Hi! I've been exploring Prefect quite a bit these days and I'm really impressed. I've been looking for a way to keep the containers where the Flows run (with 

DockerRun

) in sync with the code/package I have in the repository. One of the approaches I came up with was to develop a Storage (very similar to Git) that would do the following: 1. Clone the repository into a temp folder. 2. Extract the Flow (as it is currently done).  3. Search the repo for a specified 

Dockerfile

. (the 

Dockerfile

 would do something like an 

ADD

 of the entire repo and a 

pip install -e .

 so that the image has the latest code in it). 4. Build the image with a certain tag (the same tag that 

DockerRun

 would use). (This is all assuming that the Storage code is executed in the Docker Agent process just before the container is started which would give time to create the image before the 

run_config

 create the container and run the flow inside) I was going to do a test by creating the custom Storage but I ran into things I didn't take into account like 

serialization

. But before I explore this option further I wanted to ask: • Is there any way to achieve what i want, that I am not seeing, that doesn't involve a Container Registry or a CI pipeline? • Does the solution I propose make sense to you? Do you think it would work or am I missing something? • Would it be useful for more people and would it be worth some contribution? I would appreciate any guidance, advice or comments on the subject.

Hey @Jose Daniel Posada Montoya, I don’t have a significant amount of experience when it comes to CI/CD but from what I have seen, people trigger the Docker build and Flow registration from the Git commit. For example, you have a prod branch, and then when there is a merge into there, you trigger the CI/CD the Flow registration that uses Docker Storage. If you use this setup, you can upload the Flow container to a registry. Upon Flow execution, the container will be download and ran. I should also note that you can decouple the

Docker

container and the actual script sometimes. If you container has all of the dependencies you need, just use

DockerRun

+ your Git storage. The flow will be downloaded from Git storage and run on top of the container. This won’t need any CI pipeline at all. You can’t create a custom Storage because the storage information is pushed to our database. Upon flow run, it uses that Storage class to search for your Flow and load it. So by default, it has to be something we know how to work with.