Skip to content

Pipeline file cache publication for streamed wheels - #21333

Draft
charliermarsh wants to merge 2 commits into
charlie/wheel-mode-manifestsfrom
charlie/pipeline-wheel-file-cache
Draft

Pipeline file cache publication for streamed wheels#21333
charliermarsh wants to merge 2 commits into
charlie/wheel-mode-manifestsfrom
charlie/pipeline-wheel-file-cache

Conversation

@charliermarsh

@charliermarsh charliermarsh commented Aug 28, 2026

Copy link
Copy Markdown
Member

Summary

We currently finish streaming extraction before publishing the extracted files to files-v0. This PR publishes each completed file after checking its local header and data descriptor, so cache linking overlaps downloading and extracting the remaining payloads. A bounded queue feeds batches to blocking workers, retaining the existing grouping by cache shard.

We still keep RECORD private and wait for extraction and file publication to finish before persisting the executable manifest and exposing the wheel in archive-v0. A rejected or interrupted ZIP can leave completed file objects behind; cache pruning removes those unreferenced objects. The temporary archive stays alive while any blocking publication work is running, including when extraction is cancelled.

This uses the existing streaming download without additional requests. Executable status still comes from the central directory and is stored in the manifest; the published file objects remain non-executable.

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.

1 participant