Liveness Probes

By default, when application inside container crashes, Kubernetes will try to restart the Pod in order to serve requests.
In real-world scenarios, when application is not working due to bug and container continues to stay alive. In that scenario they are unable to serve requests.
We need a way to check periodically that the application inside the container is healthy or not
Kubernetes provides several ways to check that the application are healthy
- Set up an HTTP endpoint (for web servers)
- Use a TCP socket check (e.g., for databases)
- Write and execute a custom script inside the container to verify liveness

How to Define a Liveness Probe

In the Pod YAML definition, under the spec.containers section, define a livenessProbe:

containers:
  - name: web-server
    image: web-server-image
    livenessProbe:
      httpGet:
        path: /api/healthy
        port: 8080

livenessProbe:
  tcpSocket:
    port: 3306

livenessProbe:
  exec:
    command:
      - python3
      - check_health.py

initialDelaySeconds: If your application takes some time to warm up (e.g., 10 seconds), use this field to specify the delay before the first probe is initiated.
periodSeconds: Specifies how often the probe should be performed.
failureThreshold: By default, Kubernetes tries 3 times checking container is not ready. You can increase this threshold (e.g., to 10) to allow more retries before failing.