Merge remote-tracking branch 'origin/dev' into feat/reconciliation-dag

apps/backend/app/core/config.py

@@ -1,7 +1,7 @@
 from functools import lru_cache
 from typing import Literal, Self
 
-from pydantic import Field, SecretStr, model_validator
+from pydantic import Field, SecretStr, field_validator, model_validator
 from pydantic_settings import BaseSettings, SettingsConfigDict
 
 Environment = Literal["local", "dev", "staging", "prod"]
@@ -76,6 +76,13 @@ class Settings(BaseSettings):
     expose_api_docs: bool | None = None
     metrics_token: SecretStr | None = None
 
+    # Compose passe `APP_METRICS_TOKEN` vide quand aucun jeton n'est posé : vide vaut absent, sinon
+    # `/metrics` exigerait un `Bearer` sans valeur et plus rien ne pourrait le scruter.
+    @field_validator("metrics_token", mode="before")
+    @classmethod
+    def _jeton_vide_vaut_absent(cls, valeur: object) -> object:
+        return None if valeur == "" else valeur
+
     @property
     def allowed_origins(self) -> list[str]:
         return [origin.strip() for origin in self.cors_origins.split(",") if origin.strip()]

apps/backend/app/main.py

@@ -6,7 +6,8 @@ from fastapi import Depends, FastAPI
 from fastapi.middleware.cors import CORSMiddleware
 from fastapi.openapi.docs import get_redoc_html, get_swagger_ui_html
 from fastapi.staticfiles import StaticFiles
-from prometheus_fastapi_instrumentator import Instrumentator
+from prometheus_client import CollectorRegistry, GCCollector, PlatformCollector, ProcessCollector
+from prometheus_fastapi_instrumentator import Instrumentator, metrics
 from starlette.requests import Request
 from starlette.responses import HTMLResponse
 
@@ -37,6 +38,16 @@ async def lifespan(_: FastAPI) -> AsyncIterator[None]:
     await get_engine().dispose()
 
 
+# Pourquoi : le registre global n'accepte chaque métrique qu'une fois. Toute application créée
+# après la première, dans les tests notamment, n'aurait rien mesuré.
+def _registre_de_metriques() -> CollectorRegistry:
+    registre = CollectorRegistry()
+    ProcessCollector(registry=registre)
+    PlatformCollector(registry=registre)
+    GCCollector(registry=registre)
+    return registre
+
+
 def create_app(settings: Settings | None = None) -> FastAPI:
     resolved = settings or get_settings()
     configure_logging(resolved)
@@ -102,7 +113,14 @@ def create_app(settings: Settings | None = None) -> FastAPI:
 
     register_error_handlers(application)
 
-    Instrumentator().instrument(application).expose(
+    # Les sondes de santé tombent toutes les 30 s : comptées, elles fausseraient latences et débit.
+    # Seaux fins autour du seuil de charge (p95 < 500 ms, ADR 0015), route par route.
+    registre = _registre_de_metriques()
+    Instrumentator(
+        excluded_handlers=["/metrics", f"{resolved.api_prefix}/health/.*"], registry=registre
+    ).add(
+        metrics.default(latency_lowr_buckets=(0.05, 0.1, 0.25, 0.5, 1, 2.5), registry=registre)
+    ).instrument(application).expose(
         application,
         endpoint="/metrics",
         include_in_schema=False,

apps/backend/tests/api/test_hardening.py

@@ -72,6 +72,20 @@ async def test_metrics_stay_open_when_no_token_is_configured(client: AsyncClient
     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"}