pulp / pulp/pulp_python

Pull-through cache: `.whl.metadata` (PEP 658) requests raise `ValueError: True is not in list` on 3.29.0 — residual case of #1083

未关闭
#1,219 1 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

还没有人认领这个 Issue。

Issue
主要语言
Python
星标
49
派生
88
平均合并
1 天 10 小时
30 天内合并 PR
31

描述

Version

  • pulp_python 3.29.0
  • pulpcore 3.108.0
  • Installation: container image (Pulp operator on Kubernetes)

Describe the bug

On a pull-through PyPI distribution, when a client requests <wheel>.whl.metadata (PEP 658), pulp-content raises ValueError: True is not in list in pulp_python/app/utils.py:get_project_metadata_from_file and returns a 5xx / aborted stream. A direct GET also surfaces a misleading 404 with an upstream URL that doesn't exist (https://pypi.org/simple/<package>/<wheel>.whl.metadata — PEP 658 metadata files live next to the wheel on files.pythonhosted.org/packages/<hash>/..., not under the simple index path).

This appears to be a residual case of #1083. PR #1086 fixed the .whl fetch path by passing suffix=filename into the temp file, so endswith(".whl") matches. But when the requested artifact is the .whl.metadata companion, init_from_artifact_and_relative_path is still called, path.name ends in .metadata, the temp file ends in .metadata, and DIST_EXTENSIONS has no .metadata entry — so .index(True) raises.

To Reproduce

  1. Configure a pull-through PyPI distribution:
pulp python remote create       --name pypi --url https://pypi.org/ --policy on_demand
pulp python distribution create --name pypi-proxy --base-path pypi --remote pypi                                                                                                                                                                      
  1. From a clean state (no cached package), request the PEP 658 metadata file directly:
curl -sk -o - -w "HTTP %{http_code}\n" \               
"https://<pulp>/pulp/content/<domain>/pypi/simple/ruff/ruff-0.12.12-py3-none-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata"                                                                                                               

Equivalent: run uv lock / uv sync against this distribution from a cold uv cache — uv issues .whl.metadata requests per PEP 658.
3. Observe a 404 to the client and the following traceback in pulp-content logs:

File "pulpcore/content/handler.py", line 1033, in _save_artifact                                                                                                                                                                                      
content = c_type.init_from_artifact_and_relative_path(artifact, rel_path)                                                                                                                                                                           
File "pulp_python/app/models.py", line 223, in init_from_artifact_and_relative_path
data = artifact_to_python_content_data(path.name, artifact, domain=get_domain())                                                                                                                                                                    
File "pulp_python/app/utils.py", line 273, in artifact_to_python_content_data
metadata = get_project_metadata_from_file(temp_file.name)                                                                                                                                                                                           
File "pulp_python/app/utils.py", line 195, in get_project_metadata_from_file
pkg_type_index = [filename.endswith(ext) for ext in extensions].index(True)                                                                                                                                                                         
ValueError: True is not in list                        

Expected behavior

Either:

  • The .metadata companion is served from upstream (fetched alongside / extracted from the wheel) so PEP 658 works through the pull-through cache, or
  • The request returns a clean 404 without invoking the python-package metadata extractor on a non-package file.

In both cases get_project_metadata_from_file should not be called with a filename whose extension isn't in DIST_EXTENSIONS.

Additional context

  • Original report and partial fix: #1083, PR #1086 ("Fix pull-through metadata serving and saving", merged 2026-02-06; backported to 3.24 in #1089).
  • Related: #1101 (PR #1102) — metadata file mismatched against wheel.
  • 3.29.0 source for artifact_to_python_content_data confirms the suffix=filename fix from #1086 is present, so the .whl path works; the .whl.metadata path bypasses that protection because filename itself ends in .metadata.
  • Likely fix direction: short-circuit the companion .metadata case before artifact_to_python_content_data is called (e.g. in init_from_artifact_and_relative_path when relative_path ends in .metadata), since those bytes already are the metadata and don't need extracting.
  • Effect on clients: uv resolutions against the pull-through distribution silently report "only <old-cached-version> is available" because per-version metadata fetches all crash server-side; only versions whose metadata happens to already be cached in Postgres remain visible.

贡献指南

打开贡献指南

从这里开始

  1. 先读完整个 Issue,再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库,在一个分支上完成修改。
  4. 提交 Pull Request,并在描述里引用这个 Issue 编号。

调研方向

Start in pulp_python/app/models.py at init_from_artifact_and_relative_path and pulp_python/app/utils.py at artifact_to_python_content_data and get_project_metadata_from_file; reproduce the cold-cache .whl.metadata request with curl or uv. Trace the companion metadata path and verify it no longer invokes the extractor for an unsupported extension, while the request either serves metadata or returns a clean 404 without the ValueError.

由索引模型根据 Issue 内容生成。

评估

技术栈
python
领域
backend
Issue 类型
缺陷
难度
3/5
预计耗时
1-2 天
活跃度
冷清
描述清晰度
基本清楚
新手友好度
58/100

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。