DAG availability in an odd number of hosts.

Hi,
If two nodes fail at once, the remaining node will NOT be enough to keep the cluster quorum.

Have a read at it here:

<<

Three nodes with a witness
All nodes vote, so the witness doesn't initially vote. The majority is determined out of a total of 3 votes. After one failure, the cluster has two nodes with a witness – which is back to Scenario 2. So, now the two nodes and the witness vote.

Quorum explained in the case with three nodes with a witness

Can survive one server failure: Yes.
Can survive one server failure, then another: Yes.
Can survive two server failures at once: No.

>

Reference:
https://video2.skills-academy.com/en-us/azure-stack/hci/concepts/quorum#cluster-quorum-overview
https://video2.skills-academy.com/en-us/azure-stack/hci/concepts/quorum#how-cluster-quorum-works

If the above response was helpful, please feel free to "Accept as Answer" and "Upvote" the same so it can be beneficial to the community.

Hi, thanks for the answer.
I will clarify a little what I want to get and maybe your recommendation from the schemes I have proposed?

I plan a scheme of three physical exchange 2019 servers in DAG, in one location, two server servers. Geo-distributed and different data centers are still in the future.
As far as I understand, with an even number of hosts, a shared witness disk is involved- quorum. If odd, the quorum is not used.
A scheme with two nodes in DAG and witness - will withstand the failure of one server.
A scheme with three nodes in a DAG without witness - will withstand the failure of one server.

Please tell me, in this case, which scheme is the best to prefer if there are three physical nodes in operation, and in fact the cluster can withstand the failure of only one of them?
While considering the following options:

1. Locations in the office (two server rooms) - placing at least two of the three cluster nodes in the server room, where the split system does not fail.
2. Adding a fourth node (as a virtual one), so that the quorum would be of 4 nodes, which will allow to withstand the failure of two nodes out of four. (Although this configuration is not recommended)?
   For the future:
3. The option with a data center is to place all three servers on the data center site.
4. Two physical hosts in the office and one physical + virtual in the data center
5. As in Paragraph 2, only in different offices in different cities.

Sure, let's make sure we have the dynamic quorum behavior cleared. With dynamic quorum enabled (default), in the event of subsequent node failures, the cluster will stop counting the failed nodes' votes so that it won't lose quorum even with one node remaining. This doesn't cover you when you lose multiple nodes at once.

From the two options, having three nodes without a witness is preferred as it has a 50/50 chance that it will survive one node failure followed by another one. This is not something 2-node clusters can benefit of. Moreover, if you also add a witness resource to the three-node cluster, that 50/50 chance becomes the most probable outcome.

Reference: https://video2.skills-academy.com/en-us/azure-stack/hci/concepts/quorum#how-cluster-quorum-works

If your business goal is for your cluster to survive after two nodes fail at once, you won't be able to achieve it with 3 nodes. Having a virtual node added to the cluster might work though not recommended for production anyhow as it can be unreliable network/performance-wise.

Instead of a quorum disk, you may also use a File Share Witness (in a separate site accessible to all nodes) or a Cloud Witness (if you have internet access on your nodes or a connection between your on-prem and Azure).

Hope this helps :)

Yes, I will have a shared folder of a file resource as a witness disk in any case.

Then why sometimes implement a DAG cluster of three nodes, and not two or four?