This message was deleted.

To clarify, you want Longhorn to schedule two replicas to the same node and ensure that they use different disks, correct? I'm not 100% sure on this. I think you will definitely need to check the box for

Replica Node Level Soft Anti-Affinity

or set it to true (https://longhorn.io/docs/1.4.2/references/settings/#replica-node-level-soft-anti-affinity) to get multiple replicas to schedule to the same node. Once you get past this restriction in the "node and zone selection stage" our documentation doesn't make any clear guarantee that both replicas won't schedule to the same disk in the "disk selection stage". We might have to experiment a bit. See the scheduling policy for more detailed information (https://longhorn.io/docs/1.4.2/volumes-and-nodes/scheduling/).

got it, I think that currently it’s not possible to schedule on different disks..

This is an interesting! Have you tried it? If so, what was your experience exactly?

well, the documentation doesn’t seem to describe how to configure Longhorn to schedule replicas on two separate disks..

Replica Node Level Soft Anti-Affinity

allow the replicas to be scheduled on the same node, but not on different disks

Yes, this makes sense to me. I'm scanning the GitHub issues to see if there is an existing one that covers this use case.

people might use one node production environments for services which do not require HA, but data lose is not acceptable

I think the correct way to proceed is to add your use case to https://github.com/longhorn/longhorn/issues/5149, which has already been opened as a catch-all for considering additional factors when scheduling replicas. Would you be willing to: 1. Like that issue to help bump it in priority. 2. Add the clear/concise explanation for what you want the replica scheduler to do and why you want it that you put above as a comment to that issue?

if anyone need this functionality, please give it a 👍 here -> https://github.com/longhorn/longhorn/issues/5149#issuecomment-1559585527

Confirmed that we do not schedule the replicas to different disks on the same node (and instead are likely only looking at which disk has the most available space). Good raise @blue-kitchen-51801!

thanks Eric for your support