2

I'm setting up a simple 3rd VM to join my HA Cluster. This VM shouldn't ever run any services, except participate in the Quorum vote. In Galera Cluster there is a nice daemon for this, garbd, but for Pacemaker, it seems that its all done via constraints, which means I have to configure everything manually not to run on the 3rd node. This seems overkill and overly complicated. I wonder if I can have a node join a cluster in standby mode and this would simplify this process and work as a 3rd quorum node.

The situation with Pacemaker/Corosync etc gets even more complicated with all the different stacks that exist, I'm wondering if a newer simplified method has been developed that is burried deep in google.

I'm using CentOS 7 configuring everything with PCSD. (which is great, except lots of guides use crm etc)

Any suggestions on a simple 3rd VM for Quorom Votes in a Pacemaker HA Cluster Setup?

FreeSoftwareServers
  • 571
  • 1
  • 6
  • 25
  • Your question seems fine here, but I think you might find more experts at [sf]. If you'd like to have it moved, review the help center there, then flag your question, select "in need of moderator intervention", and ask the moderators to move it. – derobert Dec 21 '16 at 16:13
  • Let me see if I understand. You want a third node in your cluster but you don't ever want it to run any services. Just be their as a quorum participant? That kind of defeats the benefits of having a three node cluster. –  Dec 21 '16 at 16:21
  • Constraints are used to keep resources together. For example you have a file system for nginx, vip for nginx, and ngix processes on the same node. That group of resources would move together around the cluster. –  Dec 21 '16 at 16:27
  • @PaperMonkey The purpose is simple, to avoid split brain over Network related outages, which are likely the most common. And gives me the ability to take down one node, without losing quorum. There is a reason garbd for Galera exists for this exact reason. I'd prefer not to have to tell each constraint not to run on node3, but somehow say, globally "Don't run anything on node3" – FreeSoftwareServers Dec 21 '16 at 16:32
  • @FreeSoftwareServers - Yes that's what I thought you were shooting for. So with a two node cluster you need to setup Stonith (Shoot the other node in the head) properly so split brain doesn't occur. It's been nearly a year since I setup Pacemaker/Corosync stack so I can't give you the exact config. See http://clusterlabs.org/doc/crm_fencing.html –  Dec 21 '16 at 16:41
  • While I appreciate your efforts, I'm familiar with STONITH, and I just need to figure out how to setup a 3rd Quorum node, STONITH is an entirely different beast. – FreeSoftwareServers Dec 21 '16 at 17:00

1 Answers1

1

I was looking for the same solution for Pacemaker. 'standby' mode is not exactly what is needed, since resource checks will probably fail on this node. As I understand, 'standby' is for 'not to run' resource on the node, but not for 'not to check' resource status. What I suggest to use is a quorum device on RHEL 7.4: https://access.redhat.com/documentation/en-us/red_hat_enterprise_linux/7/html/high_availability_add-on_reference/s1-quorumdev-haar

Nopius
  • 359
  • 2
  • 5