Skip to content

feat: 迁移 ray examples guard - #208

Merged
hipudding merged 3 commits into
Ascend:mainfrom
Meihan-chen:feat/ray-examples-guard
Oct 13, 2026
Merged

hipudding merged 3 commits into
Ascend:mainfrom
Meihan-chen:feat/ray-examples-guard

Conversation

@Meihan-chen

Copy link
Copy Markdown
Contributor

Summary

  • 从 CI workflows 仓(cosdt-ci-test/workflows,projects/ray/,origin/main)迁移 ray 的 examples 守护管线到 examples-guards/ray/,与 accelerate 的迁移方式一致(参考 feat: 迁移 accelerate examples guard #200 / feat: 启用 accelerate-examples 每日定时触发 #203)
  • 项目载荷:examples_manifest.yaml + scripts/(setup/run_example.sh、resolve_test_requirements.py、verify_ray_test_support.py)+ 14 个 project-owned NPU 一致性测试(example/)
  • 上游测试(python/ray/tests/accelerators/test_npu.py、python/ray/train/tests/test_torch_device_manager.py)与 project-owned example 共用同一 manifest,通过 source: upstream/project 区分;上游跑 target checkout,project 跑 examples-guards/ray/example/
  • 新增薄触发器 ray-examples.yml(参考 accelerate-examples.yml 格式):schedule 先注释(与 workflows 仓当前状态一致,矩阵在 NPU runner 上全绿后再启用),default_branch: master,max_parallel: 2

说明

  • 共享引擎(examples-template.yml、check_supported_entries.py、write_example_result.py、result.schema.json)已支持 source: project / npu_devices / target_ref,本 PR 无需改动引擎
  • 载荷文件与 workflows 仓 origin/main 逐字节一致(blob SHA 校验通过);未迁移 tests/、docs/(Quick-start)与项目级 README(与 accelerate 迁移口径一致)

Test plan

  • 手动 dispatch ray-examples workflow,确认 monitor → manifest-check → run-example 矩阵(16 条)→ validate-results 全链路
  • 矩阵在 NPU runner 上全绿后再启用每日定时(北京时间 02:00)

从 CI workflows 仓迁移 ray 的 examples 守护管线:

- 新增 examples-guards/ray/ 项目载荷(examples_manifest.yaml +
  setup/run_example.sh + resolve_test_requirements.py /
  verify_ray_test_support.py + 14 个 project-owned NPU 一致性测试)
- 上游测试(python/ray/tests/accelerators/test_npu.py 等 2 项)与
  project-owned example 共用同一 manifest(source: upstream/project)
- 新增 ray-examples.yml 薄触发器(schedule 先注释,矩阵在 NPU runner
  上全绿后再启用;default_branch=master,max_parallel=2)

引擎(examples-template.yml + check_supported_entries.py +
write_example_result.py)已支持 source: project / npu_devices,无需改动。
Comment thread .github/workflows/ray-examples.yml Outdated
on:
# Enable after the expanded matrix is green on the NPU runners.
# schedule:
# - cron: '0 18 * * *'

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

要开启定时触发,一天一次

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

已在 c93b94b 开启每日定时(UTC 18:00 = 北京时间 02:00,一天一次)。

参照 examples-guards/peft 的触发器格式:

- on 增加 pull_request(opened/synchronize/reopened,paths 限定
  examples-guards/ray/** 与本触发器文件,与 quick-start 触发器同口径)
- concurrency 改三路分组:schedule 串行(保 outcome 写回)/
  每 PR 按 number 串行且新 push 取消旧跑(省 NPU runner 排队)/
  manual 独立;PR 组名带 workflow 名前缀避免跨项目互斥
取消 schedule 注释,每日北京时间 02:00(UTC 18:00)一次,与
flash-linear-attention / peft 的触发节奏对齐;引擎 monitor 仍以上游
release / 看护文件变更门控实际执行,定时仅界定检测延迟。
@hipudding
hipudding merged commit 7e7f3cc into Ascend:main Oct 13, 2026
35 checks 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.

3 participants