Scale max CPU when migrating

[voarsh]

Not sure if this has been requested, but HA or migration won't work when CPU core numbers differ on different nodes.

I think at the very least if the migration FROM node has 16 cores, migrating to a node with 8, it should scale down so it doesn't automatically fail?

 

[AllanM]

Assuming both nodes have SMT enabled cores, have you considered reducing the number of virtual cores to 16 on this VM so that it can run on your 8 core nodes? On a 16 core node this still allows the VM access to ~70-80% of compute resources of the node.

In practice, it's not usually a good idea to provision every available hardware thread to any particular VM in a cluster anyway. Better to keep some compute resources free for the hypervisor and software defined storage (if you're using ceph, for example).

 

[apoc]

In practice, it's not usually a good idea to provision every available hardware thread to any particular VM in a cluster anyway.

+1

I keep wondering that nearly 10 years in mainstream virtualization people still have not understood the concepts of cpu virtualization and think "more is more".
Assigning lots of cores adds a cost and negates the purpose of flexibility. For that reason the request imho doesn't make sense at all, because conceptional something is wrong in such a situation (no N+1)