Artemis-Eapil/tests/unit/runtime/test_queue_file_io.py

"""Real Windows handle sharing and bounded queue-write failure diagnostics."""

import errno
import json
import multiprocessing
import os
from pathlib import Path
import threading
from types import SimpleNamespace

import pytest

import artemis.runtime.device_lock as module
from artemis.runtime.device_lock import DeviceExecutionLock as Lock, DeviceQueueIOError
from artemis.runtime.queue_files import queue_file_lock


def _hold_shared_reader(path, ready, finish):
    with queue_file_lock(Path(path)), Path(path).open(encoding="utf-8") as reader:
        ready.set()
        assert finish.wait(10)
        assert json.loads(reader.read()) == {"version": 0}


def test_other_process_reader_finishes_before_atomic_replace(tmp_path):
    path = tmp_path / "票据.wait"
    path.write_text('{"version":0}', encoding="utf-8")
    context = multiprocessing.get_context("spawn")
    ready, finish = context.Event(), context.Event()
    child = context.Process(target=_hold_shared_reader, args=(str(path), ready, finish))
    child.start()
    release = threading.Timer(0.2, finish.set)
    try:
        assert ready.wait(10)
        release.start()
        assert Lock._atomic_write_json(path, {"version": 1}, raise_on_error=True)
        assert json.loads(path.read_text()) == {"version": 1}
    finally:
        finish.set()
        child.join(10)
        if release.ident is not None:
            release.join()
    assert child.exitcode == 0


def test_parallel_queue_readers_never_see_partial_json_or_block_claim(tmp_path):
    path = tmp_path / "ticket.wait"
    payload = Lock._owner_payload("token", "device", "queue stress")
    assert Lock._atomic_write_json(path, payload)
    stop = threading.Event()
    errors = []

    def scan():
        while not stop.is_set():
            owner = Lock._read_owner(path)
            if owner is None or owner.token != "token":
                errors.append(owner)

    readers = [threading.Thread(target=scan) for _ in range(3)]
    for reader in readers:
        reader.start()
    try:
        for index in range(200):
            assert Lock._atomic_write_json(path, {**payload, "description": str(index)})
    finally:
        stop.set()
        for reader in readers:
            reader.join(5)
    assert not errors
    assert not list(tmp_path.glob("*.tmp"))


@pytest.mark.skipif(os.name != "nt", reason="Windows non-sharing reader behavior")
def test_legacy_reader_released_after_40ms_is_retried(tmp_path):
    path = tmp_path / "ticket.wait"
    path.write_text("{}")
    reader = path.open()
    release = threading.Timer(0.15, reader.close)
    release.start()
    try:
        assert Lock._atomic_write_json(path, {"claimed": True}, raise_on_error=True)
    finally:
        release.join()
        reader.close()


def _inject_replace(monkeypatch, operation):
    delays = []
    monkeypatch.setattr(module, "os", SimpleNamespace(**{**vars(os), "replace": operation}))
    monkeypatch.setattr(
        module, "time", SimpleNamespace(**{**vars(module.time), "sleep": delays.append})
    )
    return delays


def test_claim_keeps_native_error_and_bounds_sharing_retry(tmp_path, monkeypatch):
    monkeypatch.setattr(module, "get_temp_dir", lambda *_: tmp_path)
    token = Lock.reserve("test")
    lock = Lock("device", queue_ticket=token)
    original = lock._find_reserved_ticket().read_text()
    error = PermissionError(errno.EACCES, "held by another process")
    error.winerror = 5
    attempts = []

    def denied(src, dst):
        attempts.append(dst)
        raise error

    delays = _inject_replace(monkeypatch, denied)
    with pytest.raises(DeviceQueueIOError) as raised:
        lock.acquire()
    assert raised.value.__cause__ is error
    assert "winerror=5" in str(raised.value)
    assert "attempts=6" in str(raised.value)
    assert str(lock._find_reserved_ticket()) in str(raised.value)
    assert sum(delays) == pytest.approx(1.55)
    assert len(attempts) == 6
    assert lock._find_reserved_ticket().read_text() == original
    assert not list(lock.queue_dir.glob("*.tmp"))


@pytest.mark.parametrize("code", [errno.ENOSPC, errno.EROFS, errno.EACCES])
def test_unrelated_replace_errors_are_not_retried(tmp_path, monkeypatch, code):
    error = OSError(code, "disk or permission failure")

    def fail(*_):
        raise error

    delays = _inject_replace(monkeypatch, fail)
    with pytest.raises(DeviceQueueIOError) as raised:
        Lock._atomic_write_json(tmp_path / "ticket", {}, raise_on_error=True)
    assert raised.value.__cause__ is error
    assert not delays
    assert not list(tmp_path.glob("*.tmp"))


def test_temporary_write_error_is_distinguished(tmp_path):
    with pytest.raises(DeviceQueueIOError, match="write temporary") as raised:
        Lock._atomic_write_json(tmp_path / "missing" / "ticket", {}, raise_on_error=True)
    assert isinstance(raised.value.__cause__, FileNotFoundError)


def test_unreadable_owner_is_never_cleaned_as_stale(tmp_path, monkeypatch):
    path = tmp_path / "owner.lock"
    path.write_text("{}")
    os.utime(path, (1, 1))
    error = PermissionError(errno.EACCES, "locked reader")

    def denied(_):
        raise error

    monkeypatch.setattr(module, "read_queue_text", denied)
    with pytest.raises(DeviceQueueIOError, match="read"):
        Lock._get_active_owner_for_path(path)
    assert path.exists()


def test_cancellation_during_retry_does_not_resurrect_ticket(tmp_path, monkeypatch):
    path = tmp_path / "ticket"
    path.write_text("{}")
    error = PermissionError(errno.EACCES, "sharing")
    error.winerror = 32

    def cancel(src, dst):
        path.unlink()
        raise error

    _inject_replace(monkeypatch, cancel)
    assert not Lock._atomic_write_json(path, {"claimed": True}, require_existing=True)
    assert not path.exists()
    assert not list(tmp_path.glob("*.tmp"))