arm-core/api/tests/test_nats_integration.py

"""Интеграционные тесты `src/core/nats.py` против NATS без JetStream из `.env.test` (spec recognition-job-exchange,
«Соединение с NATS по шаблону, без JetStream»; design D-6).

Subject'ы — уникальные на тест (`test.<uuid>`): общий `recognition.result` не трогаем, чтобы не пересекаться
с другими запущенными инстансами ядра.
"""

import asyncio
import time
import uuid
from collections import Counter
from typing import Any

import pytest
from nats.aio.client import Client as NATS
from nats.aio.msg import Msg

from src.contract.subjects import NATS_MSG_ID_HEADER
from src.core.config import NATSSettings, get_settings
from src.core.nats import RESULT_QUEUE_GROUP, NatsConnection, NatsPublishError

pytestmark = pytest.mark.integration

READY_TIMEOUT = 5.0
START_MAX_SECONDS = 0.5  # start() не ждёт подключения
STOP_MAX_SECONDS = 2.0  # stop() не висит в цикле подключения nats-py


def _nats_settings() -> NATSSettings:
    return get_settings().OPENBAO.NATS


def _unique_subject() -> str:
    return f"test.{uuid.uuid4().hex}"


async def _wait_until(predicate: Any, timeout: float = 5.0) -> None:
    deadline = time.monotonic() + timeout
    while not predicate():
        if time.monotonic() > deadline:
            raise AssertionError("условие не выполнено за отведённое время")
        await asyncio.sleep(0.02)


async def _start(conn: NatsConnection) -> NatsConnection:
    await conn.start()
    if not await conn.wait_ready(timeout=READY_TIMEOUT):
        await conn.stop()
        pytest.fail(f"NATS недоступен ({_nats_settings().NATS_URL}, .env.test)", pytrace=False)
    return conn


async def test_publish_delivers_payload_and_msg_id_header():
    subject = _unique_subject()
    received: list[Msg] = []

    async def handler(msg: Msg) -> None:
        received.append(msg)

    async def subscribe(conn: NatsConnection) -> None:
        await conn.subscribe_results(handler)

    conn = await _start(NatsConnection(_nats_settings(), on_connected=subscribe, result_subject=subject))
    try:
        assert conn.is_connected
        await conn.publish(subject, b'{"job": 1}', "job-1:1")
        await _wait_until(lambda: len(received) == 1)
    finally:
        await conn.stop()

    msg = received[0]
    assert msg.subject == subject
    assert msg.data == b'{"job": 1}'
    assert msg.headers == {NATS_MSG_ID_HEADER: "job-1:1"}


async def test_queue_group_delivers_each_message_to_exactly_one_instance():
    """Два инстанса ядра в queue group `arm-core`: каждое сообщение обработано ровно одним (без дублей)."""
    subject = _unique_subject()
    received: dict[str, list[str]] = {"a": [], "b": []}

    def make_on_connected(name: str) -> Any:
        async def handler(msg: Msg) -> None:
            assert msg.headers is not None
            received[name].append(msg.headers[NATS_MSG_ID_HEADER])

        async def on_connected(conn: NatsConnection) -> None:
            await conn.subscribe_results(handler)

        return on_connected

    instance_a = await _start(
        NatsConnection(_nats_settings(), on_connected=make_on_connected("a"), result_subject=subject)
    )
    instance_b = await _start(
        NatsConnection(_nats_settings(), on_connected=make_on_connected("b"), result_subject=subject)
    )
    assert instance_a.queue == instance_b.queue == RESULT_QUEUE_GROUP
    total = 20
    try:
        for i in range(total):
            await instance_a.publish(subject, f"result {i}".encode(), f"result:job-{i}:1")
        await _wait_until(lambda: len(received["a"]) + len(received["b"]) >= total)
        await asyncio.sleep(0.2)  # дубль, если бы он был, успел бы прийти
    finally:
        await instance_a.stop()
        await instance_b.stop()

    all_ids = received["a"] + received["b"]
    assert len(all_ids) == total
    assert Counter(all_ids) == Counter(f"result:job-{i}:1" for i in range(total))


async def test_subscription_survives_handler_error():
    subject = _unique_subject()
    received: list[bytes] = []

    async def handler(msg: Msg) -> None:
        if msg.data == b"boom":
            raise ValueError("сломанное сообщение")
        received.append(msg.data)

    async def subscribe(conn: NatsConnection) -> None:
        await conn.subscribe_results(handler)

    conn = await _start(NatsConnection(_nats_settings(), on_connected=subscribe, result_subject=subject))
    try:
        await conn.publish(subject, b"boom", "m:1")
        await conn.publish(subject, b"ok", "m:2")
        await _wait_until(lambda: received == [b"ok"])
    finally:
        await conn.stop()


async def test_start_against_unreachable_server_does_not_block_and_stops_cleanly():
    """Регрессия T4.2: недоступный NATS при старте — start() не ждёт, is_connected=False, stop() быстрый."""
    settings = NATSSettings(NATS_URL="nats://127.0.0.1:54999")
    conn = NatsConnection(settings)

    started = time.monotonic()
    await conn.start()
    assert time.monotonic() - started < START_MAX_SECONDS
    await asyncio.sleep(0.3)
    assert not conn.is_connected
    with pytest.raises(NatsPublishError):
        await conn.publish("test.unreachable", b"{}", "m:1")

    started = time.monotonic()
    await conn.stop()
    assert time.monotonic() - started < STOP_MAX_SECONDS
    assert not conn.is_connected


async def test_plain_client_receives_publish_from_connection():
    """Сторонний подписчик (как сервис распознавания) видит заголовок `Nats-Msg-Id` от `NatsConnection.publish`."""
    subject = _unique_subject()
    nc = NATS()
    await nc.connect(_nats_settings().NATS_URL, connect_timeout=2, max_reconnect_attempts=1)
    subscription = await nc.subscribe(subject)
    conn = await _start(NatsConnection(_nats_settings()))
    try:
        await conn.publish(subject, b"payload", "job-x:2")
        msg = await subscription.next_msg(timeout=5)
    finally:
        await conn.stop()
        await nc.close()

    assert msg.data == b"payload"
    assert msg.headers == {NATS_MSG_ID_HEADER: "job-x:2"}