sched: require proven ownership at poll/admin/recovery entries (F5)

This commit is contained in:
lda
2026-09-08 11:24:40 +07:00 Verified
parent 15bd472bdc
commit 993ed07fd3
7 changed files with 293 additions and 22 deletions
+191
View File
@@ -0,0 +1,191 @@
"""Ownership guards at scheduler mutation/recovery/dispatch entries (R4/F5).
The held lock must be proven at the entry points, not merely exist as a
class: ``poll``, schedule administration, and ``recover`` reject when the
caller cannot prove live ownership, before any write or side effect.
"""
from __future__ import annotations
from datetime import UTC, datetime, timedelta
from pathlib import Path
from typing import Any
import pytest
from tests.scheduling.controlled import (
DictDeployments,
ScriptedDispatcher,
fixture_environment,
)
from wf_artifacts.runs.store import FileRunStore
from wf_scheduling import recovery as sched_recovery
from wf_scheduling.calendar import OneShotSource
from wf_scheduling.models import Schedule
from wf_scheduling.ownership import SchedulerOwnership, SecondOwnerError
from wf_scheduling.poll import Scheduler
from wf_scheduling.prepare import SchedulePreparer
from wf_scheduling.store import FileScheduleStore
def ts(y: int, mo: int, d: int, h: int = 0, mi: int = 0) -> datetime:
return datetime(y, mo, d, h, mi, tzinfo=UTC)
def _sched_model(sid: str, **kw: Any) -> Schedule:
now = ts(2026, 9, 8, 12, 0)
base: dict[str, Any] = {
"id": sid,
"deployment_id": "dep-1",
"trigger": {"kind": "cron", "expression": "0 * * * *", "timezone": "UTC"},
"input_bindings": [],
"created_at": now.isoformat(),
"updated_at": now.isoformat(),
}
base.update(kw)
return Schedule.model_validate(base)
def _scheduler(
tmp_path: Path,
ownership: SchedulerOwnership,
*,
script: dict | None = None,
) -> tuple[Scheduler, FileScheduleStore, FileRunStore]:
sched_store = FileScheduleStore(tmp_path / "sched")
run_store = FileRunStore(tmp_path / "runs")
sched = Scheduler(
schedule_store=sched_store,
run_store=run_store,
sources={},
capacity=4,
preparer=SchedulePreparer(
DictDeployments({"dep-1": {"rev": 1, "required": []}}),
fixture_environment,
),
dispatcher=ScriptedDispatcher(script),
ownership=ownership,
)
return sched, sched_store, run_store
def _due_setup(sched: Scheduler, store: FileScheduleStore, intended: datetime) -> None:
store.create_schedule(_sched_model("a"))
store.save_consumed("a", intended - timedelta(hours=1))
sched.sources["a"] = OneShotSource(intended)
def test_poll_without_ownership_rejects_before_writes(tmp_path: Path) -> None:
calls: list[str] = []
def spy(admission: Any, now: datetime) -> Any:
calls.append(admission.id)
raise AssertionError("dispatcher must not run without ownership")
ownership = SchedulerOwnership(tmp_path / "sched", owner="never-acquired")
sched, store, runs = _scheduler(tmp_path, ownership)
sched.dispatcher = ScriptedDispatcher({"*": spy})
intended = ts(2026, 9, 8, 12, 0)
_due_setup(sched, store, intended)
with pytest.raises(SecondOwnerError):
sched.poll(intended)
assert calls == []
assert runs.list_runs() == []
assert runs.list_admissions() == []
assert store.list_occurrences("a", limit=100)["total"] == 0
assert store.get_consumed("a") == intended - timedelta(hours=1)
def test_poll_with_released_ownership_rejects(tmp_path: Path) -> None:
ownership = SchedulerOwnership(tmp_path / "sched", owner="test").acquire()
sched, store, runs = _scheduler(tmp_path, ownership)
intended = ts(2026, 9, 8, 12, 0)
_due_setup(sched, store, intended)
ownership.release()
assert not ownership.held
with pytest.raises(SecondOwnerError):
sched.poll(intended)
assert runs.list_runs() == []
assert store.list_occurrences("a", limit=100)["total"] == 0
def test_recover_without_ownership_rejects_before_writes(tmp_path: Path) -> None:
from tests.artifacts.test_run_store import artifact as _artifact
from tests.artifacts.test_run_store import deployment as _deployment
from wf_api.run_lifecycle import materialize_admitted_view, persist_admission
from wf_artifacts import PinnedRunEnvironment
sched_store = FileScheduleStore(tmp_path / "sched")
run_store = FileRunStore(tmp_path / "runs")
sched_store.create_schedule(_sched_model("a"))
env = PinnedRunEnvironment(
deployment=_deployment(), root_artifact=_artifact(), child_artifacts=[]
)
admission = persist_admission(
store=run_store,
run_id=run_store.allocate_run_id(),
environment=env,
resolved_input={},
max_steps=None,
scheduled_at=ts(2026, 9, 8, 12, 0),
schedule_id="a",
schedule_revision=1,
)
materialize_admitted_view(store=run_store, admission=admission)
ownership = SchedulerOwnership(tmp_path / "sched", owner="never-acquired")
with pytest.raises(SecondOwnerError):
sched_recovery.recover(
schedule_store=sched_store,
run_store=run_store,
now=ts(2026, 9, 8, 12, 0),
ownership=ownership,
)
# Rejected before any write: the admitted run is untouched.
assert run_store.get_run(admission.id).status.value == "admitted"
def test_held_ownership_allows_poll_and_recovery(tmp_path: Path) -> None:
ownership = SchedulerOwnership(tmp_path / "sched", owner="test").acquire()
try:
sched, store, runs = _scheduler(tmp_path, ownership, script={"*": "hang"})
intended = ts(2026, 9, 8, 12, 0)
_due_setup(sched, store, intended)
assert sched.poll(intended) == {"a": f"admit:{runs.list_runs()[0].id}"}
diags = sched_recovery.recover(
schedule_store=store,
run_store=runs,
now=intended,
ownership=ownership,
)
assert diags != []
finally:
ownership.release()
def test_second_owner_cannot_acquire_for_poll(tmp_path: Path) -> None:
first = SchedulerOwnership(tmp_path / "sched", owner="first").acquire()
try:
second = SchedulerOwnership(tmp_path / "sched", owner="second")
with pytest.raises(SecondOwnerError):
second.acquire()
assert not second.held
# A scheduler proving the loser's (unheld) ownership is rejected.
sched, store, _ = _scheduler(tmp_path, second)
intended = ts(2026, 9, 8, 12, 0)
_due_setup(sched, store, intended)
with pytest.raises(SecondOwnerError):
sched.poll(intended)
finally:
first.release()
def test_admin_mutations_require_ownership(tmp_path: Path) -> None:
ownership = SchedulerOwnership(tmp_path / "sched", owner="test")
sched, store, _ = _scheduler(tmp_path, ownership)
store.create_schedule(_sched_model("a", paused=True))
with pytest.raises(SecondOwnerError):
sched.resume_schedule("a", ts(2026, 9, 8, 12, 0))
with pytest.raises(SecondOwnerError):
sched.edit_schedule("a", ts(2026, 9, 8, 12, 0))
assert store.get_schedule("a").paused is True
assert store.get_schedule("a").revision == 1