cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
Announcements

Cisco Unity Connections – Split Brain

8196
Views
15
Helpful
6
Comments

The Spint Brain Recovery condition in Unity Connections is an odd state to find your Unity Connections cluster in, for sure. One thing is certain though, 'something' happened, and this is the result of whatever that 'something' was or is.

When this condition happens, what you'll notice (either from logs or by watching the cluster node status) is that the Primary and HA servers will take turns being the cluster's primary node every few minutes. This is because both cluster nodes have somehow ended up with slightly different database versions and the Server Redundancy Manager service cannot determine exactly which server should be primary. The cluster will try to resolve this issue on it's own, but often times cannot.

The good news (maybe not so good) is that in several cases, the cluster will continue to function and answer calls during a split brain. Where I have seen the split brain condition cause service outage to users is with SCCP integrations that do not have sufficient SCCP ports on both cluster nodes.

What causes this can be a number of things; often it is the result of the two nodes losing network communication between themselves and/or service failure. The resolution to this condition is fairly simple, although, you'd be best served to figure out why it happened in the first place, lest you repeat it again.

Since Unity Connections is running the same operating system as Unified Communications Manager, you can run the same health checks in Unity Connections that you would run you Unified Communications Manager. You'll want to resolve any issues discovered in those health checks before resolving the split brain issue.

Assuming you have a healthy Unity Connections cluster and/or discovered and resolved the issue that caused the split brain you'll want to move on to resolution.

  • Power off / shut down the true HA node (the node that is not supposed to be the true primary).
  • In the Cisco Unity Connection Administration section click on Cluster under System Settings in the left-side vertical navigation menu and verify that both nodes are entered in correctly (IP address/hostname/FQDN).
  • In the Cisco Unity Connection Serviceability section click on Service Management under the Tools menu and verify that the Connection Message Transfer Agent and the Connection Notifer services are started.
  • Restart the Unity Connections Cluster primary server. From the CLI, issue the command "utils system restart" and press "Y" to the proceeding prompt.
  • Once the rebooted primary server is back up, wait till you can access the GUI web page of the server before proceeding (in other words, wait till the Cisco Tomcat Service is operational).
  • From the Unified Communications Server (or other type of call control server), place a call into voicemail (all you really need is something that will signal the IVR to pickup) and then hang-up and wait 5 minutes before proceeding to the last step.
  • Power on the HA node (that should still be shutdown) and wait till you can access the GUI web page of the server before proceeding (in other words, wait till the Cisco Tomcat Service is operational).
  • Verify that the split brain status has been removed for the node status on the Cluster Management page under Cisco Unity Connection Serviceability (or issue show cuc cluster status from the CLI of the cluster's primary server).
  • Run health checks on the cluster one last time, to verify everything is healthy and operational.
  • The cluster should be back to normal status now and the split brain condition no longer listed on the cluster nodes.

In some instances (usually due to how long the cluster was in this state), even more action is necessary and you may need to reset the cluster replication (done via the CLI of the primary server with, utils dbreplication reset *) if the cluster's database replication is damaged. The need to do this will be discovered in your final set of health checks.

Comments
Beginner

Thanks Ryan for this post, I'm currently troubleshooting this exact problem.  In my case Cisco TAC is pointing to Cisco Bug CSCug53756 because of the presence of core dump ServM files on the servers. 

https://quickview.cloudapps.cisco.com/quickview/bug/CSCug53756


But I wonder whether the presence of these core dumps is a result of the problem you described, and not the cause.  During the Split Brain recovery, the system load goes through the roof on both the Publisher and Subscriber server making the entire Unity system nonresponsive to even CLI, and I'm guessing ServM is crashing because of the high load. 

Anyone have any thoughts on this?   Cisco assures me upgrading to the latest release of my Unity Connection version will resolve the issue...

Enthusiast

While I have never faced the particular bug you're referencing; if TAC is willing to assure resolution by upgrading, I would consider it. Unity Connections, IMO, is the easiest to upgrade and least likely to have compatibility issues.

However, if you are still in a split brain recovery state, you may also consider the resolution I proposed, if you have not tried that already. It may be a quicker resolution than an upgrade, especially if you're not expecting to have to do it.

Beginner

Thanks Ryan, your resolution actually is what it took to get my Unity Connection working again.  I really appreciate finding your post on this issue. 

I think my problem was caused by a couple momentary blips in network access due to spanning-tree recalculation that happened right around the time that my Unity problems started. 

Prior to shutting down both the Publisher and Subscriber the symptoms I was having were the Split Brain problem described above, along with periods of enormously high load on both servers, +200 Load Average making even the CLI unresponsive.  I'm guessing the high load is what ultimately lead to the ServM core dumps, but I have no way of knowing.  But in any case, my Unity Connection cluster was completely unusable during those periods.  For the moment my Unity Connection is back to functioning correctly while only running the Publisher and the Subscriber is still shut off, I'll probably turn it on next week.

TAC seems to be ignoring the Split Brain symptoms and only wants to focus on the core dumps.  TAC points to the Bug article I linked and simply tells me to upgrade to the latest.   I'm working on trying to get them back focused on the split brain problem instead of the core dumps.

I'm probably going to upgrade to the latest version anyway in hopes it will help, but from other searching online I see other people with very recent versions of Unity still having the Split Brain problem, so I'm doubtful the latest version of 9.x will resolve the issue...

Enthusiast

It is possible that a code weakness, makes a particular scenario more susceptible to the conditions that caused the split brain; and that by upgrading, you'll remove the weakness.

Alternatively, you've addressed the symptoms and barring any network anomalies, should be good to go. However, I do think that upgrading will best serve you in the long term.

-Ryan

Beginner

I greatly appreciate your comments on this.  Yeah it's unlikely TAC will come back with anything other than upgrade, so that's what I will do.  I won't waste any more time on troubleshooting considering they spent 3+ hours trying to get my Unity Connection system usable again to no avail, finally gave up and collected some logs and said they would have to get back to me. 

I ended up finding your fix after getting off the phone with them.   Apparently TAC is unaware of your fix for the Split Brain problem.

Beginner

The health checks link no longer functions. What were they?

CreatePlease to create content
Content for Community-Ad
August's Community Spotlight Awards