metalfrom.eu/apps/crawler/tests/test_health.py

"""
Contrôles de santé des services de fond.

Règle cardinale : une sonde ne doit JAMAIS interrompre le service qu'elle
surveille. Un contrôle de santé qui fait tomber le crawler serait pire que pas
de contrôle du tout — d'où le nombre de cas d'échec vérifiés ici.
"""

import contextlib

import bm_health as health
import pytest
from bm_health import (
    database_probe,
    freshness_probe,
    http_probe,
    maybe_run,
    probe,
    run_checks,
    write_health,
)


class FakeCursor:
    def __init__(self, results=None, fail=False):
        self._results = list(results or [])
        self.queries = []
        self.fail = fail

    def execute(self, sql, params=None):
        if self.fail:
            raise RuntimeError("base injoignable")
        self.queries.append((sql, params))

    def fetchone(self):
        return self._results.pop(0) if self._results else None

    def __enter__(self):
        return self

    def __exit__(self, *a):
        return False


def conn_factory(cursor):
    @contextlib.contextmanager
    def _get_conn():
        class Conn:
            def cursor(self_inner):
                return cursor
        yield Conn()
    return _get_conn


@pytest.fixture(autouse=True)
def _reset_interval():
    health._last_run = 0.0
    yield
    health._last_run = 0.0


class TestProbe:
    def test_transmet_le_resultat(self):
        assert probe("x", lambda: (True, "ok")) == ("x", True, "ok")

    def test_convertit_une_exception_en_echec(self):
        def boom():
            raise ValueError("cassé")
        name, ok, detail = probe("x", boom)
        assert (name, ok) == ("x", False)
        assert "ValueError" in detail and "cassé" in detail

    def test_normalise_un_detail_absent(self):
        assert probe("x", lambda: (True, None))[2] == ""


class TestDatabaseProbe:
    def test_ok_quand_la_base_repond(self):
        cur = FakeCursor([(1,)])
        assert database_probe(conn_factory(cur))()[0] is True
        # La sonde doit réellement interroger la base, pas se contenter
        # d'ouvrir une connexion.
        assert cur.queries and "SELECT 1" in cur.queries[0][0]

    def test_echoue_sans_lever_quand_la_base_est_injoignable(self):
        cur = FakeCursor(fail=True)
        name, ok, detail = probe("database", database_probe(conn_factory(cur)))
        assert ok is False
        assert "injoignable" in detail


class TestHttpProbe:
    def test_ok_sur_200(self):
        assert http_probe(lambda url, timeout: type("R", (), {"status_code": 200})(), "http://x")()[0] is True

    def test_echec_sur_500(self):
        ok, detail = http_probe(lambda url, timeout: type("R", (), {"status_code": 503})(), "http://x")()
        assert ok is False
        assert "503" in detail

    def test_une_erreur_reseau_devient_un_echec_de_sonde(self):
        def boom(url, timeout):
            raise ConnectionError("nom d'hôte introuvable")
        name, ok, detail = probe("dep", http_probe(boom, "http://x"))
        assert ok is False
        assert "ConnectionError" in detail


class TestFreshnessProbe:
    def _probe(self, latest, age_h):
        cur = FakeCursor([(latest,), (age_h,)])
        return freshness_probe(conn_factory(cur), "SELECT max(x) FROM t", 12, "test")()

    def test_ok_quand_les_donnees_sont_fraiches(self):
        assert self._probe("2026-08-18T10:00:00Z", 3.0)[0] is True

    def test_echec_quand_les_donnees_sont_trop_anciennes(self):
        ok, detail = self._probe("2026-08-01T10:00:00Z", 40.0)
        assert ok is False
        assert "40.0 h" in detail and "seuil 12 h" in detail

    def test_borne_exacte_acceptee(self):
        assert self._probe("x", 12.0)[0] is True

    # Une base vide au premier démarrage n'est pas une panne.
    def test_absence_de_donnees_nest_pas_un_echec(self):
        cur = FakeCursor([(None,)])
        ok, detail = freshness_probe(conn_factory(cur), "SELECT max(x) FROM t", 12, "test")()
        assert ok is True
        assert "aucune donnée" in detail


class TestWriteHealth:
    def test_agrege_les_sondes_et_enregistre(self):
        cur = FakeCursor()
        ok, failed = write_health(conn_factory(cur), "crawler",
                                  [("database", True, "ok"), ("dep", True, "ok")])
        assert ok is True and failed == []
        sql, params = cur.queries[0]
        assert "INSERT INTO service_health" in sql
        assert "ON CONFLICT (service) DO UPDATE" in sql   # une ligne par service
        assert params[0] == "crawler" and params[1] is True

    def test_une_sonde_en_echec_rend_le_service_degrade(self):
        cur = FakeCursor()
        ok, failed = write_health(conn_factory(cur), "crawler",
                                  [("database", True, "ok"), ("dep", False, "HTTP 503")])
        assert ok is False
        assert failed == ["dep — HTTP 503"]
        params = cur.queries[0][1]
        assert params[1] is False
        assert "dep" in params[3]          # la première panne est lisible telle quelle

    def test_le_detail_par_sonde_est_conserve(self):
        cur = FakeCursor()
        write_health(conn_factory(cur), "crawler", [("a", True, ""), ("b", False, "")])
        import json
        assert json.loads(cur.queries[0][1][2]) == {"a": True, "b": False}

    # Si la base est en panne, la sonde base l'a déjà signalé : échouer ici en
    # plus ferait tomber le service pour rien.
    def test_une_base_injoignable_ne_fait_pas_tomber_le_service(self):
        cur = FakeCursor(fail=True)
        ok, failed = write_health(conn_factory(cur), "crawler", [("database", False, "ko")])
        assert ok is False   # pas d'exception propagée


class TestRunChecks:
    def test_journalise_les_sondes_en_echec(self):
        cur = FakeCursor()
        messages = []
        run_checks(conn_factory(cur), "crawler",
                   [("dep", lambda: (False, "HTTP 503"))],
                   log_event=lambda lvl, msg: messages.append((lvl, msg)))
        assert messages and messages[0][0] == "warning"
        assert "DÉGRADÉ" in messages[0][1] and "dep" in messages[0][1]

    def test_ne_journalise_rien_quand_tout_va_bien(self):
        cur = FakeCursor()
        messages = []
        run_checks(conn_factory(cur), "crawler", [("dep", lambda: (True, "ok"))],
                   log_event=lambda lvl, msg: messages.append(msg))
        assert messages == []

    def test_un_log_event_defaillant_ne_fait_pas_tomber_le_controle(self):
        cur = FakeCursor()
        def boom(lvl, msg):
            raise RuntimeError("log cassé")
        assert run_checks(conn_factory(cur), "crawler",
                          [("dep", lambda: (False, "ko"))], log_event=boom) is False


class TestMaybeRun:
    def test_execute_au_premier_appel(self):
        cur = FakeCursor()
        assert maybe_run(conn_factory(cur), "crawler", [("d", lambda: (True, ""))]) is True

    # Appelée depuis une boucle qui tourne toutes les minutes, la fonction ne
    # doit pas sonder les dépendances externes à chaque passage.
    def test_ne_reexecute_pas_avant_lintervalle(self):
        cur = FakeCursor()
        probes = [("d", lambda: (True, ""))]
        maybe_run(conn_factory(cur), "crawler", probes, interval_s=3600)
        assert maybe_run(conn_factory(cur), "crawler", probes, interval_s=3600) is None

    def test_force_contourne_lintervalle(self):
        cur = FakeCursor()
        probes = [("d", lambda: (True, ""))]
        maybe_run(conn_factory(cur), "crawler", probes, interval_s=3600)
        assert maybe_run(conn_factory(cur), "crawler", probes, interval_s=3600, force=True) is True

    def test_reexecute_une_fois_lintervalle_ecoule(self):
        cur = FakeCursor()
        probes = [("d", lambda: (True, ""))]
        maybe_run(conn_factory(cur), "crawler", probes, interval_s=3600)
        health._last_run -= 4000          # simule le temps écoulé
        assert maybe_run(conn_factory(cur), "crawler", probes, interval_s=3600) is True