Support for other Kubernetes Applications¶

Introduction¶

Kubernetes has several built-in resources that encapsulate running Pods. These are called Kubernetes Workloads and should not be confused with Run:ai Workloads.

Examples of such resources are a Deployment that manages a stateless application, or a Job that runs tasks to completion.

Run:ai natively runs Run:ai Workloads. A Run:ai workload encapsulates all the resources needed to run, creates them, and deletes them together. However, Run:ai, being an open platform allows the scheduling of any Kubernetes Workflow.

How To¶

To run Kubernetes Workloads with Run:ai you must add the following to the YAML:

A namespace that is associated with a Run:ai Project.
A scheduler name: runai-scheduler.
When using Fractions, use a specific syntax for the nvidia/gpu limit.

Example: Job¶

job1.yaml

apiVersion: batch/v1
kind: Job # This is a Kubernetes Job.

metadata:
  name: job1
  namespace: runai-team-a # Namespace: Replace runai-team-a with the name of the Run:ai namespace for the specific Project (typically runai-<Project-Name>).

spec:
  template:
    spec:
      containers:
      - name: job1-container
        image: runai.jfrog.io/demo/quickstart
        resources:
          limits:
            nvidia.com/gpu: 1 # To run with half a GPU replace 1 with "0.5" (with apostrophes).

      restartPolicy: Never
      schedulerName: runai-scheduler # The job to be scheduled with the Run:ai scheduler.

To submit the Job run:

kubectl apply -f job1.yaml

You will be able to see the Job in the Run:ai User interface, including all metrics and lists

Example: Deployment¶

deployment1.yaml

apiVersion: apps/v1
kind: Deployment # This is a Kubernetes Deployment.

metadata:
  name: inference-1
  namespace: runai-team-a # Namespace: Replace runai-team-a with the name of the Run:ai namespace for the specific Project (typically runai-<Project-Name>).

spec:
  replicas: 1
  selector:
    matchLabels:
      app: inference-1
  template:
    metadata:
      labels:
        app: inference-1
    spec:
      containers:
        - resources:
            limits:
              nvidia.com/gpu: 1 # To run with half a GPU replace 1 with "0.5" (with apostrophes).

          image: runai/example-marian-server
          imagePullPolicy: Always
          name: inference-1
          ports:
            - containerPort: 8888
      schedulerName: runai-scheduler # The job to be scheduled with the Run:ai scheduler.


---
apiVersion: v1
kind: Service # This example also contains the creation of a service to connect to the deployment. It is not mandatory.

metadata:
  labels:
    app: inference-1
  name: inference-1
spec:
  type: ClusterIP
  ports:
    - port: 8888
      targetPort: 8888
  selector:
    app: inference-1

To submit the Deployment run:

kubectl apply -f deployment1.yaml

Example: Submit a Cron job via YAML¶

The cron command-line utility is a job scheduler typically used to set up and maintain software environments at scheduled intervals. Run:ai now supports submitting jobs with cron using a YAML file.

To submit a job using cron, run the following command:

kubectl apply -f <file_name>.yaml

The following is an example YAML file:

apiVersion: batch/v1
kind: CronJob
metadata:
  name: hello
spec:
  schedule: "* * * * *"
  jobTemplate:
    spec:
      template:
        metadata:
          labels:
          - (Mandatory) runai/queue: team-a
        spec:
          (Mandatory) schedulerName: runai-scheduler
          containers:
          - name: hello
            image: busybox:1.28
            imagePullPolicy: IfNotPresent
            command:
            - /bin/sh
            - -c
            - date; echo Hello from the Kubernetes cluster
          restartPolicy: OnFailure
          (Optional) priorityClassName: build / train / inference / interactivePreemptible

Limitations¶

The Run:ai command line interface provides limited support for Kubernetes Workloads.