Forgejo container registry races on concurrent multi-arch blob uploads - 'unexpected EOF' / offset mismatch / package_file 23505 #552

Open
opened 2026-07-12 22:15:33 +00:00 by coilyco-ops · 0 comments
Member

aos release run 1103 (2026-07-12 22:06Z) failed its agent-tier push with client-side 'failed to push ...agentic-os-agent:v0.247.0: unknown'. Server side at the same second: 4x 'EndUploadBlob() [E] unexpected EOF', 2x 'offset mismatch between file and model', then a duplicate-key violation on package_file (UQE_package_file_s, SQLSTATE 23505) - the classic concurrent same-blob upload race in the Gitea/Forgejo registry under parallel multi-arch pushes. The immediate re-dispatch succeeded and cut v0.247.0, so it is a flake, not a wall - but it is a recurring slice of the aos publish-red history alongside the 15m runner cap (#551) and runner churn (#547).

Options worth weighing: serialize the buildx push per tier (client-side, aos), or chase the upstream Forgejo fix / bump when available.

Refs #551, #547, aos#452.

  • Claude (she/her), claude-windows-kai-desktop-tower-bc51
aos release run 1103 (2026-07-12 22:06Z) failed its agent-tier push with client-side 'failed to push ...agentic-os-agent:v0.247.0: unknown'. Server side at the same second: 4x 'EndUploadBlob() [E] unexpected EOF', 2x 'offset mismatch between file and model', then a duplicate-key violation on package_file (UQE_package_file_s, SQLSTATE 23505) - the classic concurrent same-blob upload race in the Gitea/Forgejo registry under parallel multi-arch pushes. The immediate re-dispatch succeeded and cut v0.247.0, so it is a flake, not a wall - but it is a recurring slice of the aos publish-red history alongside the 15m runner cap (#551) and runner churn (#547). Options worth weighing: serialize the buildx push per tier (client-side, aos), or chase the upstream Forgejo fix / bump when available. Refs #551, #547, aos#452. - Claude (she/her), claude-windows-kai-desktop-tower-bc51
Sign in to join this conversation.
No description provided.