Merge pull request #254 from shakenfist/node-process-metrics

Support a nicer way of fetching node process metrics.

shakenfist_client/apiclient.py

@@ -844,6 +844,14 @@ def delete_node(self, node):
         r = self._request_url('DELETE', '/nodes/' + node)
         return r.json()
 
+    def get_node_process_metrics(self, node):
+        if not self.check_capability('node-process-metrics'):
+            raise IncapableException(
+                'The API server version you are talking to does not support '
+                'fetching process metrics for a node.')
+        r = self._request_url('GET', '/nodes/' + node + '/processmetrics')
+        return r.json()
+
     def get_interface(self, interface_uuid):
         r = self._request_url('GET', '/interfaces/' + interface_uuid)
         return r.json()

shakenfist_client/commandline/node.py

@@ -5,6 +5,7 @@
 import sys
 import time
 
+from shakenfist_client.apiclient import IncapableException
 from shakenfist_client import util
 
 
@@ -189,13 +190,25 @@ def node_cpuhogs(ctx, threshold=0.25):
     hogs = []
 
     for node in ctx.obj['CLIENT'].get_nodes():
-        event = ctx.obj['CLIENT'].get_node_events(node['name'], event_type='resources', limit=1)[0]
-        for resource in event.get('extra', {}):
-            value = event['extra'][resource]
-            if resource.startswith('process_cpu_fraction_') and value > threshold:
-                hogs.append('%s on node %s has consumed %.02f of a CPU, threshold is %.02f'
-                            % (resource[len('process_cpu_fraction_'):],
-                               node['name'], value, threshold))
+        try:
+            metrics = ctx.obj['CLIENT'].get_node_process_metrics(node['name'])
+            for resource in metrics:
+                value = metrics[resource]
+                if resource.startswith('process_cpu_fraction_') and value > threshold:
+                    hogs.append('%s on node %s has consumed %.02f of a CPU, threshold is %.02f'
+                                % (resource[len('process_cpu_fraction_'):],
+                                   node['name'], value, threshold))
+
+        except IncapableException:
+            # Fall back to the old way
+            event = ctx.obj['CLIENT'].get_node_events(node['name'], event_type='resources',
+                                                      limit=1)[0]
+            for resource in event.get('extra', {}):
+                value = event['extra'][resource]
+                if resource.startswith('process_cpu_fraction_') and value > threshold:
+                    hogs.append('%s on node %s has consumed %.02f of a CPU, threshold is %.02f'
+                                % (resource[len('process_cpu_fraction_'):],
+                                   node['name'], value, threshold))
 
     if hogs:
         print('\n'.join(hogs))