If we use a global pod (defined within the configure Jenkins panel).

The pod will end up in a «Error» state with the jnlp in error and the others container within still running.

So this easy example:

node("slave-go"){
  sleep 5
}

Where «slave-go» is a pod with 2 containers, a go container and a jnlp container.

The job will run with a success state, but the pod will fall in a Error state.

slave-go-2jl94             1/2       Error     0          6h

The log of the jnlp container seem to properly end:

...
Aug 22, 2017 8:30:10 PM hudson.remoting.jnlp.Main$CuiListener status
INFO: Terminated

But will immediately try to reconnect:

...
Aug 22, 2017 8:30:20 PM jenkins.slaves.restarter.JnlpSlaveRestarterInstaller$2$1 onReconnect
INFO: Restarting agent via jenkins.slaves.restarter.UnixSlaveRestarter@22f325f7
Aug 22, 2017 8:30:21 PM hudson.remoting.jnlp.Main createEngine
...

It will try all the protocols and than fail.

This is where is it interesting. If we use a podTemplate defintion, instead of a global pod definition. We will end up with a job running with sucess and a slave pod going to the proper «terminating» state.

So this example will work:

podTemplate(label: 'example_go', inheritFrom: 'slavenode',
  containers: [
    containerTemplate(name: 'golang',
    image: 'somegoimage',
    ttyEnabled: true,
    alwaysPullImage: true,
    workingDir: '/home/jenkins/workspace',
    command: 'cat')
  ]) {
node("example_go"){
  container("golang"){
    sleep 5 
  }
 }
}

Please fix it, so that slave provided by a standard global pod definition end up in the proper state.

This was properly working with version 0.10 of the kubernetes plugin.