HA question

[MaPf]

If there is a cluster with 6 nodes, and there is a set of machines in a HA group like:
node1:100,node2:20,node3:1,node4:1,node5:1,node6:1
HA is set to shutdown-policy=migrate
CRS is set to ha-rebalance-on-start=1

now when maintenance on node1 is done, VMs are migrated to node2.
What is supposed to happen if node2 is out of memory?
From my understanding, even if the priority is very low, VMs should be migrated to other nodes.

not in our case ... node2 just ran full and rebooted then

Is that expected behaviour?

 

[fiona]

Hi,
no, HA group priorities take precedence over other scheduling:

For each service that needs to be recovered or migrated, the scheduler iteratively chooses the best node among the nodes with the highest priority in the service’s group.

https://pve.proxmox.com/pve-docs/chapter-ha-manager.html#ha_manager_crs

 

[MaPf]

Ok, understood. Thanks for the link.

But is it expected for node2 to continue migration, even if there is not enough memory available?
Shouldn't the migration at least stop?

 

[fiona]

There currently is no heuristic in HA to block recovery because of memory usage. Such a feature would still need to be implemented. Note that effective memory usage can be hard to predict, in many cases it will still work even if overprovisioned: https://pve.proxmox.com/wiki/Dynamic_Memory_Management