Quorum for two node pacemaker active/standby DRBD cluster

Hi

I’m looking for the best way to setup a quorum for two node active/standby DRBD cluster.

Qdevice is not an option, because it requires an additional host, this is the very last chance.

So I’m looking for something like HPE’s serviceguard lock disk, which is disk and doesn’t require configuring an additional host.

Can external stonith block device mapped and configured on both cluster nodes with auto_tie_breaker enabled be suitable for this ?

Anton

I am not familiar with “HPE’s serviceguard lock disk”, but can I assume that this is a disk that can be written to by all members of the cluster? If so, then you could likely configure SBD (Storage Based Death) STONITH within Pacemaker. Check out this article for a quick primer: ⚡ Using SBD With Pacemaker Please note that fencing/STONITH does not equal quorum.

You really cannot implement quorum with only two nodes. However, DRBD was two node only for years and years before version 9. We have been successfully mitigating split-brains with fencing only for a long time now. The quorum features were originally implemented for clusters, with three or mode nodes, as a way to prevent split-brains in environments where fencing wasn’t possible or didn’t make sense. That is not to say that you cannot combine them if you want. You can use both quorum and fencing together when it makes sense, or one or the other, or neither (and risk potential split-brains).

1 Like

I am not familiar with “HPE’s serviceguard lock disk”, but can I assume that this is a disk that can be written to by all members of the cluster?

Yes, correct.

I configured Storage Based Death from shared storage for two cluster nodes using NVMe over TCP.

Interesting, that when I enabled sbd (pcs stonith sbd enable) it suggests enable auto_tie_breaker -

Warning: SBD fencing is enabled in the cluster. To keep it effective, auto_tie_breaker quorum option will be enabled.

I’m still in deep thinking if it is enough, or better switch to qdevice with the disadvantage as additional qdevice host.

Thank you for sharing your experience, I’m planning to stay with the current setup (SBD + fencing/stonith)

Anton

1 Like

Interesting, that when I enabled sbd (pcs stonith sbd enable) it suggests enable auto_tie_breaker -

Warning: SBD fencing is enabled in the cluster. To keep it effective, auto_tie_breaker quorum option will be enabled.

auto_tie_breaker is less suitable for two node cluster than two_node. So I edited corosync.conf,

“auto_tie_breaker: 1” → “two_node: 1”

Combining strong, multi-level fencing/stonith (sbd_fence, ipmi_fence, …) looks the best “quorum” like in two node setups.

Anton

I’m using a tiny NanoPi R3S exactly for this use case. It is powered with a Ideal Diode USB-C power supply to both nodes so it even got redundant power supply :wink: