[JENKINS-66337] Kubernetes Plugin IllegalStateException: Not expecting pod template to be null at this point. On master restart for long lasting slave.

Type: Bug
Resolution: Unresolved
Priority: Critical
Component/s: kubernetes-plugin
Labels:
None
Environment:
Jenkins 2.305
kubernetes-plugin: 1.30.1
slave-agent: 4.10

Similar Issues:
Powered by SuggestiMate

Show

We are in the process of upgrading our Jenkins instance and plugins version.

We have encountered this bug for long-lasting pods (ex: idleMinutes: "30").

If there is no slave matching the agent label. Everything is running properly.

A new agent pod is provided
Closure is properly run on new agent pod.

Please note that the agent will properly run all pipeline that will be provided from this point.

Jenkins properly find the already running agent (Yeah!)
Closure is properly executed.

If we restart Jenkins (master), while the agent was already provided (matching label):

Jenkins properly find the already running agent (Yeah!)

PodTemplate is getting the exception:

java.lang.IllegalStateException: Not expecting pod template to be null at this point

Here is a pipeline example :

def slavePodForYou(steps, body){
    String image = 'jenkins/jnlp-slave:latest'
    
    def kubeLabel = "slavenode"
    steps.podTemplate(name: "slavenode", label: kubeLabel, namespace: "default",
               containers: [
                   steps.containerTemplate(
                    name: "jnlp",
                    image: image,
                    ttyEnabled: false,
                    args: '${computer.jnlpmac} ${computer.name}',
                    resourceRequestCpu: '100m',
                    resourceRequestMemory: '250Mi'
                    )
               ],
               yaml: """
                       metadata:
                         labels:
                           job: jenkins
                     """.stripIndent(),
               showRawYaml: false,
               idleMinutes: "30",
           ){
               body.call(kubeLabel)
           }   
}
slavePodForYou(this){ kubeLabel ->
    node(kubeLabel){
        sh "echo 'chicken'"
    }
}

Note: Everything was properly working with version: 1.25.1 of the Kubernetes plugin.

Note also: jnlp agent is properly rejoining the master, after the restart.

Note also: we are not using JCasc. All our pod definition are done in a shared library. Only the cloud cluster definition is done in the UI.

Pascal Laporte created issue - 2021-08-09 17:03

Pascal Laporte made changes - 2021-08-09 17:04

Description

Original: We are in the process of upgrading our Jenkins instance and plugins version.

We have encountered this bug for long-lasting pods (ex: idleMinutes: "30").

If there is no slave matching the agent label. Everything is running properly.
* A new agent pod is provided
* Closure is properly run on new agent pod.

Please note that the agent will properly run all pipeline that will be provided
* Jenkins properly find the already running agent (Yeah!)
* Closure is properly executed.

If we restart Jenkins (master), while the agent was already provided (matching label):
* Jenkins properly find the already running agent (Yeah!)
* PodTemplate is getting the exception:
{code:java}
java.lang.IllegalStateException: Not expecting pod template to be null at this point{code}

Here is a pipeline example :

{code:java}
def slavePodForYou(steps, body){
    String image = 'jenkins/jnlp-slave:latest'

    def kubeLabel = "slavenode"
    steps.podTemplate(name: "slavenode", label: kubeLabel, namespace: "default",
               containers: [
                   steps.containerTemplate(
                    name: "jnlp",
                    image: image,
                    ttyEnabled: false,
                    args: '${computer.jnlpmac} ${computer.name}',
                    resourceRequestCpu: '100m',
                    resourceRequestMemory: '250Mi'
                    )
               ],
               yaml: """
                       metadata:
                         labels:
                           job: jenkins
                     """.stripIndent(),
               showRawYaml: false,
               idleMinutes: "30",
           ){
               body.call(kubeLabel)
           }
}
slavePodForYou(this){ kubeLabel ->
    node(kubeLabel){
        sh "echo 'chicken'"
    }
}
{code}

Note: Everything was properly working with version: 1.25.1 of the Kubernetes plugin.

Note also: jnlp agent is properly rejoining the master, after the restart.

Note also: we are not using JCasc. All our pod definition are done in a shared library. Only the cloud cluster definition is done in the UI.

New: We are in the process of upgrading our Jenkins instance and plugins version.

We have encountered this bug for long-lasting pods (ex: idleMinutes: "30").

If there is no slave matching the agent label. Everything is running properly.
* A new agent pod is provided
* Closure is properly run on new agent pod.

Please note that the agent will properly run all pipeline that will be provided from this point.
* Jenkins properly find the already running agent (Yeah!)
* Closure is properly executed.

*If we restart Jenkins* (master), while the agent was already provided (matching label):
* Jenkins properly find the already running agent (Yeah!)
* PodTemplate is getting the exception:
{code:java}
java.lang.IllegalStateException: Not expecting pod template to be null at this point{code}

Here is a pipeline example :

{code:java}
def slavePodForYou(steps, body){
    String image = 'jenkins/jnlp-slave:latest'

    def kubeLabel = "slavenode"
    steps.podTemplate(name: "slavenode", label: kubeLabel, namespace: "default",
               containers: [
                   steps.containerTemplate(
                    name: "jnlp",
                    image: image,
                    ttyEnabled: false,
                    args: '${computer.jnlpmac} ${computer.name}',
                    resourceRequestCpu: '100m',
                    resourceRequestMemory: '250Mi'
                    )
               ],
               yaml: """
                       metadata:
                         labels:
                           job: jenkins
                     """.stripIndent(),
               showRawYaml: false,
               idleMinutes: "30",
           ){
               body.call(kubeLabel)
           }
}
slavePodForYou(this){ kubeLabel ->
    node(kubeLabel){
        sh "echo 'chicken'"
    }
}
{code}

Note: Everything was properly working with version: 1.25.1 of the Kubernetes plugin.

Note also: jnlp agent is properly rejoining the master, after the restart.

Note also: we are not using JCasc. All our pod definition are done in a shared library. Only the cloud cluster definition is done in the UI.

Jenkins

Details

Description

Attachments

Activity

People

Dates