fix gpu request counts and use generic label

binderhub/resources.py

@@ -67,7 +67,7 @@ def get_gpu_availability(self, product=None, memory=None):
         elif memory:
             nodes = api.list_node(label_selector='gpu=true,nvidia.com/gpu.memory=%s' %memory)
         else: 
-            nodes = api.list_node(label_selector='gpu=true') 
+            nodes = api.list_node(label_selector='nvidia.com/gpu.product')
         for node in nodes.items:
             product = node.metadata.labels['nvidia.com/gpu.product']
             memory = int(node.metadata.labels['nvidia.com/gpu.memory'])
@@ -80,9 +80,10 @@ def get_gpu_availability(self, product=None, memory=None):
             gpu['total_requests'] = 0
             pods = api.list_pod_for_all_namespaces(field_selector='spec.nodeName=%s' %node.metadata.name).items
             for pod in pods:
-                requests = pod.spec.containers[0].resources.requests
-                if requests:
-                    gpu['total_requests'] += int(requests.get('nvidia.com/gpu', 0))
+                for container in pod.spec.containers:
+                    requests = container.resources.requests
+                    if requests:
+                        gpu['total_requests'] += int(requests.get('nvidia.com/gpu', 0))
             gpu['available'] = max(gpu['count'] - gpu['total_requests'], 0)
         return sorted(gpus.values(), key=lambda gpu : gpu['memory'])