Continuous availability.
WHAT IT REQUIRES
Several instances A load balancer Graceful shutdown Backward-compatible changes
WHAT GRACEFUL SHUTDOWN MEANS
Stop accepting new requests, finish in-flight ones, close connections, then exit.
WHY
Abrupt termination drops requests that were being served.
WHAT A ROLLING DEPLOYMENT IS
Replacing instances progressively, so some always serve.
WHAT TO CONFIGURE
Health checks distinguishing started from ready.
WHY
An instance receiving traffic before it is ready returns errors.
WHAT THE ALTERNATIVE APPROACHES ARE
Running two environments and switching traffic Releasing to a subset first, and expanding if healthy
WHAT THAT SECOND ONE PROVIDES
Limiting the blast radius of a defective release.
WHAT TO ENSURE ABOUT SCHEMA CHANGES
Compatibility with both old and new code.
WHY
Both run simultaneously during the rollout.
WHAT TO MONITOR DURING
Error rates per version.
WHAT TO PREPARE
An immediate rollback.
WHAT TO ACCEPT
That not every application needs this.
