Files
pipelines/docs/notes/storage-delivery-pending.md

1.9 KiB

Storage delivery — staged, service left running

Storage server and app source landed in main through 0f49b31c (backend 6abc3090 plus scope clarification; UI c9e1124b). Go build/vet/full suite passed on frozen backend source; scope follow-up focused tests passed. App checks: 15 focused, 118 Settings, 143 application; analysis and production web/Android builds passed. Native/browser evidence is in app/notes/storage-settings.md.

The live backend has not been replaced. The delivery gate found no nonterminal runs or queued prompts but one active Codex turn with a background task. Root's repository agreement prohibits restarting shared services over active work. No production policy was enabled or cleanup requested. Storage UI publication waits for its matching backend; a mobile-pane-only client release can proceed independently from feature/mobile-chat-panes.

Compiled backend/CLI are staged at /private/tmp/pipelines-storage-delivery/. The guarded delivery helper /private/tmp/pipelines-deliver-storage-backend.py checks all run states, chat activity.turn_running/background_tasks (not a nonexistent row.running field), queued prompts, and keep-session shutdown configuration. It exits without mutations while busy. When idle it backs up installed server/CLI and a SQLite snapshot before swapping binaries and using the existing service restart. It never activates retention or asks for cleanup. This temporary helper must be reviewed before reuse; normal deployment authority and the live gate still apply.

Native histories outside Pipelines data and known project directories are explicitly excluded from measured totals; the overview warns about scope. Measurements are logical regular-file lengths with inode deduplication, not filesystem allocated blocks. The configured size is a best-effort target: protected data and SQLite retained pages can keep usage above it.