Implementing Zero-Downtime Deployments for Spring Boot

Intermediate·2 min read·

Fast recall

Quick revision

Core answer

Zero-downtime deployment requires: rolling update strategy, Spring Boot graceful shutdown, readiness probes, backward-compatible DB migrations, and a preStop hook to drain in-flight requests.

Interview Answer2 min

Zero-downtime deployment requires four things working together: a rolling update strategy at the infrastructure level, graceful shutdown in the application, backward-compatible database changes, and correct health check configuration. Any one of these missing causes dropped requests.

Kubernetes rolling update

strategy.type: RollingUpdate with maxUnavailable: 0 means Kubernetes never takes a pod out of rotation until a new pod is ready; maxSurge: 1 allows one extra pod during the rollout — which means at least desired pods are always serving traffic
The new pod must pass its readiness probe before Kubernetes stops routing to the old pod; configure readinessProbe.httpGet.path: /actuator/health/readiness with appropriate initialDelaySeconds for Spring Boot's startup time

Graceful shutdown in Spring Boot

When Kubernetes sends SIGTERM to a pod, Spring Boot should wait for in-flight requests to complete before shutting down — configure server.shutdown=graceful and spring.lifecycle.timeout-per-shutdown-phase=30s
Without graceful shutdown, SIGTERM immediately kills the process, dropping any request that was being processed at that moment

Database migration compatibility

During a rolling deployment, old and new versions run simultaneously — which means any database change must be backward-compatible with the old version; adding a NOT NULL column without a default breaks the old version
Follow the expand-contract pattern: add the new column as nullable (old version ignores it); deploy the new version that writes to it; add the NOT NULL constraint after all pods are on the new version

Connection draining

Configure the load balancer to drain connections from pods being terminated — Kubernetes sends SIGTERM to the pod and simultaneously removes it from the service endpoints; in-flight requests to the pod continue until they complete or the drain timeout expires
Set terminationGracePeriodSeconds to at least the load balancer's drain timeout (typically 30-60 seconds)

Interview-ready guide

Interview answer

Kubernetes rolling update

strategy.type: RollingUpdate with maxUnavailable: 0 means Kubernetes never takes a pod out of rotation until a new pod is ready; maxSurge: 1 allows one extra pod during the rollout — which means at least desired pods are always serving traffic
The new pod must pass its readiness probe before Kubernetes stops routing to the old pod; configure readinessProbe.httpGet.path: /actuator/health/readiness with appropriate initialDelaySeconds for Spring Boot's startup time

Graceful shutdown in Spring Boot

When Kubernetes sends SIGTERM to a pod, Spring Boot should wait for in-flight requests to complete before shutting down — configure server.shutdown=graceful and spring.lifecycle.timeout-per-shutdown-phase=30s
Without graceful shutdown, SIGTERM immediately kills the process, dropping any request that was being processed at that moment

Database migration compatibility

During a rolling deployment, old and new versions run simultaneously — which means any database change must be backward-compatible with the old version; adding a NOT NULL column without a default breaks the old version
Follow the expand-contract pattern: add the new column as nullable (old version ignores it); deploy the new version that writes to it; add the NOT NULL constraint after all pods are on the new version

Connection draining

Configure the load balancer to drain connections from pods being terminated — Kubernetes sends SIGTERM to the pod and simultaneously removes it from the service endpoints; in-flight requests to the pod continue until they complete or the drain timeout expires
Set terminationGracePeriodSeconds to at least the load balancer's drain timeout (typically 30-60 seconds)

03 · Deep dive