core services Health checks

ansible/roles/stack-monitor/defaults/main.yml

@@ -120,6 +120,18 @@ service_blackbox_checks:
     probe_module: http_2xx
     targets:
       - http://adminutil_adminutil:4000/health
+  - service_name: 'player_player'
+    probe_module: http_2xx
+    targets:
+      - http://player_player:3000/service/health
+  - service_name: 'learner-service'
+    probe_module: http_2xx
+    targets:
+      - http://learner-service:9000/service/health
+  - service_name: 'api-manager_kong'
+    probe_module: http_2xx
+    targets:
+      - http://api-manager_kong:8001/status
   - service_name: 'api-manager_echo'
     probe_module: http_2xx
     targets:
@@ -136,14 +148,10 @@ service_blackbox_checks:
     probe_module: http_2xx
     targets:
       - "{{sunbird_content_repo_api_base_url}}/health"
-  - service_name: 'player_player'
-    probe_module: http_2xx
-    targets:
-      - "http://player_player:3000"
   - service_name: 'content-service'
     probe_module: http_2xx
     targets:
-      - http://content-service:5000/health
+      - http://content-service:5000/service/health
   - service_name: 'monit'
     probe_module: http_2xx
     targets:

ansible/roles/stack-monitor/templates/alertrules.task.yml

@@ -18,7 +18,7 @@ groups:
       summary: 'HIGH MEMORY USAGE WARNING: TASK {% raw %}{{ $labels.container_label_com_docker_swarm_task_name }}{% endraw %} on {% raw %}{{ $labels.instance }}{% endraw %}'
 
   - alert: replicas_uneven
-    expr: (docker_service_replicas_expected != docker_service_replicas_running)
+    expr: sum by (service_name) (docker_service_replicas_expected != docker_service_replicas_running) 
     for: 1m
     annotations:
       description: 'UNEVEN REPLICAS COUNT FOR {% raw %}{{ $labels.service_name }}{% endraw %}'