Running a Pod

Step 1: Creating the YAML File

We first must define all the parameters for our pod. We will use this minimalist example. There are many more optional parameters that you may be interested in depending on your compute needs. The full API documentation is here

The YAML we will use for basic Pod creation is shown below. This is identical to the sample YAML in the repo here.

apiVersion: v1
kind: Pod
metadata:
  name: pod-name-sso # YOUR POD NAME HERE
spec:
  containers:
    - name: pod-name-sso # YOUR CONTAINER NAME HERE
      image: ubuntu:20.04
      command: ["sh", "-c", "echo 'Im a new pod' && sleep infinity"]
      resources:
        limits:
          memory: 12Gi
          cpu: 2
        requests:
          memory: 10Gi
          cpu: 2

We will now walk through each line/section to understand what we are specifying to Kubernetes:

apiVersion: v1

The version of the Kubernetes API that we are using.

kind: Pod

Specify the kind of resource we want to create. In this case, a pod

metadata:
    name: pod-name-sso

The metadata of the Pod. In this minimalist example, we specify only a name. Note that this must be unique in the namespace

spec:
    containers:

We are specifying what we need (spec), and our first set of needs is containers

    - name: pod-name-sso # YOUR CONTAINER NAME HERE
      image: ubuntu:20.04
      command: ["sh", "-c", "echo 'Im a new pod' && sleep infinity"]

Here we are defining the basic paramters of our first (and only) container:

• name: The name of the container, this can be the same as the pod or it can be different. In cases with multiple containers, this needs to be unique.
• image: Which docker image to load into the container. This must be publically accessible, either in Docker Hub or a public facing container registry
• command: The command to run when the pod starts. This should almost always be sleep infinity

      resources:
        limits:
          memory: 12Gi
          cpu: 2
        requests:
          memory: 10Gi
          cpu: 2

We are defining the resources for the container, in two steps. The first is the limit, i.e., the highest amount of resources that should be made available. The second is the requested amount of resources.

Note: The pod will not be scheduled on a node (i.e., will not be created) until your requests are met, so do not specify too many resources.

Note: The max amount of resources for a pod is 12 GB of RAM and 2 CPU cores (and 2 NVIDIA GPUs). Any more, and you will receive an error. If you need more resources than that, you will need to create a job.

Step 2: Creating the Pod

Once you have your YAML spec created, you can create your pod using the kubectl apply or kubectl create commands:

kubectl create -f MYFILE.yml

Note: Creation of a Pod will take time. It is not instantaneous. You can use kubectl get pods to check when your pod is running

Step 3: Attaching to the Pod

For a running pod, you can attach to the pod using the kubectl exec command:

kubectl exec -it MYPOD -- /bin/bash

From here, you can perform any/all compute that you need. Be sure to save any artifacts to your persistent volume or to copy it to your local machine with kubectl cp

Step 4: Deleting the Pod

Once you have finished using your pod and copied all necessary data from the local pod storage, delete the pod:

kubectl delete pod MYPOD

Note: On the Nautilus cluster, pods are automatically deleted 6 hours after creation. If you need longer running compute than that, you will need to use a job.