Scheduled message delivery

QStash sends HTTP requests on a fixed schedule or after a delay, holding each job until its delivery time. A Beam agent watches upcoming schedules and payload records, and when a job is due it applies the customer rule to confirm the target endpoint and headers before releasing the call. Delivery outcomes get written back to the run log, and successful callbacks close the item on their own. When a schedule points at an endpoint the rule does not recognize, or a payload fails validation, the agent holds the job and notifies an engineer rather than sending an unverified request out to a live service.

Failure retries and dead-lettering

QStash retries requests that return errors and moves exhausted messages into a dead-letter queue for later inspection. A Beam agent reads retry counts and response codes, and on a repeated failure it applies the approved rule to requeue, delay, or park the message. Each decision is recorded against the original job so the history stays clear for anyone auditing it. Messages that reach the dead-letter queue for reasons outside the rule, such as an unfamiliar status code or a downstream outage, are escalated to a human with the response details attached for a manual call.

Queue backpressure controls

QStash limits how many messages flow to an endpoint at once so downstream services are not overwhelmed by a spike. A Beam agent reads queue depth and current rate settings, and when volume crosses an approved threshold it adjusts the parallelism or pauses intake per the customer rule. The change and its reason are logged for review afterward. If a queue keeps growing after throttling, or the target service reports sustained errors, the agent stops making automatic changes and routes the situation to an on-call engineer with the recent metrics summarized for them.