For the new ECSTask - how does the image know which command to execute. For me it reaches the state "Pending" but in the container it just executes the default image entrypoint. The task definition command is empty (

[]

) in my yaml file the command points to my prefect entry script but it needs the UUID of the flow run. So it boils down to two questions: 1. How does the queue worker provide the UUID into the ecs task 2. What does

Command

do? (in the second screenshot I changed the task definition command from [] to

python3 /app/tasks/run.py

- but it still needs the flow run uuid)

Hi Tim, the task run ID should be available in the PREFECT__FLOW_RUN_ID environment variable inside the container.

ECSTask

sets that before it starts the container. The normal process looks like: • The container starts and runs the command provided by the deployment (the default is

python -m prefect.engine

) •

prefect.engine

runs, reads the flow ID from the env variable, pulls details about the deployment, downloads the code from whichever external storage you specifed in the deployment, then runs it

Thanks for your confirmation! The only reason I use a custom script instead of

-m prefect.engine

is because I need to override

ignore_storage:

flow = await load_flow_from_flow_run(
    flow_run, client=client, ignore_storage=True
)

But the other parts are just copy pasted for now. At the beginning of the script i just dumped all env variables:

if __name__ == "__main__":
    # Standard Library
    import os
    import sys

    print(os.environ.items())

Besides my own env variables which i set through a s3 arn resource in the task definition and a few default docker env variables, no prefect variables are set.

This seems to be a bug. I think it is because when you added a command then deleted it, the command becomes

[]

, which overwrites the default command to start an agent. We raised the issue here