how to configure nova scheduler/filter and memory usage of compute nodes?

asked 2019-06-12 08:15:11 -0500

minimike gravatar image

Hi

I have got an learn/testing environment realized with OpenStack. In my case I run only 33% of my VM's but the rest of my systems are turned off. This currently results that NOVA Scheduler with AggregateRamFilter by all VM's, however running or not, calculates the amount usage from all, means not running and running VM's. To handle it, I have got overprovisioned my both Compute-Nodes with the setting "ram_allocation_ratio=9.0" on each node. This reduces my "No valid host was found. There are not enough hosts available" Errors. But this is a awfull solution. It works for some things, but not for all.

So I would like that the AggregateRamFilter could been set on default. But Filters shall only watch on free available memory but ignore the usage from not running VM's. Looking only on free available memory on my compute nodes.

Is there an setting how I could configure it? Maybe with an example?

OpenStack Queens, Ubuntu 18.04, with two compute nodes. Each haves got 128 GB RAM

cheers Darko

edit retag flag offensive close merge delete

Comments

I'm not aware of anything other than the already mentioned overprovisioning. Nova needs to make sure that it still can boot already existing instances even if they are shut off, so that new instances will be rejected by the scheduler.

eblock gravatar imageeblock ( 2019-06-20 09:17:27 -0500 )edit