Yes — the easiest option is to call the flow’s underlying function with

.fn

during local dev. Example:

python
from prefect import flow
from prefect_aws.decorators import ecs

@ecs(work_pool="my-ecs-pool")
@flow
def my_flow(x: int):
    return x * 2

# Local dev: runs locally, no ECS submission
result = my_flow.fn(5)

# Normal call: submits to ECS via the work pool
result = my_flow(5)

A few important notes: *

@ecs

changes the flow so that calling

my_flow(...)

submits it to ECS * There is not a built-in “run locally instead” switch for the decorator * So for local execution,

.fn(...)

is the simplest override If you want the same call site to behave differently in dev vs prod, a common pattern is to apply

@ecs

conditionally:

python
import os
from prefect import flow
from prefect_aws.decorators import ecs

def maybe_ecs(work_pool: str, **kwargs):
    def decorator(fn):
        wrapped = flow(fn)
        if os.getenv("RUN_ON_ECS") == "true":
            return ecs(work_pool=work_pool, **kwargs)(wrapped)
        return wrapped
    return decorator

@maybe_ecs(work_pool="my-ecs-pool")
def my_flow(x: int):
    return x * 2

Then: * local dev:

python flow.py

→ runs locally * prod:

RUN_ON_ECS=true python flow.py

→ submits to ECS Relevant docs/code: * Prefect AWS integration docs: https://prefecthq.github.io/prefect-aws/ * Prefect deployments docs: https://docs.prefect.io/v3/how-to-guides/deployments * Prefect source for infrastructure-bound flows: https://github.com/PrefectHQ/prefect * Prefect AWS source: https://github.com/PrefectHQ/prefect/tree/main/src/integrations/prefect-aws If you want, I can also show the best-practice dev/prod structure for ECS flows in Prefect 3.x.

Most likely: the code is not failing locally — the ECS container environment is missing a dependency that exists in your local env.

importlib_metadata

is a Python package backport. In modern Python, the stdlib module is

importlib.metadata

, but some dependencies in the container may still import

importlib_metadata

. If that package is missing in the ECS image, you’ll get:

text
ModuleNotFoundError: No module named 'importlib_metadata'

Why this happens: * your local environment has the full dependency set installed * the ECS image/task is using a different environment * a transitive dependency in the ECS container expects

importlib_metadata

* this is commonly caused by a custom Docker image or incomplete dependency install What to check first 1. Are you using an official Prefect AWS image or a custom one? * Official images are the safest:

prefecthq/prefect-aws:...

* Custom images often miss transitive deps 2. Are local and ECS using the same Python version? * Python 3.12 tends to avoid some of this pain * Python 3.10/3.11 environments hit this more often 3. Did your image actually install all runtime dependencies? Make sure the image build installs

prefect

and

prefect-aws

fully. Example Dockerfile:

dockerfile
FROM prefecthq/prefect:3-latest-python3.12

RUN pip install --no-cache-dir prefect-aws

Or if you’re starting from plain Python:

dockerfile
FROM python:3.12-slim

RUN pip install --no-cache-dir "prefect[aws]" prefect-aws

If you want the fastest tactical fix, explicitly install the missing package too:

dockerfile
RUN pip install --no-cache-dir "prefect[aws]" prefect-aws importlib-metadata

That said, I’d treat that as a signal that your ECS image is not matching your local env. Recommended fix * use a pinned official image if possible, like

prefecthq/prefect-aws:<tag>

* or rebuild your custom image so it installs all runtime deps * align Python versions between local and ECS * avoid relying on

latest

in production Why it deploys to ECS when you run locally Because with

@ecs

, calling the flow normally submits it to the ECS work pool. For local-only execution, use:

python
my_flow.fn(...)

instead of:

python
my_flow(...)

So if you were expecting a local run but called the decorated flow normally, that explains why you’re seeing an ECS-container import error. Helpful links: * Prefect AWS integration: https://prefecthq.github.io/prefect-aws/ * Deployments docs: https://docs.prefect.io/v3/how-to-guides/deployments * Prefect repo: https://github.com/PrefectHQ/prefect If you paste your Docker image / deployment config / ECS task image, I can help pinpoint exactly where

importlib_metadata

is being missed.