Thank you for clarifying. The use case I am looking for is placing virtual machines in terms of actual ram usage due to limitation of the way the free_ram_used is accounted for. For example, we would like to place virtual machines on hosts with actual available ram instead of computed for allocation ration and reserved memory. Is there other tools available to achieve this goal? Thank you very much. Best regards, Jiatong Shen On Fri, Mar 7, 2025 at 11:22 PM Sean Mooney <smooney@redhat.com> wrote:

On 07/03/2025 08:44, Jiatong Shen wrote:

...

Hello Experts,

I hope this email finds you well.

I am researching the "compute_monitors" option for nova and would like to find out what it can do. By some googling I find that by enabling this option, the compute_nodes table will store extra metrics and reflect the vcpu usage on the compute node. My question is is it possible to also save memory usage?

no

the only data this can collect is

https://github.com/openstack/nova/blob/master/nova/compute/monitors/cpu/virt...

...

Besides, is this option still supported in the long term?

kind of

the compute monitor interface is feature frozen we are not currently planning to every add any more.

however peopled occasionally ask use too so while it unlikely to change going forward it might.

this policy is codified in our developer policies doc

https://docs.openstack.org/nova/latest/contributor/policies.html#metrics-gat...

while we have frozen the feature we never deprecated it for removal or made it unsupported

so yes you can use this to enable the metrics weigher to function

we have downstream folks using this to avoid host with large iowait or load when sechduleign vms.

this was never intended as a monitoring solution so if your use-case is not related to scheduling

and vm placement you should look at celiometer or other monitoring solutions.

...

Thank you very much for the help.

-- Best Regards, Jiatong Shen