NVIDIA-NeMo / NVIDIA-NeMo/Gym

[checkpoint] Restore GRL Sokoban episodes

Open
#3,237 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

area:environment feature
Dominant language
Python
Stars
1.2k
Forks
349
Avg merge
1d 21h
Merged PRs (30d)
318

Description

Parent

Child of #3219 and #3024.

Problem and state owner

resources_servers/grl_sokoban/app.py owns a live Sokoban environment, board/player/box state, counters, observation, reward, done flag, and RNG state outside generic session storage.

Proposed adapter

Provide an explicit JSON codec and recreate a fresh environment from configuration plus serialized logical/RNG state.

Implement explicit checkpoint opt-in, route classification, export/restore/retire hooks, execution-to-session binding, revision reporting, atomic replacement under attempt N+1, and stale-attempt fencing.

Acceptance criteria

  • The same post-restore action yields identical board, observation, reward, counters, and terminal status.
  • Checkpoint after multiple mutations, replace the server process, restore, and compare the next operation with an uninterrupted execution.
  • Corrupt or incomplete bulk state activates no sessions.
  • Attempt N cannot mutate or retire restored attempt N+1.
  • Terminal cleanup is idempotent and releases restored state exactly once.

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 with resources_servers/grl_sokoban/app.py and read parent issues #3219 and #3024 for the surrounding session and environment design. Trace the live board, counters, observation, reward, done flag, and RNG state, then map the proposed checkpoint hooks to the acceptance criteria. Done means deterministic restore, safe rejection of corrupt bulk state, stale-attempt fencing, and idempotent cleanup.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
backend-api-design, distributed-systems
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Active
Clarity
Mostly clear
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.