Skip to content

test(settlement): qualify current Host readback in full settlement chains #6285

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
huangruiteng merged 2 commits into from
Oct 13, 2026
Merged

test(settlement): qualify current Host readback in full settlement chains #6285

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
277dab8
test(settlement): observe current Host before scheduler acknowledgement
songoow Oct 12, 2026
f6c8aae
Merge main and preserve full settlement Host readback coverage
songoow Oct 13, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
84 changes: 47 additions & 37 deletions tests/control_plane/test_quota_settlement_cli.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,6 +12,7 @@

import pytest

from examples.control_plane.quota_plan_fixtures import write_codex_heartbeat_fixture
from loopx.bootstrap_command_pack import build_start_goal_guided_packet
from loopx.control_plane.work_items.delivery_outcome import (
PROGRESS_DELIVERY_OUTCOMES,
Expand Down Expand Up @@ -47,6 +48,36 @@
SELECTED_REPLAN_TODO_ID = "todo_chain_000000000000"


@pytest.fixture
def scheduler_host_home(tmp_path: Path, monkeypatch) -> Path:
root = tmp_path / "codex-host"
monkeypatch.setenv("CODEX_HOME", str(root))
monkeypatch.setenv("CODEX_THREAD_ID", "scheduler-fixture")
return root


def _prepare_scheduler_host_readback(
registry_path: Path, runtime: Path, host_root: Path, ack_args: list[str],
) -> None:
"""Exercise the real Host reader over a synthetic manifest, not a live App."""
requested = ack_args[ack_args.index("--applied-rrule") + 1]
wrong = "FREQ=MINUTELY;INTERVAL=60"
assert requested != wrong
for observed in (None, wrong):
if observed is not None:
write_codex_heartbeat_fixture(host_root, goal_id=GOAL_ID, agent_id=AGENT_ID, rrule=observed)
rc, refused = _run_cli(registry_path, runtime, *ack_args)
assert rc == 1, refused
assert refused["scheduler_commit"]["reason_code"] == "rrule_mismatch", refused
assert refused["scheduler_commit"]["written"] is False
assert refused["scheduler_commit"]["observed_host_rrule"] == (observed or "")
assert load_scheduler_state(runtime, goal_id=GOAL_ID, agent_id=AGENT_ID,
state_key=APP_AUTOMATION_STATEFUL_BACKOFF_STATE_KEY) is None
assert _spend_run_count(runtime) == 1
# Only change the isolated Host observation, not the original receipt/hint.
write_codex_heartbeat_fixture(host_root, goal_id=GOAL_ID, agent_id=AGENT_ID, rrule=requested)


def _assert_action_selection_recovery_projections(payload: dict[str, Any]) -> None:
from loopx.control_plane.quota.turn_envelope import build_turn_envelope
from loopx.control_plane.turn_driver.host_candidate import extract_turn_authority
Expand Down Expand Up @@ -2697,13 +2728,8 @@ def test_prior_host_closeout_reads_archived_exact_todo(


def test_standard_codex_app_settlement_is_receipted_and_idempotent(
tmp_path: Path, monkeypatch: pytest.MonkeyPatch,
tmp_path: Path, scheduler_host_home: Path,
) -> None:
from examples.control_plane.quota_plan_fixtures import write_codex_heartbeat_fixture

codex_home = tmp_path / "codex-home"
monkeypatch.setenv("CODEX_HOME", str(codex_home))
monkeypatch.setenv("CODEX_THREAD_ID", "scheduler-fixture")
project, runtime, registry_path = _write_fixture(tmp_path)
_configure_read_only_todo(project)
binding = (
Expand Down Expand Up @@ -2891,10 +2917,9 @@ def test_standard_codex_app_settlement_is_receipted_and_idempotent(
TURN_ID,
"--execute",
]
# Matching Host facts cannot revive the superseded receipt; the fresh
# Turn below must still acknowledge this real isolated host observation.
# Matching Host facts cannot revive the superseded receipt.
write_codex_heartbeat_fixture(
codex_home, goal_id=GOAL_ID, agent_id=AGENT_ID,
scheduler_host_home, goal_id=GOAL_ID, agent_id=AGENT_ID,
rrule="FREQ=MINUTELY;INTERVAL=3",
)
ack_rc, ack = _run_cli(
Expand Down Expand Up @@ -2935,11 +2960,18 @@ def test_standard_codex_app_settlement_is_receipted_and_idempotent(
assert fresh_turn_rc == 0, fresh_turn
assert fresh_turn["selected_todo"]["todo_id"] == successor_id
fresh_ack_args = fresh_turn["scheduler_hint"]["app_automation"]["ack_hint"]["cli_args"]
# Remove only the isolated observation to exercise missing Host refusal.
(scheduler_host_home / "automations" / "scheduler-fixture" / "automation.toml").unlink()
_prepare_scheduler_host_readback(registry_path, runtime, scheduler_host_home, fresh_ack_args)
fresh_ack_rc, fresh_ack = _run_cli(registry_path, runtime, *fresh_ack_args)
assert fresh_ack_rc == 0, fresh_ack
assert fresh_ack["schema_version"] == "loopx_scheduler_host_followup_result_v0"
assert fresh_ack["mode"] == "scheduler-ack"
assert fresh_ack["scheduler_commit"]["written"] is True
assert load_scheduler_state(runtime, goal_id=GOAL_ID, agent_id=AGENT_ID,
state_key=APP_AUTOMATION_STATEFUL_BACKOFF_STATE_KEY)["last_applied_rrule"] == (
fresh_ack_args[fresh_ack_args.index("--applied-rrule") + 1]
)
scheduler_bytes = Path(fresh_ack["scheduler_state_path"]).read_bytes()
replay_ack_rc, replay_ack = _run_cli(registry_path, runtime, *fresh_ack_args)
assert replay_ack_rc == 0, replay_ack
Expand Down Expand Up @@ -5600,13 +5632,8 @@ def test_pending_action_selection_does_not_commit_after_new_user_gate(

@pytest.mark.parametrize("write_vision", [True, False], ids=["qualified-vision", "missing-vision"])
def test_todoless_autonomous_replan_settles_quota_refresh_spend_chain(
tmp_path: Path, write_vision: bool, monkeypatch: pytest.MonkeyPatch,
tmp_path: Path, write_vision: bool, scheduler_host_home: Path,
) -> None:
from examples.control_plane.quota_plan_fixtures import write_codex_heartbeat_fixture

codex_home = tmp_path / "codex-home"
monkeypatch.setenv("CODEX_HOME", str(codex_home))
monkeypatch.setenv("CODEX_THREAD_ID", "scheduler-fixture")
project, runtime, registry_path = _write_fixture(tmp_path)
_configure_autonomous_replan_fixture(project, runtime, registry_path)
registry = json.loads(registry_path.read_text(encoding="utf-8"))
Expand Down Expand Up @@ -5777,28 +5804,7 @@ def test_todoless_autonomous_replan_settles_quota_refresh_spend_chain(
] == "autonomous_replan"
assert _spend_run_count(runtime) == 1

# A settled receipt retains the original recipe, not authority to invent
# current Host facts. Missing and different cadences cannot mutate state.
for observed in (None, "FREQ=MINUTELY;INTERVAL=6"):
if observed is not None:
write_codex_heartbeat_fixture(
codex_home, goal_id=GOAL_ID, agent_id=AGENT_ID, rrule=observed,
)
refused_rc, refused = _run_cli(
registry_path, runtime, *original_scheduler_ack_args,
)
assert refused_rc == 1, refused
assert refused["scheduler_commit"]["reason_code"] == "rrule_mismatch"
assert refused["scheduler_commit"]["written"] is False
assert not Path(refused["scheduler_commit"]["path"]).exists()
assert _spend_run_count(runtime) == 1

# Simulate the host update through the existing isolated manifest fixture;
# command-line hints alone do not prove the actual Codex Host readback.
write_codex_heartbeat_fixture(
codex_home, goal_id=GOAL_ID, agent_id=AGENT_ID,
rrule="FREQ=MINUTELY;INTERVAL=3",
)
_prepare_scheduler_host_readback(registry_path, runtime, scheduler_host_home, original_scheduler_ack_args)
ack_rc, ack = _run_cli(
registry_path,
runtime,
Expand All @@ -5813,6 +5819,10 @@ def test_todoless_autonomous_replan_settles_quota_refresh_spend_chain(
assert ack["scheduler_commit"]["written"] is True
assert ack["scheduler_state_mutated"] is True
assert ack["appended"] is False
assert load_scheduler_state(runtime, goal_id=GOAL_ID, agent_id=AGENT_ID,
state_key=APP_AUTOMATION_STATEFUL_BACKOFF_STATE_KEY)["last_applied_rrule"] == (
original_scheduler_ack_args[original_scheduler_ack_args.index("--applied-rrule") + 1]
)
scheduler_bytes = Path(ack["scheduler_state_path"]).read_bytes()
ack_replay_rc, ack_replay = _run_cli(
registry_path, runtime, *original_scheduler_ack_args
Expand Down
Loading