feat(projection): add the fleet reset as a loop over the repository form
All checks were successful
CI Smoke / host-smoke (push) Successful in 0s
CI Smoke / container-smoke (push) Successful in 1s
Build and Publish Multi-Context Image / build-and-push (push) Successful in 25s

ADR-012 decision 7 requires the fleet form to share the per-repository
implementation: the rarely-run wide operation must be the frequently-run narrow
one, or the wide one is trusted on the strength of never having been exercised.

Failure behaviour is the substance. A refusal does not stop the pass — aborting
on the first refusal means one unresolved repository blocks reconstruction
everywhere, which in practice means permanently. An error does not stop it
either. Each repository gets its own session so one failure cannot roll back
another's work, and only repositories that applied are committed.

Refs STATE-WP-0083-T04

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

Assistant: claude-code
Assistant-Model: opus
Assistant-Process: 2583210@bnt-lap001
Assistant-Session: f2bff2d5-e9b2-4338-92ca-10282a927006
This commit is contained in:
tegwick 2026-08-26 13:23:08 +02:00
parent 76a7c7ed24
commit 532583ce17
3 changed files with 173 additions and 1 deletions

View file

@ -618,3 +618,72 @@ async def reset_repository_projection(
"canonical identifier (STATE-WP-0083-T06)."
)
return outcome
# ---------------------------------------------------------------------------
# Fleet form (STATE-WP-0083-T04)
# ---------------------------------------------------------------------------
@dataclass
class FleetResetOutcome:
results: dict[str, dict[str, Any]] = field(default_factory=dict)
errors: dict[str, str] = field(default_factory=dict)
def to_dict(self) -> dict[str, Any]:
by_status: dict[str, int] = {}
for r in self.results.values():
by_status[r["status"]] = by_status.get(r["status"], 0) + 1
return {
"schema": "state-hub.fleet-projection-reset.v1",
"repositories": len(self.results) + len(self.errors),
"by_status": by_status,
"errored": len(self.errors),
"totals": {
k: sum(r["counts"][k] for r in self.results.values())
for k in ("created", "updated", "retired", "refused")
},
"results": self.results,
"errors": self.errors,
}
async def reset_fleet_projection(
session_factory: Any,
repo_slugs: list[str],
*,
acknowledge_retirements: bool = False,
forge_base: str = DEFAULT_FORGE_BASE,
) -> FleetResetOutcome:
"""Reset every repository, one at a time, sharing the per-repository path.
The fleet form is a loop over the repository form and nothing else
(`ADR-012` decision 7). The rarely-run wide operation must be the frequently
run narrow one, or the wide one is trusted on the strength of never having
been exercised.
A repository that refuses or errors is recorded and the pass continues.
Aborting on the first refusal would mean one unresolved repository blocks
reconstruction everywhere which in practice means permanently.
Each repository gets its own session, so one failure cannot roll back
another's work or leave a poisoned transaction behind.
"""
outcome = FleetResetOutcome()
for slug in repo_slugs:
try:
async with session_factory() as session:
result = await reset_repository_projection(
session,
slug,
acknowledge_retirements=acknowledge_retirements,
forge_base=forge_base,
)
if result.status == "applied":
await session.commit()
else:
await session.rollback()
outcome.results[slug] = result.to_dict()
except Exception as exc: # noqa: BLE001 - one repo must not end the pass
outcome.errors[slug] = f"{type(exc).__name__}: {exc}"[:300]
return outcome

View file

@ -249,6 +249,9 @@ class _FakeSession:
async def commit(self):
self.committed = True
async def rollback(self):
self.rolled_back = True
class TestCollisionRefusal:
"""A colliding identifier must refuse, not raise (STATE-WP-0083-T03).
@ -299,3 +302,85 @@ class TestCollisionRefusal:
session = _FakeSession(repo=_Repo(), rows=[], foreign=[])
out = await fp.reset_repository_projection(session, "demo", derived=derived)
assert out.status == "applied" and out.created == ["DEMO-WP-0001"]
class TestFleetReset:
"""The fleet form must be a loop over the repository form (T04).
Its value is entirely in what it does with failure: a wide reset that stops
at the first refusal is one nobody can run, because there is always one
unresolved repository somewhere.
"""
class _Factory:
def __init__(self, sessions):
self._sessions = list(sessions)
def __call__(self):
session = self._sessions.pop(0)
class Ctx:
async def __aenter__(self_inner):
return session
async def __aexit__(self_inner, *a):
return False
return Ctx()
@staticmethod
def _derived(rid):
return fp.DerivedProjection(
repo_slug="x", commit="c0ffee",
workplans=[fp.DerivedWorkplan(
record_id=rid, uuid=fp.derived_record_uuid(rid), title=rid,
status="active", relative_path="workplans/a.md", archived=False, tasks=[])])
@pytest.mark.asyncio
async def test_a_refusal_does_not_stop_the_pass(self, monkeypatch):
calls = []
async def fake(session, slug, **kw):
calls.append(slug)
out = fp.ResetOutcome(repo_slug=slug, commit="c0ffee", status="applied")
if slug == "bad":
out.status = "refused"
out.refused.append({"reason": "would be retired", "slug": "x"})
else:
out.updated.append("A-WP-0001")
return out
monkeypatch.setattr(fp, "reset_repository_projection", fake)
s = [_FakeSession(repo=_Repo(), rows=[]) for _ in range(3)]
res = await fp.reset_fleet_projection(self._Factory(s), ["good", "bad", "also-good"])
assert calls == ["good", "bad", "also-good"]
d = res.to_dict()
assert d["by_status"] == {"applied": 2, "refused": 1}
assert d["totals"]["updated"] == 2
@pytest.mark.asyncio
async def test_an_error_does_not_stop_the_pass(self, monkeypatch):
async def fake(session, slug, **kw):
if slug == "boom":
raise RuntimeError("clone failed")
return fp.ResetOutcome(repo_slug=slug, commit="c", status="noop")
monkeypatch.setattr(fp, "reset_repository_projection", fake)
s = [_FakeSession(repo=_Repo(), rows=[]) for _ in range(3)]
res = await fp.reset_fleet_projection(self._Factory(s), ["a", "boom", "b"])
d = res.to_dict()
assert d["errored"] == 1 and "clone failed" in res.errors["boom"]
assert set(res.results) == {"a", "b"}
@pytest.mark.asyncio
async def test_only_applied_repositories_are_committed(self, monkeypatch):
async def fake(session, slug, **kw):
out = fp.ResetOutcome(repo_slug=slug, commit="c", status="applied")
if slug == "refuser":
out.status = "refused"
else:
out.updated.append("A-WP-0001")
return out
monkeypatch.setattr(fp, "reset_repository_projection", fake)
s = [_FakeSession(repo=_Repo(), rows=[]) for _ in range(2)]
await fp.reset_fleet_projection(self._Factory(s), ["applier", "refuser"])
assert s[0].committed is True
assert s[1].committed is False

View file

@ -184,7 +184,7 @@ so partial convergence cannot be mistaken for full. 678 tests pass.
```task
id: STATE-WP-0083-T04
status: todo
status: progress
priority: medium
```
@ -196,6 +196,24 @@ the pass.
Acceptance: a fleet dry-run reports per-repository outcomes including refusals,
and completes despite them.
**Built (2026-08-26).** `reset_fleet_projection()` is a loop over the
per-repository form and nothing else, so the rarely-run wide operation is the
frequently-run narrow one rather than a second implementation trusted on the
strength of never having been exercised.
Its behaviour on failure is the part that matters, and all three are tested:
- **A refusal does not stop the pass.** Aborting on the first refusal means one
unresolved repository blocks reconstruction everywhere, which in practice means
permanently.
- **An error does not stop the pass.** A repository that cannot be cloned is
recorded and the loop continues.
- **Each repository gets its own session**, so one failure cannot roll back
another's work or leave a poisoned transaction behind. Only repositories that
applied are committed.
694 tests pass. Needs deployment before it can run against central.
## Retire what the reset replaces
```task