What happens to the rest of a middleware chain when a hook answers a request itself instead of delegating onward?

## What short-circuiting is Every hook in a chain holds the remainder of that chain as something it may call. Delegating is therefore a choice, not an obligation. **Short-circuiting** is the case where a hook makes a decision, builds a response itself, and returns without ever making that call. It is an ordinary return, not an error: no exception is involved and nothing exceptional has to be signalled. Typical reasons to do it: the caller is not authenticated or not permitted; the caller has exceeded a quota and gets 429; a cached representation is still fresh and can be served without touching the handler; the request should be redirected; a conditional request can be answered with 304. In each case the work further inside the chain is not merely unnecessary — it may be actively undesirable, which is the whole point of putting the decision outside it. ## What is skipped and what still runs | Element | Before-phase | After-phase | |---|---|---| | hooks registered outside the short-circuiting hook | already ran | still runs | | the short-circuiting hook itself | ran, and produced the response | its own remaining code runs | | hooks registered inside it | never runs | never runs | | the route handler | never invoked | n/a | The crucial asymmetry is the last two rows against the first. Skipping is scoped to what sits **inside** the hook that stopped the chain; it says nothing about the layers enclosing it. Those layers cannot tell the difference structurally — from where they sit, a call returned with a response, which is what they expected in the first place. ## Why the outer after-phases still running matters This is a feature, not an accident, and it is what makes an outer position the right home for anything that must observe all traffic: - **Access logging and metrics stay complete.** Rejected and throttled requests are still recorded, with their real status, because the logging hook resumes normally. - **Total timing still works.** An outermost timer brackets the short-circuit exactly as it brackets a full request. - **Response-wide decoration still applies.** A hook that stamps a header onto every outgoing response continues to do so on the early response, as long as it is outside the short-circuit. The converse is the trap: anything that would have been contributed by a hook **inside** the short-circuit is simply absent from that response. If a header, an encoding step or a correlation value is attached by an inner hook, the short-circuited response goes out without it, and a test that only exercises the happy path will never notice. The fix is positional — move the contributor outward so it encloses the decision — rather than duplicating its logic into the short-circuiting hook. ## Writing one correctly 1. Decide as early as the information allows, but no earlier than the data you need to decide is available. 2. Build a **complete** response: status, the headers that response needs, and a body if it has one. Nothing downstream will fill in gaps. 3. Return without delegating — and make sure the control flow really does return, rather than falling through into the delegation call. 4. Check what you are skipping. If something inside the chain was contributing to every response, this response now lacks it. ## The classic failure: responding and delegating anyway Writing a response and then still calling the rest of the chain is the most common bug in this area. The chain runs on regardless — a response already existing does not by itself cancel the remaining pipeline in most frameworks — and the handler eventually tries to write a second response into the same exchange. Depending on the framework you then get the second write silently dropped, an error about the response already being committed, or two bodies concatenated on the wire. Symptoms are confusing precisely because the hook looks correct in isolation; the defect is in what happened after it. A related subtlety: short-circuiting from a hook that has already flushed part of the response is not really a short circuit at all. Once bytes are on the wire, the status and headers are fixed and an early decision can no longer replace them. ## What interviewers are checking They want to hear that delegation is explicit and optional, that skipping is scoped inward, that the enclosing after-phases still run, and that the candidate has thought about the second-order effect — which contributions the response now silently lacks. A candidate who answers only 'the chain stops' has described half the mechanism and none of the consequences.