Health and Ready endpoints (#3788)

feat: Adiciona health check endpoints

Co-authored-by: Edward <9326037+edwardoliveira@users.noreply.github.com>

Author: edwardoliveira

sapl/api/urls.py

@@ -1,15 +1,13 @@
-
 from django.conf.urls import include, url
 from drf_spectacular.views import SpectacularAPIView, SpectacularSwaggerView, \
     SpectacularRedocView
 from rest_framework.authtoken.views import obtain_auth_token
 
 from sapl.api.deprecated import SessaoPlenariaViewSet
-from sapl.api.views import AppVersionView, recria_token,\
-    SaplApiViewSetConstrutor
+from sapl.api.views import recria_token, SaplApiViewSetConstrutor
 
 from .apps import AppConfig
-
+from .views_health import HealthzView, ReadyzView
 
 app_name = AppConfig.name
 
@@ -38,7 +36,6 @@
     url(r'^api/', include(urlpatterns_api_doc)),
     url(r'^api/', include(urlpatterns_router)),
 
-    url(r'^api/version', AppVersionView.as_view()),
     url(r'^api/auth/token$', obtain_auth_token),
     url(r'^api/recriar-token/(?P<pk>\d*)$', recria_token, name="recria_token"),
 ]

sapl/api/views.py

@@ -1,6 +1,7 @@
 import logging
 
 from django.conf import settings
+from django.http import HttpResponse, JsonResponse
 from rest_framework.authtoken.models import Token
 from rest_framework.decorators import api_view, permission_classes
 from rest_framework.permissions import IsAuthenticated, IsAdminUser
@@ -21,20 +22,6 @@ def recria_token(request, pk):
     return Response({"message": "Token recriado com sucesso!", "token": token.key})
 
 
-class AppVersionView(APIView):
-    permission_classes = (IsAuthenticated,)
-
-    def get(self, request):
-        content = {
-            'name': 'SAPL',
-            'description': 'Sistema de Apoio ao Processo Legislativo',
-            'version': settings.SAPL_VERSION,
-            'user': request.user.username,
-            'is_authenticated': request.user.is_authenticated,
-        }
-        return Response(content)
-
-
 SaplApiViewSetConstrutor = ApiViewSetConstrutor
 SaplApiViewSetConstrutor.import_modules([
     'sapl.api.views_audiencia',

sapl/api/views_health.py

@@ -0,0 +1,106 @@
+import logging
+import traceback
+
+from django.http import HttpResponse, JsonResponse
+from django.utils import timezone
+from rest_framework.views import APIView
+from rest_framework.response import Response
+from rest_framework import status
+from django.conf import settings
+from sapl.health import check_app, check_db, check_cache
+
+COMMON_HEADERS = {
+    "Cache-Control": "no-store, no-cache, must-revalidate, max-age=0",
+    "Pragma": "no-cache",
+}
+
+
+def _format_plain(ok: bool) -> HttpResponse:
+    return HttpResponse("OK\n" if ok else "UNHEALTHY\n",
+                        status=status.HTTP_200_OK if ok else status.HTTP_503_SERVICE_UNAVAILABLE,
+                        content_type="text/plain")
+
+
+class HealthzView(APIView):
+    authentication_classes = []
+    permission_classes = []
+
+    logger = logging.getLogger(__name__)
+
+    def get(self, request):
+        try:
+            ok, msg, ms = check_app()
+            payload = {
+                "status": "OK" if ok else "UNHEALTHY",
+                "checks": {"app": {"ok": ok, "latency_ms": round(ms, 1), "error": msg}},
+                "version": settings.SAPL_VERSION,
+                "time": timezone.now().isoformat(),
+            }
+            if request.query_params.get("fmt") == "txt":
+                return _format_plain(ok)
+            return Response(payload,
+                            status=status.HTTP_200_OK if ok else status.HTTP_503_SERVICE_UNAVAILABLE,
+                            headers=COMMON_HEADERS)
+        except Exception as e:
+            self.logger.error(traceback.format_exc())
+            return "An internal error has occurred!"
+
+
+
+class ReadyzView(APIView):
+    authentication_classes = []
+    permission_classes = []
+
+    logger = logging.getLogger(__name__)
+
+    def get(self, request):
+        try:
+            checks = {
+                "app": check_app(),
+                "db": check_db(),
+                "cache": check_cache(),
+            }
+            payload_checks = {
+                name: {"ok": r[0], "latency_ms": round(r[2], 1), "error": r[1]}
+                for name, r in checks.items()
+            }
+            ok = all(r[0] for r in checks.values())
+            payload = {
+                "status": "ok" if ok else "unhealthy",
+                "checks": payload_checks,
+                "version": settings.SAPL_VERSION,
+                "time": timezone.now().isoformat(),
+            }
+            if request.query_params.get("fmt") == "txt":
+                return _format_plain(ok)
+            return Response(payload,
+                            status=status.HTTP_200_OK if ok else status.HTTP_503_SERVICE_UNAVAILABLE,
+                            headers=COMMON_HEADERS)
+        except Exception as e:
+            self.logger.error(traceback.format_exc())
+            return "An internal error has occurred!"
+
+
+class AppzVersionView(APIView):
+    authentication_classes = []
+    permission_classes = []
+
+    logger = logging.getLogger(__name__)
+
+    def get(self, request):
+        try:
+            payload = {
+                'name': 'SAPL',
+                'description': 'Sistema de Apoio ao Processo Legislativo',
+                'version': settings.SAPL_VERSION,
+            }
+            if request.query_params.get("fmt") == "txt":
+                return HttpResponse(f"{payload['version']} {payload['name']}",
+                                    status=status.HTTP_200_OK,
+                                    content_type="text/plain")
+            return Response(payload,
+                            status=status.HTTP_200_OK,
+                            headers=COMMON_HEADERS)
+        except Exception as e:
+            self.logger.error(traceback.format_exc())
+            return "An internal error has occurred!"

sapl/endpoint_restriction_middleware.py

@@ -18,7 +18,7 @@
     'ff00::/8'
 ]
 
-RESTRICTED_ENDPOINTS = ['/metrics']
+RESTRICTED_ENDPOINTS = ['/metrics', '/health', '/ready', '/version']
 
 
 class EndpointRestrictionMiddleware:

sapl/health.py

@@ -0,0 +1,36 @@
+# core/health.py
+import logging
+import time
+from typing import Tuple, Optional
+from django.db import connection
+from django.core.cache import cache
+
+logger = logging.getLogger(__name__)
+
+
+def check_app() -> Tuple[bool, Optional[str], float]:
+    t0 = time.monotonic()
+    return True, None, (time.monotonic() - t0) * 1000
+
+
+def check_db() -> Tuple[bool, Optional[str], float]:
+    t0 = time.monotonic()
+    try:
+        with connection.cursor() as cur:
+            cur.execute("SELECT 1")
+            cur.fetchone()
+        return True, None, (time.monotonic() - t0) * 1000
+    except Exception as e:
+        logging.error(e)
+        return False, "An internal error has occurred!", (time.monotonic() - t0) * 1000
+
+
+def check_cache() -> Tuple[bool, Optional[str], float]:
+    t0 = time.monotonic()
+    try:
+        cache.set("_hc", "1", 5)
+        ok = cache.get("_hc") == "1"
+        return ok, None if ok else "Cache get/set failed", (time.monotonic() - t0) * 1000
+    except Exception as e:
+        logging.error(e)
+        return False, "An internal error has occurred!", (time.monotonic() - t0) * 1000

sapl/metrics.py

@@ -0,0 +1,12 @@
+from prometheus_client import CollectorRegistry, Gauge, generate_latest, CONTENT_TYPE_LATEST
+
+
+def health_registry(check_results: dict) -> bytes:
+    """
+    check_results: {"app": True/False, "db": True/False, ...}
+    """
+    reg = CollectorRegistry()
+    g = Gauge("app_health", "1 if healthy, 0 if unhealthy", ["component"], registry=reg)
+    for comp, ok in check_results.items():
+        g.labels(component=comp).set(1 if ok else 0)
+    return generate_latest(reg)

sapl/settings.py

@@ -149,9 +149,9 @@
     'django.middleware.clickjacking.XFrameOptionsMiddleware',
     'django.middleware.security.SecurityMiddleware',
     'whitenoise.middleware.WhiteNoiseMiddleware',
-    'django_prometheus.middleware.PrometheusAfterMiddleware',
     'waffle.middleware.WaffleMiddleware',
     'sapl.middleware.CheckWeakPasswordMiddleware',
+    'django_prometheus.middleware.PrometheusAfterMiddleware',
 ]
 if DEBUG:
     INSTALLED_APPS += ('debug_toolbar',)

sapl/urls.py

@@ -36,6 +36,8 @@
 import sapl.relatorios.urls
 import sapl.sessao.urls
 
+from sapl.api.views_health import AppzVersionView, HealthzView, ReadyzView
+
 urlpatterns = []
 
 urlpatterns += [
@@ -69,6 +71,12 @@
     path("robots.txt", TemplateView.as_view(
         template_name="robots.txt", content_type="text/plain")),
 
+    # Health and Readiness
+    url(r'^version/$', AppzVersionView.as_view(), name="version"),
+    url(r"^health/$", HealthzView.as_view(), name="health"),
+    url(r"^ready/$", ReadyzView.as_view(), name="ready"),
+
+    # Monitoring
     path(r'', include('django_prometheus.urls')),
 
 ]