Merge pull request #7145 from luis5tb/kuryr-health-checks

Add support for kuryr-controller health checks

roles/kuryr/defaults/main.yaml

@@ -14,6 +14,9 @@ kuryr_openstack_project_domain_name: default
 # Kuryr OpenShift namespace
 kuryr_namespace: openshift-infra
 
+# Kuryr health check server port
+kuryr_healthcheck_port: 8082
+
 # Whether to run the cni plugin in debug mode
 kuryr_cni_debug: "false"
 

roles/kuryr/templates/cni-daemonset.yaml.j2

@@ -49,6 +49,18 @@ spec:
           mountPath: /host_proc
         - name: openvswitch
           mountPath: /var/run/openvswitch
+        readinessProbe:
+          httpGet:
+            path: /ready
+            port: {{ kuryr_healthcheck_port }}
+            scheme: HTTP
+          initialDelaySeconds: 15
+          timeoutSeconds: 5
+        livenessProbe:
+          httpGet:
+            path: /alive
+            port: {{ kuryr_healthcheck_port }}
+          initialDelaySeconds: 15
       volumes:
         - name: bin
           hostPath:

roles/kuryr/templates/configmap.yaml.j2

@@ -332,6 +332,9 @@ data:
     # Pod VIF drivers vs Pool Drivers mapping allowed
     pools_vif_drivers = nested:nested-vlan,neutron:neutron-vif
 
+    [health_server]
+    port = {{ kuryr_healthcheck_port }}
+
   kuryr-cni.conf: |+
     [DEFAULT]
 
@@ -656,3 +659,9 @@ data:
     # From kuryr_kubernetes
     #
     lock_path = {{ kuryr_openstack_lock_path | default('/var/kuryr-lock') }}
+
+    [cni_health_server]
+    port = {{ kuryr_healthcheck_port }}
+
+    # Maximum memory usage (MiB) before kuryr-daemon is marked as unhealthy
+    max_memory_usage = -1

roles/kuryr/templates/controller-deployment.yaml.j2

@@ -22,13 +22,17 @@ spec:
       - image: {{ openshift_openstack_kuryr_controller_image }}
         imagePullPolicy: IfNotPresent
         name: controller
-{% if kuryr_openstack_enable_pools | default(false) %}
         readinessProbe:
-          exec:
-            command:
-            - cat
-            - /tmp/pools_loaded
-{% endif %}
+          httpGet:
+            path: /ready
+            port: {{ kuryr_healthcheck_port }}
+            scheme: HTTP
+          timeoutSeconds: 5
+        livenessProbe:
+          httpGet:
+            path: /alive
+            port: {{ kuryr_healthcheck_port }}
+          initialDelaySeconds: 15
         terminationMessagePath: "/dev/termination-log"
         # FIXME(dulek): This shouldn't be required, but without it selinux is
         #               complaining about access to kuryr.conf.