1

I'm absolutely new to the whole cluster topic (pacemaker with DRBD, etc) and in general new to linux (using it on a daily basis for about two months now). Actually I'm playing around with a simple Apache cluster, accessing a shared DRBD mounting point, just for fun. Everything works fine now, despite the fact, that every time one of the two nodes reboots I end up with a split brain situation. My guess is, that the network connection is cut off before the primary node can declare itself as secondary (does it in fact work that way?). So my question is how to investigate what's the problem here. Any advice would be appreciated.

Thanks!

Juke
  • 133
  • 1
  • 5
  • Don't you need a majority to prevent split brain situations? how do you get a majority vote from just 2 nodes? – hookenz Jun 23 '15 at 00:17
  • Hmm, I guess not. I have a two node setup in two virtualbox VMs, both using the DRBD service to sync the htdocs of Apache. When Node A is primary and shut down via reboot command I always get the split brain situation. When Node A is secondary it won't happen. This is why I thought it may be caused by an improper shutdown order. – Juke Jun 23 '15 at 00:25
  • What you're referring to is quorum; which in two node Pacemaker clusters should be ignored by setting ` no-quorum-policy=ignore` in the `property` section of your Pacemaker configuration. DRBD itself doesn't care about quorum since version 8.x only replicated between two nodes. – Matt Kereczman Jun 25 '15 at 05:12

1 Answers1

1

Looking at the syslog on both nodes will likely help you determine where the split-brain is happening.

Understanding how DRBD uses generation IDs to decide that a split-brain has occurred may help as well. Whenever the generation IDs are updated or compared, there will be an entry placed in syslog.

Reference: https://drbd.linbit.com/users-guide/s-gi.html

EDIT: I just noticed the Pacemaker tag on this post. Do you have DRBD/Apache set to start at boot? When using a cluster manager like Pacemaker, you will want to leave all the starting and stopping of services up to Pacemaker.

Matt Kereczman
  • 1,899
  • 9
  • 12