search

ray-project / ray

Ray is an AI compute engine. Ray consists of a core distributed runtime and a set of AI Libraries for accelerating ML workloads.
https://ray.io
Apache License 2.0
34.02k stars 5.78k forks source link

[Core] Unable to run worker with virtual environment without installing dashboard #45410

Open SebastianMorawiec opened 6 months ago

SebastianMorawiec commented 6 months ago

What happened + What you expected to happen

Ive tried to minimize the dependencies inherited from Ray. I am trying to run Ray Remote with new virtualenv (adding new dependencies).

If I am only installing pip install ray==2.9.0 I get an error that there is no such package as virtualenv from worker. I need to install pip install ray[default]==2.9.0 to make it work, which brings 23 more dependencies bunch of them related to ray-dashboard which I am not interested it.

I can add virtualenv to my own dependencies, but I'd have to maintain dependency that actually Ray uses.

TLDR: Please add your virtualenv to default dependencies.

Versions / Dependencies

Ray 2.9.0 (same goes for 2.22 tho)

Reproduction script

n/a

Issue Severity

High: It blocks me from completing my task.

rynewang commented 6 months ago

Do you mean runtime environment from Ray? This feature requires ray[default], because we need those dependencies to set up things. If you don't want runtime env, you can create an image and install deps in it, thus avoiding runtime env.

SebastianMorawiec commented 5 months ago

I want to be able to run Ray Remotes with runtime environment, without installing Ray Dashboard. Ray[default] installs (AFAIR) 34 more dependencies than just Ray with no extra. To run Ray Remote with runtime environments, I need to install Ray and separately install virtualenv dependnecy. Would be much better IMHO if virutalenv just became standard dependnency without any extras