18.3.2 Enabling graceful shutdown

We have several ways in which to make Spring applications Kubernetes friendly, but the two most essential things you’ll want to do are to enable graceful shutdown as well as liveness and readiness probes.

At any time, Kubernetes may decide to shut down one or more of the pods that your application is running in. That may be because it senses a problem, or it might be because someone has explicitly requested that the pod be shut down or restarted. Whatever the reason, if the application on that pod is in the process of handling a request, it’s poor form for the pod to immediately shut down, leaving the request unhandled. Doing so will result in an error response to the client and require that the client make the request again.

Instead of burdening the client with an error, you can enable graceful shutdown in your Spring application by simply setting the server.shutdown property to "graceful". This can be done in any of the property sources discussed in chapter 6, including in application.yml like this:

server:
  shutdown: graceful

By enabling graceful shutdown, Spring will hold off on allowing the application to shut down for up to 30 seconds, allowing any in-progress requests to be handled. After all pending requests have been completed or the shutdown time-out expires, the application will be allowed to shut down.

The shutdown time-out is 30 seconds by default, but you can override that by setting the spring.lifecycle.timeout-per-shutdown-phase property. For example, to change the time-out to 20 seconds, you would set the property like this:

spring:
  lifecycle.timeout-per-shutdown-phase: 20s

While the shutdown is pending, the embedded server will stop accepting new requests. This allows for all in-flight requests to be drained before shutdown occurs.

Shutdown isn’t the only time when the application may not be able to handle requests. During startup, for example, an application may need a moment to be prepared to handle traffic. One of the ways that a Spring application can indicate to Kubernetes that it isn’t ready to handle traffic is with a readiness probe. Next up, we’ll take a look at how to enable liveness and readiness probes in a Spring application.