L'API exposait /metrics, mais aucun collecteur ne le lisait : monitoring/ ne contenait que des .gitkeep. Sous le profil Compose `monitoring` : prometheus, alertmanager, grafana, postgres-exporter, node-exporter et cadvisor. Tous ont un mem_limit, pour environ 700 Mo au total sur la VM de 8 Go, et leurs interfaces n'écoutent que sur 127.0.0.1. Le profil est actif en prod via COMPOSE_PROFILES, donc à chaque déploiement, et se lance à la demande ailleurs (make monitoring-up). - Neuf règles d'alerte (API, base, hôte, cibles). Chacune a un cas dans les tests joués par `promtool test rules`, en CI comme par make monitoring-check. - Alertmanager route les alertes par courriel vers Mailpit ; un critical masque le warning de la même cible. - Grafana est provisionné : sources Prometheus et TimescaleDB, et trois tableaux de bord (API, données et dérive du modèle, infrastructure). - Le rôle PostgreSQL `supervision` est en lecture seule sur les seules tables métier (db/roles/supervision.sql), posé par make db-ensure-supervision et par stack-up quand le profil est actif. - Le jeton de /metrics passe à Prometheus en secret Compose (APP_METRICS_TOKEN) ; provision-host.sh génère ce secret et les deux autres. Backend : - un APP_METRICS_TOKEN vide vaut absent ; - les sondes de santé ne comptent plus dans les métriques ; - seaux de latence fins autour de 500 ms ; - un registre Prometheus par application, sans quoi toute application créée après la première (dans les tests) ne mesurait rien. Réf : #26
126 lines
4.3 KiB
Python
126 lines
4.3 KiB
Python
import pytest
|
|
from httpx import ASGITransport, AsyncClient
|
|
from httpx import Response as HttpResponse
|
|
|
|
from app.main import create_app
|
|
from tests.factories import make_settings
|
|
|
|
ORIGINE = "https://enervision.fr"
|
|
|
|
|
|
async def interroge(
|
|
settings_overrides: dict[str, object], chemin: str, **kwargs: object
|
|
) -> HttpResponse:
|
|
application = create_app(make_settings(**settings_overrides))
|
|
transport = ASGITransport(app=application)
|
|
async with AsyncClient(transport=transport, base_url="http://test") as client:
|
|
return await client.get(chemin, **kwargs) # type: ignore[arg-type]
|
|
|
|
|
|
@pytest.mark.parametrize(
|
|
("entete", "valeur"),
|
|
[
|
|
("x-content-type-options", "nosniff"),
|
|
("x-frame-options", "DENY"),
|
|
("referrer-policy", "no-referrer"),
|
|
],
|
|
ids=["nosniff", "anti_iframe", "referrer"],
|
|
)
|
|
async def test_every_response_carries_the_security_headers(
|
|
client: AsyncClient, entete: str, valeur: str
|
|
) -> None:
|
|
response = await client.get("/api/v1/health/live")
|
|
|
|
assert response.headers[entete] == valeur
|
|
|
|
|
|
async def test_the_application_never_sets_hsts_itself(client: AsyncClient) -> None:
|
|
response = await client.get("/api/v1/health/live")
|
|
|
|
assert "strict-transport-security" not in response.headers
|
|
|
|
|
|
@pytest.mark.parametrize(
|
|
"env",
|
|
["staging", "prod"],
|
|
ids=["preproduction", "production"],
|
|
)
|
|
async def test_the_documentation_disappears_outside_development(env: str) -> None:
|
|
surcharges = {"env": env, "cors_origins": ORIGINE}
|
|
|
|
for chemin in ("/docs", "/openapi.json"):
|
|
assert (await interroge(surcharges, chemin)).status_code == 404
|
|
|
|
|
|
@pytest.mark.parametrize("env", ["local", "dev"], ids=["local", "developpement"])
|
|
async def test_the_documentation_stays_available_while_developing(env: str) -> None:
|
|
surcharges = {"env": env, "cors_origins": ORIGINE}
|
|
|
|
assert (await interroge(surcharges, "/openapi.json")).status_code == 200
|
|
|
|
|
|
async def test_an_explicit_override_can_reopen_the_documentation() -> None:
|
|
surcharges = {"env": "prod", "cors_origins": ORIGINE, "expose_api_docs": True}
|
|
|
|
assert (await interroge(surcharges, "/openapi.json")).status_code == 200
|
|
|
|
|
|
async def test_metrics_stay_open_when_no_token_is_configured(client: AsyncClient) -> None:
|
|
response = await client.get("/metrics")
|
|
|
|
assert response.status_code == 200
|
|
|
|
|
|
async def test_an_empty_metrics_token_means_no_token() -> None:
|
|
assert (await interroge({"metrics_token": ""}, "/metrics")).status_code == 200
|
|
|
|
|
|
async def test_metrics_ignore_health_probes_but_count_business_routes(client: AsyncClient) -> None:
|
|
await client.get("/api/v1/health/live")
|
|
await client.get("/api/v1/sites")
|
|
|
|
exposition = (await client.get("/metrics")).text
|
|
|
|
assert 'handler="/api/v1/health/live"' not in exposition
|
|
assert 'handler="/api/v1/sites"' in exposition
|
|
|
|
|
|
async def test_metrics_demand_the_token_once_one_is_configured() -> None:
|
|
surcharges = {"metrics_token": "un-jeton-de-supervision-assez-long"}
|
|
|
|
assert (await interroge(surcharges, "/metrics")).status_code == 401
|
|
|
|
|
|
async def test_metrics_answer_to_the_right_token() -> None:
|
|
surcharges = {"metrics_token": "un-jeton-de-supervision-assez-long"}
|
|
entetes = {"Authorization": "Bearer un-jeton-de-supervision-assez-long"}
|
|
|
|
response = await interroge(surcharges, "/metrics", headers=entetes)
|
|
|
|
assert response.status_code == 200
|
|
|
|
|
|
async def test_metrics_refuse_a_token_that_is_almost_right() -> None:
|
|
surcharges = {"metrics_token": "un-jeton-de-supervision-assez-long"}
|
|
entetes = {"Authorization": "Bearer un-jeton-de-supervision-assez-lon"}
|
|
|
|
response = await interroge(surcharges, "/metrics", headers=entetes)
|
|
|
|
assert response.status_code == 401
|
|
|
|
|
|
async def test_an_unhandled_error_returns_a_correlation_id_and_no_traceback() -> None:
|
|
application = create_app(make_settings())
|
|
|
|
@application.get("/api/v1/essai-panne")
|
|
async def _casse() -> None:
|
|
raise RuntimeError("secret interne de la pile")
|
|
|
|
transport = ASGITransport(app=application, raise_app_exceptions=False)
|
|
async with AsyncClient(transport=transport, base_url="http://test") as client:
|
|
response = await client.get("/api/v1/essai-panne")
|
|
|
|
assert response.status_code == 500
|
|
assert "secret interne de la pile" not in response.text
|
|
assert response.json()["correlation"]
|