Connection Draining – Concept, Features, Benefits, Limitations, and
Use Cases
Concept
Connection Draining (also known as deregistration delay in AWS ELB) is
the process of gracefully removing or shutting down backend servers
(e.g., in a load balancer or auto-scaling group) without interrupting active
connections. It allows existing requests to complete while preventing new
ones from being sent to the target instance or service.
Key Features
- Graceful Decommissioning: Allows in-flight requests to complete before
terminating the instance or deregistering it from the load balancer.
- Timeout Control:
A maximum timeout period can be set after which existing connections are
forcefully closed.
- State Awareness:
Integrates with health checks and scaling policies to avoid routing
traffic to unhealthy or decommissioned instances.
Benefits
|
Benefit |
Description |
|
Improved User Experience |
Prevents dropped or abruptly
terminated user sessions. |
|
Safe Deployments |
Enables rolling deployments and
updates with zero downtime. |
|
Smooth Scaling |
Facilitates seamless removal of
instances during scale-in events. |
|
Operational Stability |
Helps maintain service reliability
and uptime during maintenance or upgrades. |
Limitations
|
Limitation |
Description |
|
Added Latency |
Waiting for draining to complete
can delay instance shutdown. |
|
Client Awareness |
Clients may still experience
issues if they retry requests during the draining phase without proper retry
logic. |
|
Not Foolproof |
Long-lived connections may delay
or prevent instance termination if not handled correctly. |
|
Configuration Complexity |
Requires proper integration and
tuning (e.g., drain timeout, app support for graceful termination). |
Use Cases
|
Use Case |
Description |
|
Rolling Deployments |
Replace app versions one by one
without dropping traffic. |
|
Auto Scaling (Down) |
Remove instances without
interrupting user sessions. |
|
Maintenance Windows |
Perform updates or reboots without
affecting users. |
|
Fault Isolation |
Remove unhealthy instances while
keeping active requests intact. |
twtech - Example for: AWS Network Load Balancer
In AWS:
- Connection Draining for ALB
is called "Deregistration Delay".
- Default is 300 seconds.
- During this time, no new connections are sent to the instance, but existing ones are allowed to complete.
- Connection Draining is a direct concept associated with CLB (classic load balancer)
How
twtech configures the frequency of Deregistration delay or routing to the target group:
Edit
target group attributes:twtech-nlb-tg
From: Deregistration delay (draining interval).
As of May 2025,this settings vary from 0 - 3600 seconds
The time to wait for in-flight requests to
complete while deregistering a target. During this time, the state of the
target is draining.
To:
No comments:
Post a Comment