I have a base task for each pipeline step. When I initialize a pipeline, for each step I clone the corresponding task, modify it and add it as a step. Tasks are launched from a pipeline, not cli. I'm absolutely sure docker argument is not empty (I specify it with export_data['container']['image'] = ' http://registry.gitlab.com/cherrylabs/ml/clearml-demo:clearml ' , and it shows on Web UI)

It worked, thank you!

I understand but how do you launch the cleaml-agent itself:
clearml-agent daemon --detached --queue default --docker

MelancholyElk85 what do you have under "Installed Packages" for this specific Task ?

You're right, I re-launched the agents, now it pulls images

MelancholyElk85 can you provide some logs on the pip install fails ?

MelancholyElk85 , did you start the task using task.init() or did you start it through the pipeline?

MelancholyElk85 , please try the following:
task_overrides={'script.branch': 'main', 'script.version_num': '', }

I specifically set is as empty with export_data['script']['requirements'] = {} in order not to reduce overhead during launch. I have everything installed inside the container

This doesn't seem to be running inside a container...
What's the clearml-agent launch command you are using ? (i.e. do you have --docker flag)

When I launch tasks with a pipeline, they keep complaining about missing pip packages. I run it inside a docker container, and I'm sure these packages are present inside it (when I launch the container locally, run python3 and import them, it works like charm). Any ideas how to fix this?

This is odd, Can you send the full Task log? (remove any pass/user/repo that you think is sensitive)

👍

I specifically set is as empty with 

export_data['script']['requirements'] = {}

 in order not to reduce overhead during launch. I have everything installed inside the container

Do you have everything inside the container Inside a venv ?

It doesn't install anything with pip during launch, I'm assuming it should take everything from the container itself (otherwise there would be a huge overhead). It simply fails trying to import things in the script

File "preprocess.py", line 4, in <module> from easydict import EasyDict as edict ModuleNotFoundError: No module named 'easydict'