search

SUSE / doc-caasp

SUSE Container as a Service Platform Documentation
https://documentation.suse.com/suse-caasp/
Other
27 stars 55 forks source link

Please add a note that we should not saturate nodes #131

Closed hwoarang closed 5 years ago

hwoarang commented 5 years ago

We should have a suggestion note that nodes should not be overloaded (or even committed) so workloads can have more chances from moving from one node to another during an update or maintenance window.

r0ckarong commented 5 years ago

Is there a good rule of thumb how oversized a cluster should be planned then?

hwoarang commented 5 years ago

I would say 70-80% utilized but that's just me. @ereslibre any other ideas?

ereslibre commented 5 years ago

We could encourage 70% and set 80% as the maximum reasonable limit.

However, assuming a cluster with two worker nodes of the same size if both are at 51% Kubernetes will run into troubles moving those workloads from one to another. Same applies if worker nodes are of different size and at different occupation rate, but only in the case of two workers. With more workers it's harder to hit this problem unless they vary a lot in the specs and one is fairly bigger than a big subset of the rest.

r0ckarong commented 5 years ago

The immediate question that comes to my mind is then: How do I limit resources to 80% max? I have an open task to describe RBAC resource limits but I need more input to know if this will cover this sufficiently.

hwoarang commented 5 years ago

I am not sure if we want to go into such a depth with resource limits. I would leave it as an exercise to the cluster administrator or just point them to upstream docs https://kubernetes.io/docs/concepts/policy/resource-quotas/

ereslibre commented 5 years ago

I agree with @hwoarang, every cluster is different in size and what workloads they run. We should just point to the upstream docs, just wanted to state the previous though.

r0ckarong commented 5 years ago

https://github.com/SUSE/doc-caasp/pull/132 Please review