I have a sporadic bug occuring on my Jenkins installation for months now:

java.net.ProtocolException: Expected HTTP 101 response but was '500 Internal Server Error'
at okhttp3.internal.ws.RealWebSocket.checkResponse(RealWebSocket.java:229)
at okhttp3.internal.ws.RealWebSocket$2.onResponse(RealWebSocket.java:196)
at okhttp3.RealCall$AsyncCall.execute(RealCall.java:206)
at okhttp3.internal.NamedRunnable.run(NamedRunnable.java:32)
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
at java.lang.Thread.run(Thread.java:748)
io.fabric8.kubernetes.client.KubernetesClientException: error dialing backend: EOF

I believe it was already reported in these threads and I understand that this is caused by an HTTP 500 returned by the kubernetes API:

• https://issues.jenkins-ci.org/browse/JENKINS-39844
• https://stackoverflow.com/questions/50949718/kubernetes-gke-error-dialing-backend-eof-on-random-exec-command

However, after further investigation, I am sure now that the bug occurs only when the cluster autoscaler is on and more precisely when the autoscaler scales down while a Jenkins build is running. It maybe an edge case.

To fix this, I set the annotation on all my pods in the podTemplate yaml:

cluster-autoscaler.kubernetes.io/safe-to-evict: "false" 

However, it didn't protect them. So I am trying now to setup a PodDisruptionBudget for each of my slave pod to protect them from eviction.

But, when passing the PDB into the podTemplate yaml it is just totally ignored. How can I protect my jenkins slave pods from eviction?