Add Zookeeper metrics proxy

api-daemon/pvcapid/flaskapi.py

@@ -643,12 +643,13 @@ class API_Metrics(Resource):
         health_output, health_retcode = api_helper.cluster_health_metrics()
         resource_output, resource_retcode = api_helper.cluster_resource_metrics()
         ceph_output, ceph_retcode = api_helper.ceph_metrics()
+        zookeeper_output, zookeeper_retcode = api_helper.zookeeper_metrics()
 
         if health_retcode != 200 or resource_retcode != 200 or ceph_retcode != 200:
             output = "Error: Failed to obtain data"
             retcode = 400
         else:
-            output = health_output + resource_output + ceph_output
+            output = health_output + resource_output + ceph_output + zookeeper_output
             retcode = 200
 
         response = flask.make_response(output, retcode)
@@ -761,6 +762,40 @@ class API_Metrics_Ceph(Resource):
 api.add_resource(API_Metrics_Ceph, "/metrics/ceph")
 
 
+# /metrics/zookeeper
+class API_Metrics_Zookeeper(Resource):
+    def get(self):
+        """
+        Return the current PVC Zookeeper Prometheus metrics
+
+        Proxies a metrics request to the current primary node, since all coordinators
+        run an active Zookeeper instance and we want one central location.
+        ---
+        tags:
+          - root
+        responses:
+          200:
+            description: OK
+          400:
+            description: Bad request
+        """
+        zookeeper_output, zookeeper_retcode = api_helper.zookeeper_metrics()
+
+        if zookeeper_retcode != 200:
+            output = "Error: Failed to obtain data"
+            retcode = 400
+        else:
+            output = zookeeper_output
+            retcode = 200
+
+        response = flask.make_response(output, retcode)
+        response.mimetype = "text/plain"
+        return response
+
+
+api.add_resource(API_Metrics_Zookeeper, "/metrics/zookeeper")
+
+
 # /faults
 class API_Faults(Resource):
     @RequestParser(

api-daemon/pvcapid/helper.py

@@ -199,6 +199,38 @@ def ceph_metrics(zkhandler):
     return output, status_code
 
 
+@pvc_common.Profiler(config)
+@ZKConnection(config)
+def zookeeper_metrics(zkhandler):
+    """
+    Obtain current Zookeeper Prometheus metrics from the active coordinator node
+    """
+    primary_node = zkhandler.read("base.config.primary_node")
+    if primary_node is not None:
+        # Get the data from the endpoint
+        # We use the default port of 9141
+        zookeeper_prometheus_uri = f"http://{primary_node}:9141/metrics"
+        response = get(zookeeper_prometheus_uri)
+
+        if response.status_code == 200:
+            output = response.text
+            # Parse the text to remove annoying ports (":2181")
+            output = output.replace(":2181", "")
+            # Sort the output text
+            output_lines = output.split("\n")
+            output_lines.sort()
+            output = "\n".join(output_lines) + "\n"
+            status_code = 200
+        else:
+            output = f"Error: Failed to obtain metric data from {primary_node} primary node daemon\n"
+            status_code = 400
+    else:
+        output = "Error: Failed to find an active primary node\n"
+        status_code = 400
+
+    return output, status_code
+
+
 #
 # Fault functions
 #