Files
PaperClipAI/doc
DottaandPaperclip 349b8b3f7b Migrate settled controller checkpoints while unrelated work runs (#19631)
## Thinking Path

> - Paperclip manages work performed by AI agents.
> - Server replacement must preserve retained Runner sessions.
> - Older sessions can retain controller checkpoints on the server
volume.
> - The existing importer requires all execution writers to stop,
including unrelated runs.
> - This change adds a separate online path for successful, settled
sources.
> - It locks the source task and finalizer while it reads and publishes
that source.
> - Unrelated work continues. The offline importer keeps its existing
authority.

## Linked Issues or Issue Description

**What happened?**

A retained controller checkpoint cannot be converted by the existing
maintenance importer while another task runs. Requiring a busy server to
become globally idle prevents online migration.

**Expected behavior**

An operator can convert a successful retained source after its finalizer
releases ownership, while unrelated tasks continue. An expired owner
remains an owner. Related active work and uncertain evidence remain
blocked.

**Steps to reproduce**

1. Retain a valid legacy controller for a successful run with an
accepted result and a released committed finalizer.
2. Start a different task.
3. Run controller preparation. The old importer refuses because
execution writers are active.

Related preservation contracts: #19038 and #19024. This change does not
replace workspace migration or authorize volume removal.

## What Changed

- Add `prestageNativeControllers` with task, finalizer, and source-run
row locks. Recheck related runs and exact checkpoint identity under
those locks.
- Allow legacy checkpoint material through normal settled-owner import
checks. Keep offline authority separate.
- Compare captured local controller and supervisor bytes before commit.
A changed source rolls back publication.
- Add `--controllers-only` to the existing finite native preparation
entry. This avoids re-reading provider archives during controller
conversion.
- Add PostgreSQL cases for unrelated work, expired owners, related
active work, source-lock contention, and source changes.
- Document the finite command and its limits.

## Verification

- Real PostgreSQL affected suites: 56 passed across controller migration
and recovery. A first per-server run was interrupted because it was
slow; the completed run used the repository's shared PostgreSQL test
wrapper.
- Initial online cases: 3 passed, covering unrelated work, an expired
owner, and related active work.
- Source lock and changed-file cases: 2 passed. Publication rolled back
on a source change.
- Final source-evidence regression: 2 passed. A first unproven source
deferred and a later valid source imported; changed-file rollback
remained intact.
- Server TypeScript check passed with the SDK and Runner declarations
built.
- Server package build passed, including the Runner vendor checks and
compiled command.
- Runtime bundling suite: 10 passed.
- Full repository CI and exact-head automated review remain required
before merge. This draft is not staging acceptance.


- PASS: 7 CLI cases after adding the controller-only dependency fixture
and explicit-mode coverage. The first CI run exposed missing mock/export
coverage; the fixture repair passed after rebase.
- The unrelated model-default failures were fixed on master; this branch
now includes that fix.

## Risks

- A source task can wait briefly on its own migration lock. A one-second
lock acquisition budget defers a contended source. No global admission
hold is acquired.
- Only successful sources with a released committed finalizer are
eligible. Failed or unsettled sources stay deferred and require their
preservation owners.
- The operator must run the command on the host that owns the original
files. The command does not certify enrollment readiness or detach any
volume.
- The existing offline path remains unchanged and still requires stopped
writers.

## Model Used

OpenAI Codex, GPT-6 family, with reasoning, tool use, and code
execution. The exact serving model ID and context window are not exposed
in this session.

## Checklist

- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [ ] All Paperclip CI gates are green
- [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge

---------

Co-authored-by: Paperclip <noreply@paperclip.ing>
2026-11-06 17:19:23 -06:00
..
…
…
…
…
…