"""Integration: Steward learns each metric's normal per hour (#5739).

The operator's case (System 182): a nightly backup that runs hot at 02:00
must read as normal at 02:00, and the same level at 14:00 must not, so the
normal has to be per hour and built from what each hour held, not its peak.
Requires STEWARD_DATABASE_URL.
"""
from __future__ import annotations

import asyncio
import os
import uuid
from datetime import datetime, timedelta, timezone

import pytest

pytestmark = pytest.mark.integration


@pytest.fixture
def app():
    if not os.environ.get("STEWARD_DATABASE_URL"):
        pytest.skip("needs Postgres")
    from steward.app import create_app
    return create_app(testing=False)


@pytest.fixture
def learned():
    from steward.core import norms
    norms.learn_normal("normtest", "cpu_pct", 5)
    yield
    norms.LEARNED.pop(("normtest", "cpu_pct"), None)


# Monday 00:00 UTC, so week slots are easy to read.
NOW = datetime(2026, 10, 12, 0, 0, tzinfo=timezone.utc)


def _hours(res: str, days: int, level):
    """An hourly rollup row for each of the last `days` days' hours, held at
    level(day, hour) -> (p95, max)."""
    from steward.models.metrics import PluginMetricHourly
    rows = []
    for d in range(1, days + 1):
        for h in range(24):
            bucket = NOW - timedelta(days=d) + timedelta(hours=h)
            p95, peak = level(bucket, h)
            rows.append(PluginMetricHourly(source_module="normtest", resource_name=res,
                                           metric_name="cpu_pct", bucket=bucket,
                                           value_avg=p95, value_max=peak, value_p95=p95,
                                           sample_count=60))
    return rows


def _learn(app, rows):
    from steward.core.norms import norms_at, rebuild_norms

    async def _go():
        async with app.db_sessionmaker() as s:
            async with s.begin():
                s.add_all(rows)
            async with s.begin():
                await rebuild_norms(s, NOW)
            res = rows[0].resource_name
            out = {}
            for at in (NOW.replace(hour=2), NOW.replace(hour=3), NOW.replace(hour=14),
                       NOW + timedelta(days=1, hours=2)):
                got = await norms_at(s, [("normtest", res, "cpu_pct")], at)
                out[at] = got.get(("normtest", res, "cpu_pct"))
            return out

    return asyncio.run(_go())


def test_a_nightly_backup_is_normal_at_its_hour_and_not_at_another(app, learned):
    res = f"night-{uuid.uuid4().hex[:8]}"
    # Every night 02:00–03:00 holds 90%; the rest of the day holds 10% and
    # spikes briefly to 95% (a high max, a low p95).
    got = _learn(app, _hours(res, 7, lambda b, h: (90.0, 95.0) if h == 2 else (10.0, 95.0)))
    at2, at3, at14 = (got[NOW.replace(hour=h)] for h in (2, 3, 14))
    assert at2.kind == "day" and at2.days_seen == 7
    assert at2.usual_hold == 90.0
    # 03:00 takes in 02:00 as a neighbour: a backup that runs late stays normal.
    assert at3.usual_hold == 90.0
    # 14:00 holds 10: the spikes raised its peak, never its hold.
    assert at14.usual_hold == 10.0 and at14.usual_peak == 95.0


def test_too_little_history_learns_no_normal(app, learned):
    res = f"young-{uuid.uuid4().hex[:8]}"
    got = _learn(app, _hours(res, 4, lambda b, h: (10.0, 10.0)))
    assert all(n is None for n in got.values())


def test_three_weeks_of_history_learns_the_hour_of_the_week(app, learned):
    res = f"weekly-{uuid.uuid4().hex[:8]}"
    # A weekly job: Tuesdays 02:00 hold 80; every other hour holds 5.
    got = _learn(app, _hours(res, 22, lambda b, h: (80.0, 80.0) if b.weekday() == 1 and h == 2
                                                    else (5.0, 5.0)))
    tuesday_2 = got[NOW + timedelta(days=1, hours=2)]
    monday_2 = got[NOW.replace(hour=2)]
    assert tuesday_2.kind == "week" and tuesday_2.slot == 26 and tuesday_2.usual_hold == 80.0
    # Monday 02:00 is far from Tuesday's job: on an hour-of-day normal it
    # would have been learned as the job's level.
    assert monday_2.usual_hold == 5.0