DRBD - Split Brain issue

1
votes

Im using 2 node DRBD(v9.0) cluster with pacemaker. STONITH is disabled. I have restarted my Node 2(which is acting as a primary). Then after restart, I can't able to login and it shows the split brain occurred.

  1. How can I avoid this kind of situation?
  2. If I enabled STONITH will it be fixed?
  3. What happens if both nodes have written some data?
1
linuxcluster-computinghigh-availabilitydrbd
Unable to login to what after the reboot? – Matt Kereczman
Yes unable to login, I'm using AWS EC2(there is an option to get screen shot of the screen, from there I got this error message – TheDataGuy
What are you trying to replicate? – Matt Kereczman
It's for postgresql database high availability – TheDataGuy

1 Answers

1
votes

DRBD split braining shouldn't cause any login issues...

To answer your questions:

  1. STONITH is the only way to avoid split-brains.
  2. Enabling STONITH in Pacemaker is literally the last step in configuring proper node-level fencing (aka. STONITH). Which type of STONITH device you'll use is heavily dependent on your hardware/hypervisor/environment's configuration.
  3. Then you have a diverged data set. You'll need to merge the data (if you care about it), and then chose which node will be the split-brain survivor, which will be the victim, and finally manually resolve the split-brain. This is why STONITH is such an important concept in clustering.