Configurable rescheduling when waking from suspend, or starting from shutdown

[Tobias Rydberg]

As a follow-on from the above, re-evaluate a node's CPU and RAM utilization when a guest starts (after stop) or resumes (after suspend), and have a configurable option for what to do if CPU and RAM thresholds are breached:

Start/resume the guest in place, ignoring the overcommit (this is the current behavior)
Kick the guest back to the scheduler, to find a better-suited (i.e. non-overcommitted) node
Presumably, operators would be more likely to use the second option if and only if guests run on shared storage, RBD, etc., but it could also be used with non-shared storage, equivalent to block migration.

[Matt Riedemann (mriedem)]

> As a follow-on from the above

Please be specific.

[Matt Riedemann (mriedem)]

I'm guessing the reference is to https://bugs.launchpad.net/openstack-publiccloud-wg/+bug/1791679 ?

[Dan Smith (danms)]

Unfortunately, what is described here goes against the fundamental way nova works. Stopping an instance does not de-commit resources as far as nova is concerned, which means starting it should always work. If you want to release resources on the compute node while your instance is stopped, and re-evaluate the placement on resume, then the shelve operation is what you want.