https://prefect.io logo
Join Slack
Powered by
Hello folks, I'm running flows on Prefect 1.2 usin...
# prefect-community
s
Hello folks, I'm running flows on Prefect 1.2 using ECSRun and my setup uses the approach as described in this issue under the Conclusions pt. 5. It uses the 
task_definition_arn
with the containers named as 
flow
While I'm able to trigger the flows, the flow crashes with the error
Copy code
copilot/flow/8d31faa7f1ba   File "/usr/local/lib/python3.8/importlib/__init__.py", line 127, in import_module
copilot/flow/8d31faa7f1ba     return _bootstrap._gcd_import(name[level:], package, level)
copilot/flow/8d31faa7f1ba   File "<frozen importlib._bootstrap>", line 1014, in _gcd_import
copilot/flow/8d31faa7f1ba   File "<frozen importlib._bootstrap>", line 991, in _find_and_load
copilot/flow/8d31faa7f1ba   File "<frozen importlib._bootstrap>", line 961, in _find_and_load_unlocked
copilot/flow/8d31faa7f1ba   File "<frozen importlib._bootstrap>", line 219, in _call_with_frames_removed
copilot/flow/8d31faa7f1ba   File "<frozen importlib._bootstrap>", line 1014, in _gcd_import
copilot/flow/8d31faa7f1ba   File "<frozen importlib._bootstrap>", line 991, in _find_and_load
copilot/flow/8d31faa7f1ba   File "<frozen importlib._bootstrap>", line 973, in _find_and_load_unlocked
copilot/flow/8d31faa7f1ba ModuleNotFoundError: No module named '/root/'
However, i've ensured the flows folder from my project are under the PYTHONPATH and I can't entirely figure out what (if anything) that cloudpickle is trying to do to access those flows at the time of execution.
a
Part 5 of this conclusion section doesn't include 
task_definition_arn
, though - it's mainly focused on 
task_role_arn
and 
execution_role_arn
. - can you share your ECSRun?
Re your error "ModuleNotFoundError: No module named '/root/' " - this is a Dockerfile issue. Can you share your Dockerfile? Is it based on Prefect base image? You may also check this example
I can't entirely figure out what (if anything) that cloudpickle is trying to do to access those flows at the time of execution.
You don't necessarily need to use cloudpickle - check this doc and this example using script storage rather than pickle
s
Yes. My Dockerfile is this
Copy code
FROM prefecthq/prefect:latest-python3.8

ENV ANALYTICS_APP_DIR /app

WORKDIR ${ANALYTICS_APP_DIR}

RUN apt-get update -y \  
    && apt-get update && apt-get install -y curl build-essential libssl-dev libffi-dev lib32ncurses5-dev git libsnappy-dev postgresql-client\
    && apt-get -y install telnet vim unzip

ADD poetry.lock $ANALYTICS_APP_DIR/
ADD pyproject.toml $ANALYTICS_APP_DIR

ENV PYTHONPATH=$ANALYTICS_APP_DIR
ENV POETRY_VIRTUALENVS_PATH=$ANALYTICS_APP_DIR/

RUN pip install poetry

RUN poetry install

ADD . $ANALYTICS_APP_DIR/


CMD ["/bin/bash", "-c", "./ops/launch-ecs-agent.sh"]
And the ECSRun looks like this
Copy code
ecs_run_config = ECSRun(
        labels=[f"{cfg('ENV')}"],
        task_definition_arn=task_definition_arn,
        task_role_arn=task_role_arn,
        execution_role_arn=execution_role_arn,
        cpu=1024,
        memory=2048,
        run_task_kwargs=dict(            
            cluster=f"{cfg('ECS_CLUSTER_NAME')}",
            launchType=f"{cfg('LAUNCH_TYPE')}",
            overrides=dict(
                containerOverrides=[
                    dict(name="flow",
                        command=["poetry", "run", "prefect", "run", "hello-flow"],
                        cpu=1024,
                        memory=2048,                        
                    )
                ]
            ),
        ),
    )
My ECS agent container image also contains the flow code. So I use the same image and task_definition_arn in my ECSRun code and I override the command using the overrides argument
Thank you, the modules approach to storage got me to a working flow. Thank you very much Anna. Also thanks for the repos. Very very helpful.
🙌 2
16 Views
Open in Slack
PreviousNext
Powered by