Hey all, I am trying to migrate from Agents with Infrastructure and Storage blocks (kubernetes jobs, and S3 file storage) to Workers/work pools. We build our own docker images because we have a lot of internal packages that we use, and have always baked our source code into the images. We now want to actually use the baked-in source code rather than pulling from external storage. Our source code ends up in

/workdir/src/<package>

in the image (where

/workdir

is set as the working directory in the image build process), and when registering a deployment to the server, it says the entrypoint is

src/path/to/file:flow_function

as I would expect. However, when trying to actually run the deployment, it immediately fails with

Flow could not be retrieved from deployment.
Traceback (most recent call last):
  File "", line 991, in exec_module
  File "", line 1128, in get_code
  File "", line 1186, in get_data
FileNotFoundError: [Errno 2] No such file or directory: '/opt/prefect/src/path/to/file.py'

which causes

prefect.exceptions.ScriptError: Script at 'src/path/to/file.py' encountered an exception: FileNotFoundError(2, 'No such file or directory')

So I am confused as to why Prefect is looking for it in

/opt/prefect

. I am trying creating the deployment with

my_flow.to_deployment(name="test-deployment", work_pool_name="my-work-pool")

and don't see anything in the documentation about changing the entrypoint or path when using

.to_deployment()

.

hi @Nick Hoffmann - how are you actually creating the deployment? are you using

.deploy

? if so, you shouldn't need

to_deployment

I have a few dozen deployments that I am trying to programmatically generate and deploy, all with different parameters, so I am doing basically:

import prefect
from dataclasses import dataclass
from typing import Optional, Any, List

@dataclass
class MyDeployment:
    flow: prefect.Flow
    name: str
    kwargs: Dict[str, Any]

def generate_deployment_list() -> List[MyDeployment]:
    # My custom logic for building parameter sets for each deployment
    ...

dep_list: List[MyDeployment] = generate_deployment_list()
dep_list = [dep.flow.to_deployment(name=dep.name, **dep.kwargs) for dep in dep_list]
prefect.deploy(*dep_list,
               work_pool_name="my-work-pool",
               image="my-image-built-in-CI",
               build=False)

just checking quick, are you averse to

prefect.yaml

? this seems like a natural use case for it

not necessarily averse to it, but it doesn't really fit our use case as nicely as we'd like. We have a set of deployments that we make and run for each of our customers, and we want to be able to deploy for a new customer by simply adding them to our database and not have to update and commit that

prefect.yaml

file

hmm this is working for me if thats helpful

I found out that actually my flow runs were using a standard Prefect image, not my custom image, so that solved that problem

ahh i see, glad you figured it out

ye, thank you for your time