What Is Deregistration Delay?
Deregistration delay, also called connection draining, is the interval a load balancer waits before fully removing a backend from rotation. During this window it stops sending new requests but lets existing ones complete. This prevents cutting off active connections when an instance is being replaced or scaled in.
Why it matters
Draining avoids dropped requests and broken responses during a deploy or scale-down. Setting it too short cuts off slow requests; too long stalls the rollout.
Where this shows up in a real pipeline
A definition is only useful if you can recognise the thing in your own logs. In practice this concept surfaces when a job behaves differently between a local run and a runner, which is where most CI debugging starts.
- Compare a local run against a CI run of the same command before assuming a tool is at fault.
- Runner environments differ in shell, PATH, TTY availability, available memory, and disk. Most CI-only behaviour traces to one of those five.
- Pin tool versions so an environment difference cannot be introduced by an unrelated update.