[Experimental] Implemented Hostname AntiAffinity #1221
Conversation
|
✔️ Deploy Preview for karpenter-docs-prod canceled. 🔨 Explore the source changes: d59991e 🔍 Inspect the deploy log: https://app.netlify.com/sites/karpenter-docs-prod/deploys/61f1a750e9192a0007563326 |
ellistarn
changed the title
| return errs | ||
| } | ||
|
|
||
| // Transform pod anti affinity into topology rules |
There was a problem hiding this comment.
Oh this is interesting. Is this how this works under the hood in K8s as well or a new approach that gets us there quicker in Karpenter?
There was a problem hiding this comment.
This is absolutely not how it works under the hood. It's a bit of a hack that takes advantage of simplifying antiaffinity enough until it's translatable to topology.
There was a problem hiding this comment.
From https://kubernetes.io/docs/concepts/scheduling-eviction/assign-pod-node/ I saw this.
"Note: Pod anti-affinity requires nodes to be consistently labelled, in other words every node in the cluster must have an appropriate label matching topologyKey. If some or all nodes are missing the specified topologyKey label, it can lead to unintended behavior."
So it seems that we also need to create a mechanism to label the existing nodes with the topologKey in order for this to work.
1. Issue, if available:
#942
2. Description of changes:
Implements hostname antiaffinity by translating these rules into topology rules.
Requested feedback:
3. How was this change tested?
4. Does this change impact docs?
By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.