Disable cluster.memory.allocated.capacity.disablethreshold check? Bad interaction with mem.overprovisioning.factor

[bhouse-nexthop]

I am running Cloudstack 4.22.0.0 + PR #11932. I have these configurations set:

• mem.overprovisioning.factor: 4
• cluster.memory.allocated.capacity.disablethreshold: 0.85
• vm.min.memory.equals.memory.divided.by.mem.overprovisioning.factor: false

Allocated capacity looks like this:

But actual memory usage looks like this:

This proves the PR #11932 + the fact that I tuned KSM on the host is working nicely.

That said, with cluster.memory.allocated.capacity.disablethreshold at 85%, I get:

2025-12-18 22:35:07,143 WARN  [c.c.d.FirstFitPlanner] (API-Job-Executor-67:[ctx-c19a2c63, job-23855, ctx-d91b5367]) (logid:8ecbd546) Cannot allocate cluster list [1] for VM creation since their allocated percentage crosses the disable capacity threshold defined at each cluster at Global Settings Configuration [name: cluster.memory.allocated.capacity.disablethreshold, value: 0.85] for capacity Type : MEMORY, skipping these clusters

I tried setting the value to 0 thinking that means disable (not sure why you'd ever use a value of 0 as that would make it always fail), but it didn't fix the error.

I currently have it set to 1 for 100% but I have a feeling in a few days I'm going to hit the same error again even though I haven't hit my overprovision ratio.

Either this check needs to have the ability to be completely disabled, or it needs to be made to honor the overprovisioning factor.

Anyone have any guidance on this?

[weizhouapache]

you can increase the mem.overprovisioning.factor, or set threshold value to a number larger than 1 by cmk （not sure it affects anything）

however, you might face OOM errors, so please be very careful.

[bhouse-nexthop]

I'm monitoring actual memory usage per hypervisor separately. These are for developer VMs so we know they spike and release. It is understood that any sort of overcommit could lead to significant issues if not properly monitored.

It appears setting the mem.overprovisioning.factor only applies to VMs created newly. It doesn't appear there is any way to change this setting. Even stopping and starting VMs doesn't change it, we can see the overprovisioning factor set in the VM settings stay at whatever the original value was.

I had to make a direct DB edit:

update vm_instance_details set value=4 where name='memoryOvercommitRatio';

So I guess this is expected behavior, but doesn't appear to be obvious. Its not entirely clear why it works this way.

[DaanHoogland]

thanks @bhouse-nexthop , if you care to add a documentation PR?!

[rajujith]

Its already documented.

Only Instances deployed after the change are affected by the new setting. If you want Instances deployed before the change to adopt the new over-provisioning factor, you must stop and restart the Instances. When this is done, CloudStack recalculates or scales the used and reserved capacities based on the new over-provisioning factors, to ensure that CloudStack is correctly tracking the amount of free capacity.

https://docs.cloudstack.apache.org/en/4.22.0.0/adminguide/hosts.html#setting-over-provisioning-factors

[bradh352]

This is NOT the actual behavior though. Stopping a VM and starting it again does NOT recalculate. Only newly created instances are impacted. The DB edit route was the only successful way I could do this.