[SOLVED] [HA Cluster] Prevent hosting two VMs on the same node

[Skibbi]

Hello,
Recently I created my first 3 node proxmox cluster. And I started experimenting. I have two questions that bother me:

1. I've created replication jobs every 2 hours for one VM. Now if I shutdown proxmox node manually on which VM is running let's say 15 minutes before next sync, will the replication job force synchronization before shutdown or not? If not, then what about data consistency of that affected VM? Will the new instance of VM be using last replication snapshot from 1h45m ago? What about unexpected shutdowns?
2. I have two VMs: postgres and streaming replica. Both are hosted on different nodes. How can I ensure that when I shutdown one proxmox node the VMs will be distributed on different nodes. I thought about grouping, but I not sure how should I configure it?

 

[aaron]

I've created replication jobs every 2 hours for one VM. Now if I shutdown proxmox node manually on which VM is running let's say 15 minutes before next sync, will the replication job force synchronization before shutdown or not? If not, then what about data consistency of that affected VM? Will the new instance of VM be using last replication snapshot from 1h45m ago? What about unexpected shutdowns?

There will be no replication on shutdown. It will run on the schedule.
This is also something you need to think about if you want to use HA. How long can the potential data-loss be, that you are fine with. The shortest possible interval is each minute.

If that is a complete no-go, then look into a fully shared storage, Ceph, network shares, ...

1. I have two VMs: postgres and streaming replica. Both are hosted on different nodes. How can I ensure that when I shutdown one proxmox node the VMs will be distributed on different nodes. I thought about grouping, but I not sure how should I configure it?

You can use HA groups. Select the nodes on which each group should reside on and enable the "restricted" checkbox. Make sure that there is no overlap. Though on a 3-node cluster that might be a bit problematic ;-)

 

[Skibbi]

Thanks for reply, as for now I'm just playing around and experimenting. Ceph is my next goal, however I need to create dedicated data store for that.
Regarding grouping - yes I was afraid that it might not be that easy with 3 nodes. On the contrary do I really need replica to be 100% available? I can always re-sync it with main DB if needed.