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 # (1)
metadata:
  name: job1
  namespace: runai-team-a # (2)
spec:
  template:
    spec:
      containers:
      - name: job1-container
        image: gcr.io/run-ai-demo/quickstart
        resources:
          limits:
            nvidia.com/gpu: 1 # (4)
      restartPolicy: Never
      schedulerName: runai-scheduler # (3)

This is a Kubernetes Job.
Namespace: Replace runai-team-a with the name of the Run:ai namespace for the specific Project (typically runai-<Project-Name>).
The job to be scheduled with the Run:ai scheduler.
To run with half a GPU replace 1 with "0.5" (with apostrophes).

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 # (1)
metadata:
  name: inference-1
  namespace: runai-team-a # (2)
spec:
  replicas: 1
  selector:
    matchLabels:
      app: inference-1
  template:
    metadata:
      labels:
        app: inference-1
    spec:
      containers:
        - resources:
            limits:
              nvidia.com/gpu: 1 # (4)
          image: runai/example-marian-server
          imagePullPolicy: Always
          name: inference-1
          ports:
            - containerPort: 8888
      schedulerName: runai-scheduler # (3)

---
apiVersion: v1
kind: Service # (5)
metadata:
  labels:
    app: inference-1
  name: inference-1
spec:
  type: ClusterIP
  ports:
    - port: 8888
      targetPort: 8888
  selector:
    app: inference-1

This is a Kubernetes Deployment.
Namespace: Replace runai-team-a with the name of the Run:ai namespace for the specific Project (typically runai-<Project-Name>).
The job to be scheduled with the Run:ai scheduler.
To run with half a GPU replace 1 with "0.5" (with apostrophes).
This example also contains the creation of a service to connect to the deployment. It is not mandatory.

To submit the Deployment run:

kubectl apply -f deployment1.yaml

Limitations¶

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