https://github.com/frankkkkk/k8s-nodeselector-ensurer
A POC to remove pods when their nodeSelector no longer matches
https://github.com/frankkkkk/k8s-nodeselector-ensurer
kubernetes nodeselector pods watcher
Last synced: 2 months ago
JSON representation
A POC to remove pods when their nodeSelector no longer matches
- Host: GitHub
- URL: https://github.com/frankkkkk/k8s-nodeselector-ensurer
- Owner: Frankkkkk
- Created: 2020-05-18T11:29:36.000Z (almost 5 years ago)
- Default Branch: master
- Last Pushed: 2022-12-09T05:19:27.000Z (over 2 years ago)
- Last Synced: 2025-02-25T12:56:08.021Z (2 months ago)
- Topics: kubernetes, nodeselector, pods, watcher
- Language: Python
- Homepage:
- Size: 69.3 KB
- Stars: 2
- Watchers: 3
- Forks: 1
- Open Issues: 8
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# ☸️ Kubernetes NodeSelector ensurer
`k8s-nodeselector-ensurer` is a watcher whose role is to ensure that all
the pods running on the kubernetes cluster adhere to their `nodeSelector` label list.
That is, when a node stops offering a particular label (e.g. `foo=bar`) that a pod
needs, `k8s-nodeselector-ensurer` will detect this and thus kill the aforementionned pod.
# Example
```bash
$ kubectl apply -f nodeselector-ensurer.yml
deployment.apps/nodeselector-ensurer created
$ kubectl apply -f example.yml
deployment.apps/test-nodeselector created
$ kl get pod -l name=test-nodeselector
NAME READY STATUS RESTARTS AGE
test-nodeselector-5c68758d5f-9dhlw 0/1 Pending 0 3s
$ kl label node node-c24v-2 foo=bar
node/node-c24v-2 labeled
$ kl get pod -l name=test-nodeselector
NAME READY STATUS RESTARTS AGE
test-nodeselector-5c68758d5f-9dhlw 1/1 Running 0 28s
$ kl label node node-c24v-2 foo=notbar --overwrite
node/node-c24v-2 labeled
$ kl get pod -l name=test-nodeselector
NAME READY STATUS RESTARTS AGE
test-nodeselector-5c68758d5f-287sj 0/1 Pending 0 88s
```
# How to use it ❓
First of all, this is a POC, so don't use it. But if you still want to, follow these steps:
## Installation
To install this watcher, simply `kubectl apply -f nodeselector-ensurer.yml`
## Pod usage
By default, this watcher does not change the existing nodeSelector logic.
In order to mark pods as being killable when their `nodeSelector` no longer matches their node's,
just annote the pods with `frankkkkk.nodeSelectorDuringExecution: "true"`
# Why this watcher ⁉️
Kubernetes plans on offering the extra `requiredDuringSchedulingRequiredDuringExecution` affinity type.
However it is still (as of k8s 1.18) still not implemented in the node scheduler 😞.
At first I wanted the watcher to implement this feature, however the Pod resource definition is
not customizable.