kubernetes
/
autoscaler
mirror of https://github.com/kubernetes/autoscaler.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Update FAQ on overprovisioning to account for k8s 1.11 

In k8s 1.11, pod priority, and preemption is enabled by default.
The API is under `v1beta`, and they do not need to be enabled.
This commit is contained in:
Ian Lai 2018-10-26 11:54:38 +01:00
parent 3880db92f9
commit 1b50ebdc1e
No known key found for this signature in database
GPG Key ID: 4DAF99C3633B3F5D
1 changed files with 16 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
cluster-autoscaler/FAQ.md
View File

@ -317,7 +317,7 @@ number of replicas when cluster grows and decrease the number of replicas if clu
Configuration of dynamic overprovisioning:
1. Enable priority preemption in your cluster. It can be done by exporting following env
1. (For 1.10, and below) Enable priority preemption in your cluster. It can be done by exporting following env
variables before executing kube-up (more details [here](https://kubernetes.io/docs/concepts/configuration/pod-priority-preemption/)):
```sh
export KUBE_RUNTIME_CONFIG=scheduling.k8s.io/v1alpha1=true
@ -327,7 +327,9 @@ export ENABLE_POD_PRIORITY=true
2. Define priority class for overprovisioning pods. Priority -1 will be reserved for
overprovisioning pods as it is the lowest priority that triggers scaling clusters. Other pods need
to use priority 0 or higher in order to be able to preempt overprovisioning pods. You can use
following definitions:
following definitions.
**For 1.10, and below:**
```yaml
apiVersion: scheduling.k8s.io/v1alpha1
@ -339,6 +341,18 @@ globalDefault: false
description: "Priority class used by overprovisioning."
```
**For 1.11:**
```
apiVersion: scheduling.k8s.io/v1beta
kind: PriorityClass
metadata:
name: overprovisioning
value: -1
globalDefault: false
description: "Priority class used by overprovisioning."
```
3. Change pod priority cutoff in CA to -10 so pause pods are taken into account during scale down
and scale up. Set flag ```expendable-pods-priority-cutoff``` to -10. If you already use priority
preemption then pods with priorities between -10 and -1 won't be best effort anymore.