UCCX Hostname and HA Island Mode

[ualbdp2c1]

We have struggled with an issue for more than a year related to HA over WAN. Even a brief network outage (1-2 seconds) between data centers would cause the master role to switch from the publisher to the subscriber. Mastership changed any time the nodes went into island mode. We finally found the root cause after numerous TAC cases, and I think it's worth sharing since changing the hostnames is supported.

The logic to elect the master node when recovering from island mode is:

Step 1. Check the service status of the nodes. If node 1 is IN_SERVICE and node 2 is in PARTIAL_SERVICE, node 1 becomes the master. If the states are the same (IN_SERVICE or PARTIAL_SERVICE), go to step 2.

Step 2. The hardware specification of both nodes will be checked. The server with the better specification will be handed the mastership. If the hardware specifications are the same, go to step 3.

Step 3. The publisher (i.e. node 1) will become the master if the hostname matches the PrimaryEngineComputerName in the ClusterSpecificConfig. If there is no match, go to step 4.

Step 4. Make the subscriber master.

Alright, now let's back up about 18 months. We decided to build a new cluster to migrate from version 9.0 to 10.6. The intent was to simplify the transition from CAD to Finesse. We built a new cluster on 9.0, restored from backup, changed the hostnames and IPs, and then upgraded to 10.6.  

Back to the present (now on version 11.6), we finally lucked out with a TAC engineer that explained the island mode recovery process above. The engineer used the CET (Configuration Editor Tool) to check the ClusterSpecificConfig file where we discovered the primaryEngineComputerName was the hostname of our 9.0 publisher. So now we know step 3 above always fails and the subscriber is elected master as a result.

TAC advised this value is populated at the time of install and cannot be changed (even with root access). The 9.0 cluster had been offline for over a year. Therefore, the best solution was to rename the 11.6 publisher to match the primaryEngineComputerName.

In conclusion, changing the hostname is supported, but it will cause issues with HA. Ideally, the set network hostname command should be updated to also change the primaryEngineComputerName. I asked to have a caveat added to the appropriate section of the admin guide. It's unclear if this request reached the technical writing team – I never received confirmation.

[Doug Johnstone]

Hi,

I have also come across this bug but was wondering if it really needs to be fixed during a software update.

Has anyone carried out a name change to the  primaryEngineComputerName parameter using Cisco and the CET?

Did it go ok?

Thanks

Doug

[ualbdp2c1]

Hi Doug,



I was told by TAC that it wasn't possible to change it from CET, but you might want to open a case to double-check. Otherwise, it wasn't too painful to use the documented procedures for changing the hostname. Remember to update your certificates and DNS records.



Good luck!



Brad

[Doug Johnstone]

Hi Brad,

one of my colleagues raised a case with TAC but the engineer didn't sound too confident on this so I wanted to check to see if anyone else has gone through this.

TAC reckon they can now update the primary engine name during any outage period using CET.

There is now a bug id for this and Cisco have the workaround as now being:

Workaround:
Contact Cisco TAC for the workaround.
( Using the CET Tool (Remote support access) manually update the Primary Engine Computer Name to the correct hostname of the UCCX 1 ).

This doesn't tally with what you were informed so thought it worth checking further.

Thanks

Doug

[ualbdp2c1]

This is a secure message. To read it, open the attachment.

[Doug Johnstone]

Hi Brad, did you mean to send this; there is no attachment.
Doug

[ualbdp2c1]

Hi Doug,

No, I did not intend to send an attachment.

Brad