Liveness Checks (Kubernetes)

Kubernetes can use a function called a livenessProbe to help identify unhealthy pods or pods that are having issues.

The below livenessProbe uses a httpGet to check for response codes between 200 and 400. The probe waits 3 seconds (initialDelaySeconds) before starting. The probe checks for a response every 3 seconds (periodSeconds). The probe flags an unhealthy container after 3 response failures (failureThreshold), e.g. it gets three 500 server errors.

livenessProbe:
          httpGet:
            path: /
            port: 5000
          initialDelaySeconds: 3
          periodSeconds: 3
          failureThreshold: 3

Here is the livenessProbe in the context of my previous deployment:

apiVersion: apps/v1
kind: Deployment
metadata:
  name: dep-flask-hello
spec:
  replicas: 5
  selector:
    matchLabels:
      name: geektechstuff-flask
  template:
    metadata:
      labels:
        name: geektechstuff-flask
    spec:
      containers:
      - name: flask-hello-world
        image: geektechstuff/flask-hello-world:latest
        livenessProbe:
          httpGet:
            path: /
            port: 5000
          initialDelaySeconds: 3
          periodSeconds: 3
          failureThreshold: 3
        ports:
        - containerPort: 5000
        resources:
          requests:
            cpu: 0.2
            memory: 128Mi
          limits:
            cpu: 1.0
            memory: 256Mi

Using the describe command on a pod will show what liveness options the pod is using:

kubectl describe pod PODNAME_HERE

Liveness Checks (Kubernetes) — kubectl describe pod, with Liveness showing towards the bottom of the image

Kubernetes also offers other types of liveness probes, for example if the container does not handle HTTP requests. A tcpSocket probe may be more appropriate in these situations:

livenessProbe:
          tcpSocket:
            port: 5000
          initialDelaySeconds: 5
          periodSeconds: 5
          failureThreshold: 3

Replacing the httpGet in my previous example, the full deployment now looks like:

apiVersion: apps/v1
kind: Deployment
metadata:
  name: dep-flask-hello
spec:
  replicas: 5
  selector:
    matchLabels:
      name: geektechstuff-flask
  template:
    metadata:
      labels:
        name: geektechstuff-flask
    spec:
      containers:
      - name: flask-hello-world
        image: geektechstuff/flask-hello-world:latest
        livenessProbe:
          tcpSocket:
            port: 5000
          initialDelaySeconds: 5
          periodSeconds: 5
          failureThreshold: 3
        ports:
        - containerPort: 5000
        resources:
          requests:
            cpu: 0.2
            memory: 128Mi
          limits:
            cpu: 1.0
            memory: 256Mi

A kubectl describe will now show that tcp socket is being used to check the liveness.

kubectl describe now shows liveness using tcp-socket

More information on Liveness can be found at: https://kubernetes.io/docs/tasks/configure-pod-container/configure-liveness-readiness-startup-probes/