# Scheduling AWS Step Functions State Machine

> Step Functions has no off state, so the schedule enforces quiet hours by calling states:StopExecution on in-flight executions during the window. A stopped execution is terminal: nothing resumes at start, and reruns begin from step 1. The halt works on Standard workflows only, since Express executions cannot be stopped.

Source: https://zop.dev/integrations/aws/step-function/schedule
Updated: 2026-08-19

---

## There is no off switch on a state machine

A state machine is a definition, not a running resource. It bills nothing while idle and
cannot be stopped, paused or deallocated. What a schedule can control is executions. During the
off-hours window, in-flight executions are halted via `states:StopExecution`, which is a
fundamentally different contract from every stop-and-resume schedule on this directory.

## Halted means finished, permanently

A stopped execution transitions to a terminal state and will never continue. The morning start
does not resume anything; if the work still matters, something must start a new execution, and
that new execution begins at the first state. Workflows living under this schedule need to be
built for it: idempotent steps, progress checkpointed outside the execution (DynamoDB, S3, a
database), and an input shape that lets a rerun skip completed work.

## Standard only

`states:StopExecution` applies to Standard workflows. Express executions cannot be stopped from
outside; they run to completion or time out within their 5-minute ceiling. An Express-heavy
system simply has nothing for this schedule to grab, which is fine: five minutes of overrun
past the window boundary rarely matters.

## The saving lives downstream

Standard workflows bill per state transition, and a halted execution stops accruing them, but
transition costs are rarely the point. The real money is in what the states do: the Glue jobs,
the SageMaker training, the fleets of Lambda invocations, the polled APIs. Halting the
orchestrator stops it driving those resources through the night. One caution: a job already
started by a state may continue on its own after the execution stops, so verify whether the
integration cancels children or abandons them, per service.

## Use the window as a change freeze

Because the halt is enforced rather than advisory, the off-hours window doubles as a
deployment-safe period: no execution can be mid-flight at 3am while a schema migration runs.
Teams that fight long-running workflows during maintenance windows get that guarantee for free.