admin: monitor worker version (#12463)

* root: include version in celery ping

Signed-off-by: Jens Langhammer <jens@goauthentik.io>

* check version in worker endpoint

Signed-off-by: Jens Langhammer <jens@goauthentik.io>

* include worker version in prom metrics

Signed-off-by: Jens Langhammer <jens@goauthentik.io>

* format

Signed-off-by: Jens Langhammer <jens@goauthentik.io>

* fix tests

Signed-off-by: Jens Langhammer <jens@goauthentik.io>

---------

Signed-off-by: Jens Langhammer <jens@goauthentik.io>

authentik/admin/api/workers.py

@@ -1,12 +1,16 @@
 """authentik administration overview"""
 
+from socket import gethostname
+
 from django.conf import settings
 from drf_spectacular.utils import extend_schema, inline_serializer
-from rest_framework.fields import IntegerField
+from packaging.version import parse
+from rest_framework.fields import BooleanField, CharField
 from rest_framework.request import Request
 from rest_framework.response import Response
 from rest_framework.views import APIView
 
+from authentik import get_full_version
 from authentik.rbac.permissions import HasPermission
 from authentik.root.celery import CELERY_APP
 
@@ -16,11 +20,38 @@ class WorkerView(APIView):
 
     permission_classes = [HasPermission("authentik_rbac.view_system_info")]
 
-    @extend_schema(responses=inline_serializer("Workers", fields={"count": IntegerField()}))
+    @extend_schema(
+        responses=inline_serializer(
+            "Worker",
+            fields={
+                "worker_id": CharField(),
+                "version": CharField(),
+                "version_matching": BooleanField(),
+            },
+            many=True,
+        )
+    )
     def get(self, request: Request) -> Response:
         """Get currently connected worker count."""
-        count = len(CELERY_APP.control.ping(timeout=0.5))
+        raw: list[dict[str, dict]] = CELERY_APP.control.ping(timeout=0.5)
+        our_version = parse(get_full_version())
+        response = []
+        for worker in raw:
+            key = list(worker.keys())[0]
+            version = worker[key].get("version")
+            version_matching = False
+            if version:
+                version_matching = parse(version) == our_version
+            response.append(
+                {"worker_id": key, "version": version, "version_matching": version_matching}
+            )
         # In debug we run with `task_always_eager`, so tasks are ran on the main process
         if settings.DEBUG:  # pragma: no cover
-            count += 1
-        return Response({"count": count})
+            response.append(
+                {
+                    "worker_id": f"authentik-debug@{gethostname()}",
+                    "version": get_full_version(),
+                    "version_matching": True,
+                }
+            )
+        return Response(response)

authentik/admin/apps.py

@@ -1,11 +1,10 @@
 """authentik admin app config"""
 
-from prometheus_client import Gauge, Info
+from prometheus_client import Info
 
 from authentik.blueprints.apps import ManagedAppConfig
 
 PROM_INFO = Info("authentik_version", "Currently running authentik version")
-GAUGE_WORKERS = Gauge("authentik_admin_workers", "Currently connected workers")
 
 
 class AuthentikAdminConfig(ManagedAppConfig):

authentik/admin/signals.py

@@ -1,14 +1,35 @@
 """admin signals"""
 
 from django.dispatch import receiver
+from packaging.version import parse
+from prometheus_client import Gauge
 
-from authentik.admin.apps import GAUGE_WORKERS
+from authentik import get_full_version
 from authentik.root.celery import CELERY_APP
 from authentik.root.monitoring import monitoring_set
 
+GAUGE_WORKERS = Gauge(
+    "authentik_admin_workers",
+    "Currently connected workers, their versions and if they are the same version as authentik",
+    ["version", "version_matched"],
+)
+
+
+_version = parse(get_full_version())
+
 
 @receiver(monitoring_set)
 def monitoring_set_workers(sender, **kwargs):
     """Set worker gauge"""
-    count = len(CELERY_APP.control.ping(timeout=0.5))
-    GAUGE_WORKERS.set(count)
+    raw: list[dict[str, dict]] = CELERY_APP.control.ping(timeout=0.5)
+    worker_version_count = {}
+    for worker in raw:
+        key = list(worker.keys())[0]
+        version = worker[key].get("version")
+        version_matching = False
+        if version:
+            version_matching = parse(version) == _version
+        worker_version_count.setdefault(version, {"count": 0, "matching": version_matching})
+        worker_version_count[version]["count"] += 1
+    for version, stats in worker_version_count.items():
+        GAUGE_WORKERS.labels(version, stats["matching"]).set(stats["count"])

authentik/admin/tests/test_api.py

@@ -34,7 +34,7 @@ class TestAdminAPI(TestCase):
         response = self.client.get(reverse("authentik_api:admin_workers"))
         self.assertEqual(response.status_code, 200)
         body = loads(response.content)
-        self.assertEqual(body["count"], 0)
+        self.assertEqual(len(body), 0)
 
     def test_metrics(self):
         """Test metrics API"""