Merge pull request #4422 from Yue021130/codex/3700-claim-work-parity

test(coordination): characterize file-backed claim_work with a parity fixture

Author: huangruiteng

tests/control_plane/test_coordination_provider_parity.py

@@ -0,0 +1,361 @@
+"""Characterize the shipped file-backed ``claim_work`` path against the contract.
+
+``tests/control_plane/test_coordination_executor.py`` proves the authority
+semantics with an in-memory provider, and
+``test_coordination_file_provider.py`` proves the storage verbs. This fixture
+closes the gap between them: one scenario matrix drives the real
+``CoordinationAuthorityExecutor`` through the shipped
+``FileCoordinationProvider``.
+
+The fixture is provider-neutral by construction. Every scenario speaks only the
+storage protocol (``store_identity`` / ``load`` / ``compare_and_put``) and
+receives providers from a handle factory, so registering another provider later
+requires no change to the matrix or to the expectations.
+
+Expected outcomes are declared from the coordination domain contract (RFC
+section 10, checks 1-5) before anything runs. They are never read back from
+provider output, and no authority rule is re-derived here: the executor is the
+only decision maker under test.
+
+In scope: characterization only. No production code changes, no new provider,
+no default-provider or public-behavior change, and no live NoKV, credential,
+service-startup, or provider-promotion surface.
+"""
+
+from __future__ import annotations
+
+import copy
+from dataclasses import dataclass, field
+from typing import Any, Callable
+
+import pytest
+
+from loopx.control_plane.coordination.executor import (
+    CoordinationAuthorityExecutor,
+    sample_claim_envelope,
+)
+from loopx.control_plane.coordination.file_provider import FileCoordinationProvider
+from loopx.control_plane.coordination.head import bootstrap_head
+
+
+GOAL_ID = "goal-a"
+# Fixed wall clock: every lease expiry in this matrix is minted from it.
+NOW = 1_800_000_000.0
+ELIGIBLE_AGENTS = ("agent-a", "agent-b")
+
+
+# ---- synthetic, public-safe head fixtures -----------------------------------
+
+
+def eligibility(allowed: tuple[str, ...] = ELIGIBLE_AGENTS) -> dict[str, Any]:
+    return {
+        "authorization_projection_revision": 3,
+        "authorization_projection_digest": "sha256:bootstrap-auth",
+        "allowed_agent_ids": list(allowed),
+        "dependencies_satisfied": True,
+        "dependency_revision": 12,
+        "gates_open": True,
+        "gate_revision": 5,
+    }
+
+
+def todo(**overrides: Any) -> dict[str, Any]:
+    base: dict[str, Any] = {
+        "todo_revision": 7,
+        "status": "open",
+        "claimed_by": None,
+        "eligibility": eligibility(),
+        "repository": "git:example/repo",
+        "code_revision": "0123456789abcdef",
+        "last_lease_epoch": 6,
+    }
+    base.update(overrides)
+    return base
+
+
+def claim(
+    agent: str = "agent-a",
+    todo_id: str = "todo-1",
+    operation_id: str | None = None,
+    **overrides: Any,
+) -> dict[str, Any]:
+    return sample_claim_envelope(
+        goal_id=GOAL_ID,
+        operation_id=operation_id or f"op-{agent}-{todo_id}",
+        agent_id=agent,
+        device_id=f"dev-{agent}",
+        todo_id=todo_id,
+        expected_todo_revision=7,
+        expected_preconditions={
+            "authorization_projection_revision": 3,
+            "authorization_projection_digest": "sha256:bootstrap-auth",
+            "dependency_revision": 12,
+            "gate_revision": 5,
+        },
+        lease_ttl_seconds=600,
+        **overrides,
+    )
+
+
+def bootstrap(provider: Any, todo_ids: tuple[str, ...] = ("todo-1", "todo-2")) -> None:
+    head = bootstrap_head(
+        GOAL_ID,
+        {todo_id: todo() for todo_id in todo_ids},
+        store_binding=provider.store_identity(),
+    )
+    assert provider.compare_and_put(0, head)["result"] == "applied"
+
+
+def executor_for(provider: Any) -> CoordinationAuthorityExecutor:
+    return CoordinationAuthorityExecutor(provider, goal_id=GOAL_ID, now=lambda: NOW)
+
+
+class RecordingProvider:
+    """Storage-only recorder: delegates every verb and records CAS expectations.
+
+    It holds no authority semantics; it exists so a scenario can prove that a
+    stale generation reached ``compare_and_put`` without duplicating a write.
+    """
+
+    def __init__(self, inner: Any) -> None:
+        self._inner = inner
+        self.cas_expectations: list[int] = []
+
+    def store_identity(self) -> str:
+        return self._inner.store_identity()
+
+    def load(self) -> tuple[dict[str, Any] | None, int]:
+        return self._inner.load()
+
+    def compare_and_put(
+        self, expected_provider_generation: int, head: dict[str, Any]
+    ) -> dict[str, Any]:
+        self.cas_expectations.append(expected_provider_generation)
+        return self._inner.compare_and_put(expected_provider_generation, head)
+
+
+# ---- observation and expectation --------------------------------------------
+
+
+@dataclass
+class Observation:
+    """What one scenario actually produced, in contract terms only."""
+
+    outcomes: list[dict[str, Any]] = field(default_factory=list)
+    head: dict[str, Any] = field(default_factory=dict)
+    flags: dict[str, bool] = field(default_factory=dict)
+
+
+@dataclass(frozen=True)
+class Expectation:
+    """The contract-derived verdict one scenario must produce."""
+
+    results: tuple[str, ...]
+    reasons: tuple[str | None, ...]
+    authority_revision: int
+    receipts: tuple[str, ...]
+    claimed_by: tuple[tuple[str, str | None], ...]
+    todo_revisions: tuple[tuple[str, int], ...]
+    flags: tuple[tuple[str, bool], ...] = ()
+
+
+# ---- scenarios --------------------------------------------------------------
+
+
+def same_target_competition(handles: Callable[[], Any]) -> Observation:
+    """Check 1: two actors on one target leave exactly one winner."""
+
+    provider = handles()
+    bootstrap(provider)
+    executor = executor_for(provider)
+    observation = Observation()
+    observation.outcomes.append(executor.apply(claim("agent-a", "todo-1")))
+    observation.outcomes.append(executor.apply(claim("agent-b", "todo-1")))
+    observation.head, _ = provider.load()
+    return observation
+
+
+def independent_targets_rebase(handles: Callable[[], Any]) -> Observation:
+    """Check 2: independent targets both apply through internal rebase."""
+
+    provider = handles()
+    bootstrap(provider)
+    observation = Observation()
+    observation.outcomes.append(
+        executor_for(provider).apply(claim("agent-a", "todo-1"))
+    )
+    observation.outcomes.append(
+        executor_for(provider).apply(claim("agent-b", "todo-2"))
+    )
+    observation.head, _ = provider.load()
+    return observation
+
+
+def replay_after_interleaved_write(handles: Callable[[], Any]) -> Observation:
+    """Check 3: `A -> B -> replay A` returns the exact original receipt.
+
+    The replay runs through a freshly constructed executor and a fresh provider
+    handle, so nothing is served from in-process state.
+    """
+
+    provider = handles()
+    bootstrap(provider)
+    executor = executor_for(provider)
+    observation = Observation()
+    request_a = claim("agent-a", "todo-1", operation_id="op-A")
+    first = executor.apply(request_a)
+    observation.outcomes.append(first)
+    observation.outcomes.append(
+        executor.apply(claim("agent-b", "todo-2", operation_id="op-B"))
+    )
+    reconstructed = executor_for(handles())
+    replay = reconstructed.apply(copy.deepcopy(request_a))
+    observation.outcomes.append(replay)
+    observation.flags["replay_returns_original_receipt"] = (
+        replay.get("original_receipt") == first.get("original_receipt")
+        and replay.get("original_receipt") is not None
+    )
+    observation.head, _ = provider.load()
+    return observation
+
+
+def operation_identity_reuse(handles: Callable[[], Any]) -> Observation:
+    """Check 5: one operation id with different semantics changes nothing."""
+
+    provider = handles()
+    bootstrap(provider)
+    executor = executor_for(provider)
+    observation = Observation()
+    request = claim("agent-a", "todo-1", operation_id="op-A")
+    observation.outcomes.append(executor.apply(request))
+    before, _ = provider.load()
+    mutated = copy.deepcopy(request)
+    mutated["command"]["lease_ttl_seconds"] = 601
+    observation.outcomes.append(executor.apply(mutated))
+    observation.head, _ = provider.load()
+    observation.flags["state_unchanged_after_rejection"] = observation.head == before
+    return observation
+
+
+def stale_generation_does_not_duplicate(handles: Callable[[], Any]) -> Observation:
+    """Stale provider generation is refused rather than replayed.
+
+    A handle observes a generation, the document advances behind it, and the
+    stale expectation is then offered for a raw CAS. The refused attempt must
+    not write, and the executor's bounded reload must land exactly one further
+    authority transition with no duplicate receipt.
+    """
+
+    provider = handles()
+    bootstrap(provider)
+    stale = RecordingProvider(handles())
+    stale_generation = stale.load()[1]
+
+    observation = Observation()
+    observation.outcomes.append(
+        executor_for(provider).apply(claim("agent-a", "todo-1"))
+    )
+    current_head, current_generation = provider.load()
+    stale_attempt = stale.compare_and_put(stale_generation, current_head)
+    observation.flags["stale_cas_refused"] = stale_attempt["result"] == "conflict"
+    observation.flags["stale_cas_wrote_nothing"] = provider.load()[1] == current_generation
+    observation.flags["stale_expectation_offered"] = bool(
+        stale.cas_expectations and stale.cas_expectations[0] == stale_generation
+    )
+
+    observation.outcomes.append(
+        executor_for(stale).apply(claim("agent-b", "todo-2"))
+    )
+    observation.head, _ = provider.load()
+    return observation
+
+
+SCENARIOS: dict[str, Callable[[Callable[[], Any]], Observation]] = {
+    "same_target_competition": same_target_competition,
+    "independent_targets_rebase": independent_targets_rebase,
+    "replay_after_interleaved_write": replay_after_interleaved_write,
+    "operation_identity_reuse": operation_identity_reuse,
+    "stale_generation_does_not_duplicate": stale_generation_does_not_duplicate,
+}
+
+EXPECTED: dict[str, Expectation] = {
+    "same_target_competition": Expectation(
+        results=("applied", "conflict"),
+        reasons=(None, "todo_revision_mismatch"),
+        authority_revision=1,
+        receipts=("op-agent-a-todo-1",),
+        claimed_by=(("todo-1", "agent-a"), ("todo-2", None)),
+        todo_revisions=(("todo-1", 8), ("todo-2", 7)),
+    ),
+    "independent_targets_rebase": Expectation(
+        results=("applied", "applied"),
+        reasons=(None, None),
+        authority_revision=2,
+        receipts=("op-agent-a-todo-1", "op-agent-b-todo-2"),
+        claimed_by=(("todo-1", "agent-a"), ("todo-2", "agent-b")),
+        todo_revisions=(("todo-1", 8), ("todo-2", 8)),
+    ),
+    "replay_after_interleaved_write": Expectation(
+        results=("applied", "applied", "already_applied"),
+        reasons=(None, None, None),
+        authority_revision=2,
+        receipts=("op-A", "op-B"),
+        claimed_by=(("todo-1", "agent-a"), ("todo-2", "agent-b")),
+        todo_revisions=(("todo-1", 8), ("todo-2", 8)),
+        # A replayed operation is a read, not a second transition.
+        flags=(("replay_returns_original_receipt", True),),
+    ),
+    "operation_identity_reuse": Expectation(
+        results=("applied", "rejected"),
+        reasons=(None, "operation_identity_mismatch"),
+        authority_revision=1,
+        receipts=("op-A",),
+        claimed_by=(("todo-1", "agent-a"), ("todo-2", None)),
+        todo_revisions=(("todo-1", 8), ("todo-2", 7)),
+        flags=(("state_unchanged_after_rejection", True),),
+    ),
+    "stale_generation_does_not_duplicate": Expectation(
+        results=("applied", "applied"),
+        reasons=(None, None),
+        authority_revision=2,
+        receipts=("op-agent-a-todo-1", "op-agent-b-todo-2"),
+        claimed_by=(("todo-1", "agent-a"), ("todo-2", "agent-b")),
+        todo_revisions=(("todo-1", 8), ("todo-2", 8)),
+        flags=(
+            ("stale_expectation_offered", True),
+            ("stale_cas_refused", True),
+            ("stale_cas_wrote_nothing", True),
+        ),
+    ),
+}
+
+
+@pytest.fixture
+def handles(tmp_path) -> Callable[[], Any]:
+    """Return a factory for fresh provider handles onto one shared store."""
+
+    directory = tmp_path / "coordination"
+    return lambda: FileCoordinationProvider(directory, GOAL_ID)
+
+
+@pytest.mark.parametrize("name", sorted(SCENARIOS))
+def test_claim_work_contract_holds_on_the_shipped_provider(name, handles):
+    expected = EXPECTED[name]
+    observation = SCENARIOS[name](handles)
+
+    assert tuple(item["result"] for item in observation.outcomes) == expected.results
+    assert tuple(item.get("reason") for item in observation.outcomes) == expected.reasons
+    assert observation.head["authority_revision"] == expected.authority_revision
+    assert tuple(sorted(observation.head["receipt_index"])) == expected.receipts
+
+    todos = observation.head["coordination"]["todos"]
+    assert tuple(
+        (todo_id, todos[todo_id]["claimed_by"]) for todo_id, _ in expected.claimed_by
+    ) == expected.claimed_by
+    assert tuple(
+        (todo_id, todos[todo_id]["todo_revision"])
+        for todo_id, _ in expected.todo_revisions
+    ) == expected.todo_revisions
+
+    for flag, value in expected.flags:
+        assert observation.flags.get(flag) is value, flag