Fix KubernetesSlave NPE on getting Workspace after Jenkins Master Restart

[jeffersongirao]

@carlossg Thanks a lot for your work on the kubernetes-plugin, it is greatly appreciated! 🙂 Could you please let me know your thoughts on this?

How to reproduce the issue

Kubernetes-plugin 1.14.9
Jenkins LTS 2.164.1

• Configure a Kubernetes PodTemplate in the way it stays up after the job finishes (idleMinutes setting)
• Run a Job that uses such Kubernetes PodTemplate to provision an Agent (that should work good)
• Restart Jenkins Master (for instance killing the Jenkins process inside the pod)
• When the Jenkins Master comes back online, the Agent should be able to connect back to it and it is reported as being on-line both from the Agent logs and in the Jenkins master /computer/<agent_name>/
• When running Job that tries to use that agent you should get an error saying:

ERROR: Issue with creating launcher for agent <agent name>. Computer has been disconnected
Building remotely on <agent name> (k8s)ERROR: <agent name> seems to be offline
Finished: FAILURE

Root cause

When fetching a Workspace for a Job we get a null value that triggers the aforementioned error. If we go deep on how Jenkins fetch the Workspace root path for Slaves, Jenkins queries a Map with Computers and KubernetesSlave (Node) is a key for such Map.

When Jenkins restarts and the agent gets reconnected that Map is re-instantiated, the KubernetesSlave instance is one of the keys but it is not equals() and has a different hashCode(). They differ on the template private field given that PodTemplate does not implements an equals()/hashCode().

Proposed Fix

Perhaps it is reasonable to assume that if agents have the same name they are the same independently of the PodTemplate definition. I have changed both equals() and hasCode() for KubernetesSlave in tune with that.

An alternative would be to implement equals()/hashCode() on the PodTemplate, let me know if you would like me to do it that way instead.

[Vlatombe]

Hi @jeffersongirao it would be great if you could provide a test showing this bug in order to avoid future regressions. It could be added to https://github.com/jenkinsci/kubernetes-plugin/blob/master/src/test/java/org/csanchez/jenkins/plugins/kubernetes/pipeline/RestartPipelineTest.java

[carlossg]

Thanks! I think you can have multiple agents with the same name so equals()/hashCode() on the PodTemplate would be better

[jeffersongirao]

@carlossg I think that I actually need to take it back about the possibility of solving this at the PodTemplate level, the comparison there happens in Map of <Node,Computer>, that is up in the hierarchy and we don't have the template field available (https://github.com/jenkinsci/jenkins/blob/jenkins-2.164.1/core/src/main/java/hudson/model/AbstractCIBase.java#L176). How do you suggest to proceed?

[jeffersongirao]

@Vlatombe I have made an attempt to reproduce it in the tests how you suggested, unfortunately it seems like the RestartableJenkinsRule does not have the exact same behaviour as when I'm running a real instance. Wondering if there is anything cached in terms of reference to registered computers in that, any pointers?

[Vlatombe]

If I understand correctly the original report, this should have been fixed as a side-effect of #783 (1.25.6 and later)