This message was deleted.
# harvester
a
This message was deleted.
b
I think they have to be exact. If the values are numerical, you can do a Gt or Lt to do more, but it's probably easier to set up a new label you're looking for and matching exactly.
m
One thing I found was that we cannot add tags to VMs being created in Rancher. I'd have to add them to the template, and have two templates for worker vs. control-plane
b
I'm not sure if that's true. I thought you could, but let me check.
Yeah you totally can
You just do two different pools for CP and WorkerNodes
m
The problem with VM Scheduling (which would give us affinity vs. anti-affinity like we want) is the lack of the actual tag when building it. That's why I'd have to tag it via two different templates
b
The tags that would give the affinity would be on the Harvester nodes and not on the VMs.
👍 1
m
If I wanted to do Node scheduling, yes. But I need to prevent VMs from being on the same host, so I need to utilize VM Scheduling for that.
b
If you're deploying the cluster in it's own namespace, then this would work:
m
Would that only apply to VMs of that type, then? If I deploy from Rancher to Harvester with 4 workers and 3 control-planes, wouldn't that just do it for all 7 of them?
b
It depends ™️
It's by pool, so you could do it just for the CP.
Or have it Required for the CP and preferred for the workers
Or use a different key
m
And, for clarification - topologykey vs rule?
b
Rules get AND'd
vs doing additional selectors.
m
Cool, thanks. I'll give these a shot, I appreciate it.
Does harvester automatically move stuff to a different node, or is it scheduled? I migrated a VM to be stacked with another for testing to see if it would be evacuated back
b
¯\_(ツ)_/¯
I don't think the scheduler looks at already running pods/vms.
m
lulz 😄
Perhaps descheduler could help in this regard ... hmmm