kube-safe: maxpodcount

This project is Admission webhook-based fix for Kubernetes 1.10-bug that causes cluster to overuse it's resources if Job fails.

Logic

1. Intercept all admission webhooks related to Pods
2. If the request is related to anything else than Job-related pod just accept and return
3. Check the count of pods related to the Job
  - If less than Job.Spec.BackoffLimit: accept and return
  - Else: Deny and return

This will cause Job to stop respawning new pods but will also freeze the job after spawning X-amount of failed pods. All pods are still there and must be manually deleted.

Installation

kubectl apply -f deploy.yml

# When the deployment starts it will automatically register itself to cluster