Connection Draining / Deregistration Delay

• CLB calls it Connection Draining

• ALB and NLB call it Deregistration Delay

• Ensures in-flight requests complete before an instance is removed (deregistered or marked unhealthy)

• LB stops sending new requests to the instance while it’s draining

• Configurable time: 1–3600 seconds (default 300s)

• Disable by setting to 0

• Use a shorter delay for short-lived requests to speed up deployments

Use Case

• Gracefully shut down instances during:
• Deployments
• Auto Scaling down events
• Maintenance operations

• Prevents user disruption by letting current requests finish

Behavior

• Draining state:
• Existing connections are served until completion
• No new connections sent to that target
• New traffic is routed to healthy targets only