Skip to content

feat(cli): unify job run orchestration for dataset tasks - #1235

Merged
zhangjaycee merged 6 commits into
alibaba:masterfrom
berstpander:feat/job-run-all
Jul 15, 2026
Merged

feat(cli): unify job run orchestration for dataset tasks#1235
zhangjaycee merged 6 commits into
alibaba:masterfrom
berstpander:feat/job-run-all

Conversation

@berstpander

@berstpander berstpander commented Jul 13, 2026

Copy link
Copy Markdown
Contributor

Summary

  • Extend rock job run into a unified entrypoint for single-task, multi-task, and full dataset execution via --task, --tasks, and --all
  • Add CLI-level orchestration in rock.cli.job_run for task planning, run/job metadata writes, concurrency control, JSONL progress events, and score summaries
  • Keep the SDK focused on atomic primitives: task-bound planning, single-job execution, run/job metadata repositories, and reconnecting to existing sandbox jobs
  • Add --resume <run_id> support that skips completed tasks, reconnects to still-running sandbox jobs when metadata contains sandbox_id/session/pid, and marks unrecoverable attempts before rerun
  • Add run-level query support with rock job runs, rock job status --run-id, and run-aware rock job show
  • Support both Harbor and Bash task binding while keeping one task per sandbox at the SDK execution layer

Validation

  • uv run pytest tests/unit/sdk/job/test_planner.py tests/unit/sdk/job/test_run_meta_repository.py tests/unit/sdk/job/test_executor_concurrent.py tests/unit/cli/command/test_job.py tests/unit/cli/test_job_run.py — 26 passed
  • uv run pytest tests/unit/sdk/job/test_meta.py tests/unit/sdk/job/test_viewer.py tests/unit/sdk/job/test_executor.py tests/unit/sdk/job/test_job.py tests/unit/sdk/job/test_config.py — 208 passed
  • Built wheel from feat/job-run-all: dist/rl_rock-1.10.1.dev2-py3-none-any.whl
  • Verified wheel metadata consistency: filename, .dist-info, and METADATA Version are all 1.10.1.dev2

Closes #1231

berstpander and others added 6 commits July 2, 2026 16:54
…ied job metadata

- Add JobViewer client to read job results, trials, trajectories,
  artifacts and logs from OSS (supports AK/SK and admin STS auth)
- Add rock_meta.json: unified metadata written inside the sandbox
  by both BashTrial and HarborTrial wrapper scripts, enabling
  job status tracking (running/completed/failed) for both job types
- Extend CLI with `rock job list/show/trials/trial` subcommands
- Add JobMeta model and render_meta_json helper
- 70 new unit tests (test_viewer.py + test_meta.py)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Harbor's own _oss_mirror_job_root_files() handles rock_meta.json
upload. Only BashTrial needs wrapper-level meta writing.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…ection and invalid JSON

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
@zhangjaycee
zhangjaycee merged commit 7aee406 into alibaba:master Jul 15, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Remote HTTP requests and scheduler worker execution can block indefinitely

3 participants