using Postgresql 9.4, but the 2ndQuadrant BDR extension (MultiMaster Replication)

For some reason I am being restricted from implementing an external HA Load Balancer (or Virtual IP as this is on a LInux Box), and was tasked to find a solution to this.