ClearML（原Trains）能否本地运行？无Docker/K8s可行吗？

Absolutely! I’ve actually helped a few folks get ClearML (formerly Trains) running locally on Ubuntu 18.04 without Docker or Kubernetes—here’s how you can pull it off:

Prerequisites First #

First, make sure you have the core dependencies installed on your system:

• Install essential system packages:

  sudo apt update && sudo apt install -y python3-pip python3-dev git libssl-dev libffi-dev build-essential
  

• Upgrade pip to avoid version conflicts:

  python3 -m pip install --upgrade pip
  

Step 1: Install & Configure ClearML Server Components #

ClearML Server relies on three core services: API, Web UI, and File Storage. You can run all of them directly via Python without containers.

Install the ClearML Server Package #

pip3 install clearml-server

Generate Local-Focused Configuration #

Run the init command to set up local storage (no remote services needed):

clearml-server init

When prompted, select local storage options for every component (database, file storage, etc.). The config file will save to ~/.clearml/server.conf—you can tweak it later, but the default local setup works for basic experiments.

Step 2: Start the Server Services #

You’ll need to run three separate processes. Use tmux or screen to keep them running in the background, or open three terminal windows:

1. Start the API Service (default port 8008):

   clearml-server api
   

2. Start the Web UI (default port 8080—access it via http://localhost:8080 once running):

   clearml-server web
   

3. Start the File Storage Service (default port 8081):

   clearml-server files
   

Pro Tip: If you don’t want to keep terminals open, tmux is great for splitting a single window into panes for each service—easy to monitor logs if something goes wrong.

Step 3: Point Your ClearML Client to the Local Server #

Now configure your client to use your local instance instead of the cloud platform:

1. Initialize client settings:

   clearml-init
   

2. When prompted, enter your local server details:
   • API server: http://localhost:8008
   • Web server: http://localhost:8080
   • File server: http://localhost:8081
   • You can skip credential prompts for local use (just press enter) unless you want to set up user authentication later.

Step 4: Test Your Setup #

Run a simple test script to confirm everything works:

from clearml import Task

# Initialize a local task
task = Task.init(project_name="Local Test Project", task_name="First Local Run")

# Dummy work to generate logs
for i in range(10):
    print(f"Completed step {i}")
    task.get_logger().report_scalar("Progress", "Current Step", i, iteration=i)

# Task auto-completes when the script finishes

Save this as test_clearml.py and run it:

python3 test_clearml.py

Then open http://localhost:8080 in your browser—you should see the task listed under the "Local Test Project" with the scalar logs visible.

Troubleshooting Common Issues #

• Port conflicts: If default ports (8008, 8080, 8081) are in use, edit the ~/.clearml/server.conf file to change them, then restart the services.
• Permission errors: Ensure you have write access to ~/.clearml/data (the default local storage directory). Adjust permissions with chmod instead of using sudo if possible.
• Python version issues: Ubuntu 18.04’s default Python 3.6 is supported, but if you hit errors, install Python 3.7/3.8 via ppa:deadsnakes/ppa and use that version for installation.

I’ve seen this setup work smoothly for small-scale local experiments—no need for container orchestration when you just want to test ClearML on your own machine.

内容的提问来源于stack exchange，提问作者DalyaG