One Broken Endpoint Shouldn't Slow Down Everyone Else
Every outbound endpoint gets its own circuit breaker, on every plan, and so does every inbound route on Pro and Business. When one fails repeatedly, Hookbase stops delivering to it and holds its deliveries, waits, sends probes to check whether it has recovered, and resumes automatically once it has — without you touching anything.
Closed → Open → Half-Open → Closed
A circuit breaker is a small state machine that sits in front of each endpoint. It has three states, and it moves between them on its own based on what your endpoint actually does.
Closed
The normal state. Every routed event is delivered to the endpoint as usual, and the breaker just counts consecutive failures in the background.
Open
After 5 consecutive delivery failures (the default, configurable per endpoint), the circuit opens. Nothing is sent to the endpoint: new deliveries wait as held and spend no attempts.
Half-Open
Once the cooldown elapses — 60 seconds by default on outbound endpoints, also configurable — Hookbase sends probe deliveries to test whether the endpoint has come back: as many held deliveries as the success threshold.
Closed Again
After 2 consecutive successful probes (the default success threshold), the circuit closes, normal delivery resumes and the held deliveries are released, paced. A failed probe sends it back to open, and the probe waits as held again with its attempt given back.
Closed, open, and half-open are the three real states — the fourth step above is simply the circuit returning to closed. The whole cycle runs per endpoint, so recovery for one destination never waits on another.
What the Breaker Actually Does
The point isn't to give up on an endpoint. It's to stop spending outbound delivery capacity on an endpoint that is definitively down, so the endpoints that are up keep getting their events on time.
One Breaker Per Endpoint
Every outbound endpoint gets its own independent circuit breaker, on every plan, and so does every inbound route on Pro and Business. A failing endpoint is isolated on its own — the state of one endpoint never affects delivery to any other.
Configurable Thresholds
The failure threshold, the cooldown before a probe is allowed, and the number of successful probes required to close again are all configurable per endpoint or route. Outbound endpoints default to 5 failures, 60 seconds and 2 successes; inbound routes to 5 failures, 300 seconds and 2 successes.
Held, Not Attempted
While a circuit is open, new deliveries to that endpoint wait as held rather than being attempted and burning a retry against an endpoint you already know is down. Turn holds off for an endpoint or route (holdOnCircuitOpen) and they go straight to the Dead Letter Queue instead.
Nothing Loops Forever
A hold lasts no longer than the plan's retry window, from about 13 minutes on Free to about 5 days on Business. Anything still held then moves to the Dead Letter Queue, as "circuit_breaker_exhausted" for an outbound message and "circuit_breaker_open" for an inbound delivery, instead of waiting indefinitely.
Works With Retries
Retries pause once a circuit is open, so the two features reinforce each other — the retry schedule handles a blip, and the breaker takes over when the endpoint is properly down. A held delivery keeps the attempts it had left for when the endpoint comes back.
Built for Self-Serve Endpoints
If your customers register their own endpoints, one flaky customer integration is exactly the failure mode circuit breakers exist to contain.
Separate from the breaker, an outbound endpoint that has failed every delivery for 5 days, and is still failing, is disabled and its held messages go to the Dead Letter Queue. The organization's owners and admins get an email, and the endpoint can be re-enabled once it's fixed.
Circuit breakers pair directly with webhook retries, and they matter most when you run a self-serve customer portal where your customers register endpoints you don't control.
Why This Matters
Without a breaker, a single misconfigured or down endpoint keeps getting attempted — and that work has to come out of the same outbound capacity every other endpoint is waiting on.
Without circuit breakers
With Hookbase
All three thresholds — failures to open, cooldown before a probe, and successes to close — are configurable per endpoint, so a critical internal destination and a flaky third-party one can behave differently. Some deliveries are never held: messages to FIFO and polling endpoints, outbound replays and DLQ retries, and inbound deliveries on routes with failover, to warehouse destinations, from sources in transient mode or from edited replays.
Keep Delivering, Even When an Endpoint Doesn't
Add an outbound endpoint, or an inbound route on Pro, and the circuit breaker is already watching it. Get started free — no credit card required.