Hello all,
I am trying to have a grasp on ClearML agents, and I'm struggling with docker mode. Let's say, I have a Dockerfile for my custom made docker image. However, I want to build that image on the fly, just because of having the freedom to have full control over the image. However, I couldn't figure out that how can I send a dockerfile to agent while initializing an experiment so agent can build that image and run experiment on it. I'm aware I can workaround this one by sending bash script to the container so everything I want to be installed will be installed before my experiment actually starts, but is there a way to send a Dockerfile to the agent so it can build an image from it and run the experiment in a container that built on top of that image?
ZanyPig66 , Hi.
To use a custome docker image with ClearML you need to have that image locally on the machine that is runing the agent. In the experiment you can specify the docker image either via UI (in the Execution -> Container sections)
Or in code like this:task.set_base_docker(docker_image="python:3.9-bullseye")
Also a small clarification:
ClearML doesn't build the docker image itself. You need to have a docker image already built to be used by ClearML
