mars-project / mars-project/mars

[BUG] Distributed training failed on Ray cluster

Open
#2,828 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

mod: learn type: bug
Dominant language
Python
Stars
2.7k
Forks
325
PR merge metrics
No merged PRs in 30d

Description

Describe the bug
Mars integrates some deep learning frameworks(PyTorch, TensorFlow), these frameworks usually need to set some environments for distributed training, TF_CONFIG for TensorFlow, MASTER_ADDR for PyTorch. We use ctx.get_worker_addresses() to collect all worker addresses, it works well for Oscar backend, while for Ray, the addresses start with ray:// which is invalid for them, we need a method to get worker's host IP not address to address the issue.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by tracing the use of ctx.get_worker_addresses() in the distributed-training path and compare its behavior on the Oscar and Ray backends. Check how TF_CONFIG and MASTER_ADDR consume the collected values. Done means Ray workers provide usable host IPs rather than ray:// addresses, while existing Oscar behavior remains intact.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, pytorch, tensorflow
Domain
distributed-systems, machine-learning
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.