Bring your towel and join one of the fastest growing data communities. Welcome to our second-generation open source orchestration platform, a completely rethought approach to dataflow automation.

Prefect Community

Hi, I'm trying to work through flow registration with cloud using Docker storage

I keep getting an exception from docker.py line 544 `push_image` that says "no basic auth credentials"

So I created some code using the python docker library to log into the registry. That succeeds, and the next thing I do is call `flow.register()` . Same error

So this may be more of a docker question, but any thoughts on how to get past this issue?

Hey <@ULAGV364C> - I recommend working on this with Docker directly instead of through Prefect’s `.register()` call; I suggest starting by trying to pull your `base_image` or push to your configured `registry_url` and seeing if that raises the same error

Hey <@ULAGV364C>, I think I had a similar error message at first when trying to register DockerStorage with a private ECR. Double-check the Docker-related configurations that you pass to the Storage object, especially stuff like `base_url` -- the defaults may not be optimal.
For me in the end, it turned out I had to pass `tls_config` since the running Docker was configured with that level of security. Otherwise the authentication failed

Thanks <@UR914ST7H>. By walking through the steps using the docker CLI, I was able to get past the issue

One thing I have to do is call `boto3.client('ecr').create_repository()` to ensure it exists before trying to push to it. That was the only additional thing I ran into

I believe that python docker.login is not saving credentials. If I log in through docker CLI, the creds are saved to ~/.docker/config.json

Not so if I do the login through python `docker.login()`

Was an issue in the past; could still be today <https://github.com/docker/docker-py/issues/1061>

&gt; One thing I have to do is call `boto3.client('ecr').create_repository()`
Ah, interesting. I'm basically reusing the same repository but assign unique tags to each flow's docker image -- so no need to create new repos all the time

That's a good idea. In our use case, we need many repos

So what is the right way to login via docker before calling `flow.register()` ? If I don't login at all, I get the "no basic auth credentials" error.

If I manually log in via the command line, then the creds are saved

But manually logging in doesn't seem like the best way to go

I ended up with this `subprocess.run(['docker', 'login', '-u', username, '-p', password, url])` :face_with_raised_eyebrow: