domain controller - single point of failure?

[lloydfischer]

I'm investigating EAP 6.0 (7.1.2?) in domain mode - non-HA. I have it up and running fine with a domain controller host and a second slave (is there a better name?) server host. But if I try to start the server before the domain controller it fails with a log message

[Host Controller] 09:58:04,320 ERROR [org.jboss.as.host.controller] (Controller Boot Thread) JBAS010901: Could not connect to master. Aborting. Error was: java.lang.IllegalStateException: JBAS010951: Could not connect to master in 5 attempts within 30000 ms

I see references to command line options --backup and --cached-dc that should allow the server to start without a DC, but bug https://issues.jboss.org/browse/AS7-4281 wold indicate this doesn't work.

So is non-ha domain mode really dependent on the domain controller for the operation of any server? If so, this is not really production ready. I must be missing something.

[emuckenhuber]

Lloyd Fischer wrote:

So is non-ha domain mode really dependent on the domain controller for the operation of any server? If so, this is not really production ready. I must be missing something.

It is obviously required to get the initial configuration model and for the manageability, however not required for the operation of the actual servers.

[emuckenhuber]

Lloyd Fischer wrote:

I see references to command line options --backup and --cached-dc that should allow the server to start without a DC, but bug https://issues.jboss.org/browse/AS7-4281 wold indicate this doesn't work.

You can use either of those. You can either use --backup to keep a local copy of the domain model. If the domain-controller is down and you shutdown the slave host you would need to use the --cached-dc to start the local slave host-controller using the backed-up model.

[lloydfischer]

If I understand, in production I would need two operation modes. Normal mode will start up with --backup. In an emergency if the domain controller is unavailable I would need to startup the slave hosts with --cached-dc. Then when the DC is restored I would have to restart the slave hosts in --backup.

I may not understand all the details, but since the slave host knows whether the DC is available or not it would seem capable of choosing the mode itself. 

[lloydfischer]

https://issues.jboss.org/browse/AS7-4412 is about this issue. When --cached-dc is specified, the DC is not contacted at all.

[emuckenhuber]

Yes, that sounds about right. --backup tells the host-controller to backup the model locally. --cached-dc to not connect to the remote domain-controller and use the locally cached model instead.

I am really just stating what it does at the moment. I haven't really looked into how much work would be required to have a combined mode, donig what you described that automatically.