Yes, I realized I didn't put "http://" before the ip (host.docker.internal).

The error is a NewConnectionError caused by the fact that the serving server has to download the artifacts needed to deploy the model (aka the model itself and the preprocess script), but these have an address starting with localhost (since they are on the clearml server on my host machine), hence are not accessible by the serving, because the serving needs the url to start with "host.docker.internal" and not localhost

It appears that the containers of clearml-serving couldn't access my host machine, because locahost referred to the container and not the host. However , changing the server address in the docker-compose.yml from localhost to host.docker.internal gives me another error:
clearml.backend_api.session.session.LoginError: Unrecognized Authentication Error: <class 'requests.exceptions.InvalidSchema'> No connection adapters were found for 'host.docker.internal:8008/auth.login'

Any idea what i am m...

Hi @<1523701087100473344:profile|SuccessfulKoala55> , sure, here it is
I've just hidden the url of the server, which is the self deployed version

Unfortunately, now the problem is that, although the serving server can see the clearml server and I can create an endpoint to expose a model, the serving server can't access the needed artifacts (like the preprocessing script and the model chosen) because their url starts with "localhost", while the serving server needs the url to start with "host.docker.internal" to actually access them...

Hi @<1524560082761682944:profile|MammothParrot39> , I have the same problem. Could you elaborate on the solution you found?

Hello Jack and thank you for your answer.
The server is reachable, curl gives me HTTP/1.1 200 OK.
I'm running both the clearml server and (attempting to run) the serving server on the same machine with Windows11. Both of them use docker containers