Skip to content

Latest commit

 

History

History
66 lines (37 loc) · 3.47 KB

README_PROBES.md

File metadata and controls

66 lines (37 loc) · 3.47 KB
Raw

Readiness and Liveness Probes

Kubernetes supports three types of probes, that all serve different purposes, and have different pros and cons.

This article is here to describe what recommendations kube-score will make in what situations.

Different probe types

readinessProbe

This probe roughly answers the question

Is it a good idea to send traffic to this Pod right now?

A common misunderstanding is that, since Kubernetes manages the Pods, you don't need to do graceful draining of Pods during shutdown.

Without a readinessProbe you're risking that:

  • Traffic is sent to the Pod before the server has started.
  • Traffic is still sent to the Pod after the Pod has stopped.

kube-score recommends:

Every application that is targeted by a Service to:

  • Setup a readinessProbe that responds with a healthy status when: The application is fully booted, and ready to serve traffic.
  • Set interval (default: 10s), timeout (default: 1s), successThreshold (default: 1), failureThreshold (default: 3) to your needs. In the default configuration, your application will fail for 30s (+ the time it takes for the network to react), for clients to stop sending traffic to your application.
  • Don't depend on downstream dependencies, such as other services or databases in your probe. If the dependency has a hickup, or for example, a database is restarted, removing your Pods from the Load Balancers rotation will likely only make the downtime worse.
  • Handle shutdowns gracefully: Applications should start to fail the readinessProbe after receiving SIGTERM, and wait until the service is unregistered from all Load Balancers, before shutting down. As an alternative to this, a preStop hook with a sleep can be used.

livenessProbe

This probe roughly answers the question

Is the container healthy right now, or do we need to restart it?

It can be used to let Kubernetes know if your application is deadlocked, and needs to be restarted. Only the container with the failing probe will be restarted, other containers in the same Pod will be unaffected.

kube-score recommends:

  • If you don't know why you need a livenessProbe, don't configure it.
  • It should never, be the same as your readinessProbe.
  • The livenessProbe should never depend on downstream dependencies, such as databases or other services.

startupProbe (alpha since v1.16, beta since v1.17)

This probe roughly answers the question

Should we start running the livenessProbe now?

For applications that take a longer time to boot than the livenessProbes initialDelaySeconds + periodSeconds * failureThreshold, a startupProbe can be configured.

The startupProbe allows you to decrease the liveness probes initialDelaySeconds, and catch application deadlocks earlier.

As soon as the startupProbe has succeeded once, the livenessProbe will start to be executed.

kube-score recommends:

  • Configure a startupProbe if you have a livenessProbe configured.

Further reading