Skip to content

Frequently Asked Questions

Installation

Do I need docker compose to use Labtasker?

No. Depends on your needs:

  1. If you want the simplest setup, you can use labtasker-server serve to run the server out-of-the-box. All you only need is a Python environment.
  2. If you need more scalable and robust setup, use the docker compose method to deploy the server.
  3. If you have the mongod installed locally, you can specify in the configuration file and choose --db-mode external when you run labtasker-server serve.

Do I need to deploy the Labtasker server?

It depends:

  1. If you’re using Labtasker on your own, you need to deploy the Labtasker server for the client tools to work. It is very convenient: simply run labtasker-server serve in the background and you’re good to go.
  2. If you’re working with others or trust someone who has already deployed the Labtasker server, you don’t need to—just connect to their server. (Make sure you trust the provider, as your task information is not encrypted.)

Do I need to deploy Labtasker server on the cloud? With a domain name?

No and no. You can run Labtasker server at localhost (could be your GPU server or your laptop). The only requirement is that the workers can access the server.

Usage

How to run job command with environment variables using labtasker loop?

❌ The wrong way:

labtasker loop -- CUDA_VISIBLE_DEVICES=0 python '%(script)' --dataset '%(dataset)'

✅ The right way:

CUDA_VISIBLE_DEVICES=0 labtasker loop -- python '%(script)' --dataset '%(dataset)'

or else, you may get an error like this:

FileNotFoundError: [Errno 2] No such file or directory: 'CUDA_VISIBLE_DEVICES=0'

What does the "traceback hook" option do in labtasker init?

The "traceback hook" replaces Python’s default exception handler (sys.excepthook) to format error messages and hide sensitive information like passwords.

If you disable this option, Labtasker won’t modify sys.excepthook, which can help avoid conflicts with other libraries that also use it.