Graceful Service Shutdown Strategies in Spring Cloud

In production environments, ensuring that service upgrades do not disrupt user experience is crucial; graceful shutdown aims to make the upgrade invisible to users.

The concept of graceful shutdown is relative—while kill PID is less abrupt than kill -9 PID, it is still not fully graceful.

Method 1: kill PID

Use kill on the Java process ID. This leverages Spring Boot’s shutdown hook, but when Eureka is used, other services may still see the instance for up to 90 seconds, making it insufficiently graceful.

Method 2: /shutdown endpoint

Spring Boot provides a /shutdown actuator endpoint. Enable it by adding the following configuration to application.yml:

management:
  endpoint:
    shutdown:
      enabled: true
  endpoints:
    web:
      exposure:
        include: shutdown

Trigger the shutdown with a POST request, e.g., curl -X POST http://<your-service>/actuator/shutdown.

This method fundamentally uses the same shutdown hook as Method 1.

Method 3: /pause endpoint

Spring Boot also offers a /pause endpoint. Enable it (and the required restart endpoint) in application.yml:

management:
  endpoint:
    # enable pause endpoint
    pause:
      enabled: true
    # enable restart endpoint, required by pause
    restart:
      enabled: true
  endpoints:
    web:
      exposure:
        include: pause,restart

Send a POST request: curl -X POST http://<your-service>/actuator/pause. The instance is marked DOWN in the service registry but continues to run, allowing traffic to drain before a final kill.

Method 4: /service-registry endpoint

Expose the /service-registry endpoint via configuration:

management:
  endpoints:
    web:
      exposure:
        include: service-registry

Mark the service as DOWN with a POST request, for example:

curl -X "POST" "http://localhost:8000/actuator/service-registry?status=DOWN" \
  -H "Content-Type: application/vnd.spring-boot.actuator.v2+json;charset=UTF-8"

This approach is considered the most graceful: first mark the instance DOWN, monitor traffic until it reaches zero, then safely upgrade or stop the service.

In practice, these methods work best when multiple instances are deployed; with a single instance the benefit of graceful shutdown diminishes.

The author also promotes related PDF collections on Spring Cloud, Spring Boot, and MyBatis, encouraging readers to follow the public account for more resources.