Step Functions StopExecution: ABORTED Status Without Guaranteed Resource Termination

Answer the question first

Stopping a Standard workflow with StopExecution always changes the execution status to ABORTED and emits a Stopped event. It also blocks any new state transitions. However, the API does not automatically terminate the underlying AWS resource that was started by the current state.

Confirmed facts

• The StopExecution call only affects the Step Functions state machine. It does not send a cancel or stop signal to the invoked service.
• State‑machine timers (TimeoutSeconds and HeartbeatSeconds) only limit how long the state waits for a response; they do not influence the lifetime of the underlying resource.
• When the execution is aborted, Step Functions emits a Stopped event with cause ‘User initiated stop’ and the execution history shows no further state transitions.

Which .sync integrations propagate cancellation?

Some service integrations implement a “stop” operation that Step Functions forwards when the execution is aborted. The following .sync integrations currently support this behaviour (see the Service integration patterns table):

• ECS – StopTask
• Batch – TerminateJob
• SageMaker – StopTrainingJob
• Glue – StopJobRun
• Step Functions (nested executions) – StopExecution

Integrations that do not support cancellation (e.g., Lambda, HTTP, DynamoDB, SNS, SQS) will continue to run to completion or failure. The only limits are the service‑side timeouts or the Step Functions timers.

Pattern to bound the backend job lifetime

1. Use a .sync integration that supports cancellation if you need the job to terminate instantly when the parent execution stops.
2. For integrations that lack built‑in cancellation, wrap the job in a Lambda or Step Functions task that periodically checks the execution status (via DescribeExecution) and calls a service‑specific stop API if the execution is aborted.
3. Alternatively, design the job to be idempotent and tolerant of orphaned instances; let it finish but ensure it cleans up resources automatically.

Steps to verify what happened after StopExecution

1. In the Step Functions console, open the execution and confirm the Stopped event and the ABORTED status.
2. Check the logs or metrics of the integrated service (e.g., CloudWatch Logs for ECS tasks, Batch job status page, SageMaker console) to see if a stop/terminate request was received.
3. If the service shows the job still running, the integration does not support cancellation for that resource type.

Missing diagnostic detail

To give a more precise recommendation, we need to know which .sync integration you are using. If you can share the integration type, we can point you to the exact stop API it supports (if any) and any best‑practice patterns for that service.