One slow webhook endpoint no longer delays everyone's events

Released: October 9, 2026

Webhook delivery Changed

On October 9 an endpoint that stopped answering held a large share of our shared delivery capacity, and webhooks for other integrations were delayed by up to an hour. Events were not lost.

Deliveries are now isolated per endpoint host:

• At most 5 requests in flight to one host. Further deliveries wait a few seconds before their attempt starts, without using up an attempt.
• After 5 timeouts or connection errors in a row, the host is treated as unresponsive for 30 seconds (doubling up to 5 minutes). Attempts due in that window fail with error_code: "connection_error" and an error_message starting with Not sent, without a request being made, and are retried on the normal schedule.

The retry schedule, the 8-attempt limit, auto-disable and payloads are unchanged. Endpoints that answer within a couple of seconds are not affected.

Retries and unresponsive endpoints