Job-scripts-jiriaf

Job Scripts

Job scripts include the definitions for both the configMap and the pod associated with a particular job.

Computing Sites Running Containers in User Space (Common in HPC Environments Using Singularity or Shifter)

kind: ConfigMap
apiVersion: v1
metadata:
  name: shifter-stress
data:
  stress.sh: |
    #!/bin/bash
    export NUMBER=$2
    export TIME=$1
    shifter --image="jlabtsai/stress:latest" --entrypoint
---
apiVersion: v1
kind: Pod
metadata:
  name: some-name # Job Name Here
spec:
  containers:
    - name: c1
      image: shifter-stress
      command: ["bash"]
      args: ["300", "2"] # Time and cpu for stress
      volumeMounts:
        - name: shifter-stress
          mountPath: shifter-stress
      resources:
        limits:
          cpu: "2"
          memory: 1Gi
        requests:
          cpu: "1" # Number of CPUs Here as well
          memory: 1Gi # Memory Here 
  volumes:
    - name: shifter-stress
      configMap:
        name: shifter-stress
  nodeSelector:
    kubernetes.io/role: agent
  affinity:
    nodeAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:
        nodeSelectorTerms:
        - matchExpressions:
# Below are the labels for the node, corresponding to the jiriaf labels
          - key: jiriaf.nodetype
            operator: In
            values:
            - "cpu"
          - key: jiriaf.site
            operator: In
            values:
            - "Local"
# Below should be commented out if the JIRIAF_WALLTIME is set to 0
    ###
        - key: jiriaf.alivetime 
            operator: Gt
            values:
            - "30"
    ###
  tolerations:
    - key: "virtual-kubelet.io/provider"
      value: "mock"
      effect: "NoSchedule"
  restartPolicy: Never

Compute Sites Utilizing Docker or Other Container Runtimes Operating in Root Space

Two containers are instantiated in this process. The first container is dedicated to the user's job, while the second container's role is to adjust the PGID of the first container. This adjustment ensures that metrics from the correct processes running in the first container are accurately collected.

kind: ConfigMap
apiVersion: v1
metadata:
  name: docker-stress
data:
  stress.sh: |
    #!/bin/bash
    export PGID_FILE=$3
    docker run -d --rm -e NUMBER=$2 -e TIME=$1 jlabtsai/stress:latest > /dev/null
    ## find the last container id
    export CONTAINER_ID=$(docker ps -l -q)
    docker inspect -f '{{.State.Pid}}' $CONTAINER_ID > $3
    sleep $1
---
kind: ConfigMap
apiVersion: v1
metadata:
  name: get-pgid
data:
  stress.sh: |
    #!/bin/bash
    sleep 3
    cp $1 $2
---
apiVersion: v1
kind: Pod
metadata:
  name: p # Job Name Here
spec:
  containers:
    - name: c1
      image: docker-stress
      command: ["bash"]
      args: ["300", "2", "~/p/containers/c1/p"]
      volumeMounts:
        - name: docker-stress
          mountPath: docker-stress
      resources:
        limits:
          cpu: "2"
          memory: 1Gi
        requests:
          cpu: "1" # Number of CPUs Here as well
          memory: 1Gi # Memory Here 
    - name: c2
      image: get-pgid
      command: ["bash"]
      args: ["~/p/containers/c1/p", "~/p/containers/c1/pgid"]
      volumeMounts:
        - name: get-pgid
          mountPath: get-pgid
      resources:
        limits:
          cpu: "2"
          memory: 1Gi
        requests:
          cpu: "1" # Number of CPUs Here as well
          memory: 1Gi # Memory Here 
  volumes:
    - name: docker-stress
      configMap:
        name: docker-stress
    - name: get-pgid
      configMap:
        name: get-pgid
  nodeSelector:
    kubernetes.io/role: agent
  affinity:
    nodeAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:
        nodeSelectorTerms:
        - matchExpressions: 
# Below are the labels for the node, corresponding to the jiriaf labels
          - key: jiriaf.nodetype
            operator: In
            values:
            - "cpu"
          - key: jiriaf.site
            operator: In
            values:
            - "Local"
# Below should be commented out if the JIRIAF_WALLTIME is set to 0
    ###
        - key: jiriaf.alivetime 
            operator: Gt
            values:
            - "30"
    ###
  tolerations:
    - key: "virtual-kubelet.io/provider"
      value: "mock"
      effect: "NoSchedule"
  restartPolicy: Never

Reference

Virtual-kubelet-cmd