[Bug]: `host_anti_affinity_required = false` in `kube_deployment` + `kube_descheduler` lead to regular preemption
jlevydev opened this issue · comments
Josh Levy commented
Prior Search
- I have already searched this project's issues to determine if a bug report has already been made.
What happened?
I deployed a kube_deployment
with host_anti_affinity_required = false
to allow workloads to run on the same nodes. I immediately noticed a sharp increase in workload disruption. Upon investigation of my cluster's events I found that the kube_descheduler
was removing duplicate pods on nodes as is the Panfactum
default, but if that's the case it probably isn't worth having host_anti_affinity_required
be configurable since you're basically signing up to have the kube_descheduler
nuke a pod every five minutes lol
Steps to Reproduce
- Create a
kube_deployment
withhost_anti_affinity_required = false
- Scale that deployment such that you have more than one pod per node in your cluster (or at least have one set of pods sharing a node)
- Observe them get mercilessly de-scheduled
Relevant log output
No response
Jack Langston commented
@jlevydev I verified this is an issue and it should be resolved in the next release.
Jack Langston commented
Resolved.