Fix get_version() to match the latest run-time environment.

Closes issue google#76.

collector/collector_test.py

@@ -316,10 +316,8 @@ def test_version(self):
     self.assertTrue(result.get('success'))
     version = result.get('version')
     self.assertTrue(isinstance(version, types.StringTypes))
-    # TODO(eran): restore this test once the /version endpoint is fixed.
-    # self.assertEqual(
-    #    'kubernetes/cluster-insight ac933439ec5a 2015-03-28T17:23:41', version)
-    self.assertEqual('_unknown_', version)
+    self.assertEqual(
+       'kubernetes/cluster-insight ac933439ec5a 2015-03-28T17:23:41', version)
 
   def test_minions_status(self):
     """Test the '/minions_status' endpoint."""

collector/context.py

@@ -604,7 +604,16 @@ def _do_compute_graph(gs, input_queue, output_queue, output_format):
   assert utilities.valid_string(output_format)
 
   g = ContextGraph()
-  g.set_version(docker.get_version(gs))
+  try:
+    version = docker.get_version(gs)
+  except Exception as e:
+    exc_type, value, _ = sys.exc_info()
+    msg = ('get_version() failed with exception %s: %s' %
+           (exc_type, value))
+    gs.logger_error(msg)
+    version = '_unknown_'
+
+  g.set_version(version)
   g.set_relations_to_timestamps(gs.get_relations_to_timestamps())
 
   # Nodes

collector/docker.py

@@ -23,6 +23,7 @@
 """
 
 import json
+import os
 import re
 import sys
 import time
@@ -681,9 +682,6 @@ def get_version(gs):
     CollectorError: in case of any error to compute the running image
       information.
   """
-  return '_unknown_'
-  # TODO(eran): fix the error
-  # could not find an entry for \"cpu:/docker/...\" in /proc/self/cgroup
   version, timestamp_secs = gs.get_version_cache().lookup('')
   if timestamp_secs is not None:
     assert utilities.valid_string(version)
@@ -709,17 +707,41 @@ def get_version(gs):
     gs.logger_exception(msg)
     raise collector_error.CollectorError(msg)
 
-  # The file must contain an entry for 'cpu:/docker/...'.
-  m = re.search(r'\b[0-9]+:cpu:/docker/([0-9a-fA-F]+)\b', cgroup)
+  # The file must contain an entry for '\d+:cpu:/...'.
+  m = re.search(r'\b\d+:cpu:/([0-9a-fA-F]+)\b', cgroup)
   if not m:
     msg = 'could not find an entry for "cpu:/docker/..." in %s' % fname
     gs.logger_error(msg)
     raise collector_error.CollectorError(msg)
 
   hex_container_id = m.group(1)
+  if gs.get_testing():
+    # This pod name is guaranteed to match a pod in the testdata directory.
+    my_pod_name = 'kube-dns-bqw5e'
+  else:
+    my_pod_name = os.uname()[1]
+  assert utilities.valid_string(my_pod_name)
+
+  # Find my node name from my pod.
+  my_node_name = None
+  for pod in kubernetes.get_pods(gs):
+    assert utilities.is_wrapped_object(pod, 'Pod')
+    if pod['id'] == my_pod_name:
+      my_node_name = utilities.get_attribute(
+          pod, ['properties', 'spec', 'nodeName'])
+      break
+
+  if not utilities.valid_string(my_node_name):
+    msg = ('could not find pod %s or this pod does not contain a valid '
+           'node name' % my_pod_name)
+    gs.logger_error(msg)
+    raise collector_error.CollectorError(msg)
+
   # inspect the running container.
-  url = 'http://localhost:{port}/containers/{container_id}/json'.format(
-      port=gs.get_docker_port(), container_id=hex_container_id)
+  # Must specify an explicit host name (not "localhost").
+  url = 'http://{host}:{port}/containers/{container_id}/json'.format(
+      host=my_node_name, port=gs.get_docker_port(),
+      container_id=hex_container_id)
   container = fetch_data(gs, url, 'container-' + hex_container_id[:12])
 
   # Fetch the image symbolic name and hex ID from the container information.
@@ -735,8 +757,10 @@ def get_version(gs):
     raise collector_error.CollectorError(msg)
 
   # Fetch image information.
-  url = 'http://localhost:{port}/images/{image_id}/json'.format(
-      port=gs.get_docker_port(), image_id=hex_image_id)
+  # Must specify an explicit host name (not "localhost").
+  url = 'http://{host}:{port}/images/{image_id}/json'.format(
+      host=my_node_name, port=gs.get_docker_port(),
+      image_id=hex_image_id)
   image = fetch_data(gs, url, 'image-' + hex_image_id[:12])
 
   # Fetch the image creation timestamp.

collector/docker_proxy.py

@@ -132,8 +132,7 @@ def get_response(req, cache=None):
         requests.codes.ok,
         {'Content-Type': 'application/json'})
 
-  except Exception as e:
-    app.logger.error(e, exc_info=True)
+  except:
     exc_type, value, _ = sys.exc_info()
     msg = ('Failed to retrieve %s with exception %s: %s' %
            (req, exc_type, value))
@@ -167,8 +166,7 @@ def fill_cache(cache):
     app.logger.error('invalid response format from "/containers/json"')
     return
 
-  except Exception as e:
-    app.logger.error(e, exc_info=True)
+  except:
     exc_type, value, _ = sys.exc_info()
     msg = ('Failed to fetch /containers/json with exception %s: %s' %
            (exc_type, value))
@@ -196,8 +194,7 @@ def fill_cache(cache):
       cache.update(req, json.dumps(result))
       app.logger.info('caching result of request=%s', req)
 
-    except Exception as e:
-      app.logger.error(e, exc_info=True)
+    except:
       exc_type, value, _ = sys.exc_info()
       msg = ('Failed to fetch %s with exception %s: %s' %
              (req, exc_type, value))

collector/testdata/proc-self-cgroup.txt

@@ -1,10 +1,10 @@
 10:net_prio:/
-9:perf_event:/docker/21bef774fd5e952dc07d9b81b96fd0b5d2b0b0cfdeb68e6262eb8f74c3555a00
-8:blkio:/docker/21bef774fd5e952dc07d9b81b96fd0b5d2b0b0cfdeb68e6262eb8f74c3555a00
+9:perf_event:/21bef774fd5e952dc07d9b81b96fd0b5d2b0b0cfdeb68e6262eb8f74c3555a00
+8:blkio:/21bef774fd5e952dc07d9b81b96fd0b5d2b0b0cfdeb68e6262eb8f74c3555a00
 7:net_cls:/
-6:freezer:/docker/21bef774fd5e952dc07d9b81b96fd0b5d2b0b0cfdeb68e6262eb8f74c3555a00
-5:devices:/docker/21bef774fd5e952dc07d9b81b96fd0b5d2b0b0cfdeb68e6262eb8f74c3555a00
-4:memory:/docker/21bef774fd5e952dc07d9b81b96fd0b5d2b0b0cfdeb68e6262eb8f74c3555a00
-3:cpuacct:/docker/21bef774fd5e952dc07d9b81b96fd0b5d2b0b0cfdeb68e6262eb8f74c3555a00
-2:cpu:/docker/21bef774fd5e952dc07d9b81b96fd0b5d2b0b0cfdeb68e6262eb8f74c3555a00
-1:cpuset:/docker/21bef774fd5e952dc07d9b81b96fd0b5d2b0b0cfdeb68e6262eb8f74c3555a00
+6:freezer:/21bef774fd5e952dc07d9b81b96fd0b5d2b0b0cfdeb68e6262eb8f74c3555a00
+5:devices:/21bef774fd5e952dc07d9b81b96fd0b5d2b0b0cfdeb68e6262eb8f74c3555a00
+4:memory:/21bef774fd5e952dc07d9b81b96fd0b5d2b0b0cfdeb68e6262eb8f74c3555a00
+3:cpuacct:/21bef774fd5e952dc07d9b81b96fd0b5d2b0b0cfdeb68e6262eb8f74c3555a00
+2:cpu:/21bef774fd5e952dc07d9b81b96fd0b5d2b0b0cfdeb68e6262eb8f74c3555a00
+1:cpuset:/21bef774fd5e952dc07d9b81b96fd0b5d2b0b0cfdeb68e6262eb8f74c3555a00