feat(neutron): use oslo.healthcheck for readiness probe

The neutron-server readiness probe now targets the oslo.middleware
/healthcheck endpoint instead of the API root /. This routes the
probe through a uWSGI worker and enables graceful draining by creating
/var/lib/neutron/healthcheck_disable to remove a pod from Service
rotation without restarting it. The liveness probe is intentionally
left on the uWSGI stats endpoint (port 1717), which is served by the
master process, so a dependency outage cannot trigger a restart storm
across neutron-server pods.

Assisted-by: Claude Opus 4.8
Change-Id: I09a4e7c4b4f174c463797094e72acec8b706d0ee
Signed-off-by: Doug Goldstein <cardoe@cardoe.com>

Author: cardoe

neutron/templates/deployment-server.yaml

@@ -15,7 +15,7 @@ limitations under the License.
 {{- define "serverReadinessProbeTemplate" }}
 httpGet:
   scheme: HTTP
-  path: /
+  path: {{ tuple "network" "healthcheck" "internal" . | include "helm-toolkit.endpoints.keystone_endpoint_path_lookup" }}
   port: {{ tuple "network" "service" "api" . | include "helm-toolkit.endpoints.endpoint_port_lookup" }}
 {{- end }}
 {{- define "serverLivenessProbeTemplate" }}

neutron/values.yaml

@@ -1337,6 +1337,7 @@ conf:
     composite:neutron:
       use: egg:Paste#urlmap
       /: neutronversions_composite
+      /healthcheck: healthcheck
       /v2.0: neutronapi_v2_0
     composite:neutronapi_v2_0:
       use: call:neutron.auth:pipeline_factory
@@ -1368,6 +1369,11 @@ conf:
       paste.app_factory: neutron.pecan_wsgi.app:versions_factory
     app:neutronapiapp_v2_0:
       paste.app_factory: neutron.api.v2.router:APIRouter.factory
+    app:healthcheck:
+      paste.app_factory: oslo_middleware:Healthcheck.app_factory
+      backends: disable_by_file
+      # state_path is writable; drain a pod by touching this file.
+      disable_by_file_path: /var/lib/neutron/healthcheck_disable
     filter:osprofiler:
       paste.filter_factory: osprofiler.web:WsgiMiddleware.factory
   neutron_api_uwsgi:
@@ -2527,6 +2533,7 @@ endpoints:
       #     key: null
     path:
       default: null
+      healthcheck: /healthcheck
     scheme:
       default: 'http'
       service: 'http'

releasenotes/notes/neutron-8a3f1c2d5e7b0a94.yaml

@@ -0,0 +1,12 @@
+---
+neutron:
+  - |
+    The neutron-server readiness probe now targets the oslo.middleware
+    ``/healthcheck`` endpoint (backed by ``disable_by_file``) instead of the
+    API root ``/``. This routes the probe through a uWSGI worker and enables
+    graceful draining: create ``/var/lib/neutron/healthcheck_disable`` to
+    remove a pod from Service rotation without restarting it. The liveness
+    probe is intentionally left on the uWSGI stats endpoint (port 1717), which
+    is served by the master process, so a dependency outage cannot trigger a
+    restart storm across neutron-server pods.
+...