hey everyone! We’re on Prefect Cloud with ECS workers (Prefect 3.6.5, prefect-aws 0.7.1). ECS worker runs on prefect-ecs-cpu-cluster (Fargate). infrastructure_pid is set on runs (e.g., prefect-ecs-cpu-cluster:arnawsecsus-east-2:.../task/...). IAM for the worker task role includes ecs:StopTask. Region/queue are set (AWS_REGION/AWS_DEFAULT_REGION=us-east-2, PREFECT_INTEGRATIONS_AWS_ECS_OBSERVER_SQS_QUEUE_NAME=prefect-ecs-events). SQS/EventBridge queues are empty (no backlog). Problem: Cancelling a flow run leaves it in “Cancelling”; the ECS task keeps running. Worker logs show no StopTask or cancel handling. Even with infrastructure_pid present, no stop is issued. Forcing state via UI/CLI doesn’t stop the task. Any known issues or fixes to get Cancel to issue ECS StopTask? Should we tweak anything on the worker side, or is there a known bug in the cancel path?

Hey @João Pedro Boufleur! Are you able to share any logs from a hung ECS task around the time that you cancelled the task?

thanks for the quick response Alex! Going to DM you the event logs for the flow that I cancelled mid run.

Have you customized the start up command in the container? By default it is

prefect flow-run execute

@alex sorry for late response, but thanks a billion, that was it! it was using

prefect.engine

instead have a blessed end of week!