[ARVADOS] updated: 2e919859109fe27d552b81b13d47aed61e80eca6

[git at public.curoverse.com]

Summary of changes:
 services/nodemanager/arvnodeman/computenode/__init__.py |  5 ++++-
 services/nodemanager/arvnodeman/daemon.py               |  3 ++-
 services/nodemanager/tests/test_daemon.py               | 17 +++++++++++++++--
 3 files changed, 21 insertions(+), 4 deletions(-)

       via  2e919859109fe27d552b81b13d47aed61e80eca6 (commit)
      from  be81c03a3c26f365eba35b91e4f0827244a02ef7 (commit)

Those revisions listed above that are new to this repository have
not appeared on any other notification email; so we list those
revisions in full, below.


commit 2e919859109fe27d552b81b13d47aed61e80eca6
Author: Peter Amstutz <peter.amstutz at curoverse.com>
Date:   Wed Sep 30 17:16:09 2015 -0400

    7286: Fix double count of missing nodes in shutdown

diff --git a/services/nodemanager/arvnodeman/computenode/__init__.py b/services/nodemanager/arvnodeman/computenode/__init__.py
index b47866d..e7bd7bf 100644
--- a/services/nodemanager/arvnodeman/computenode/__init__.py
+++ b/services/nodemanager/arvnodeman/computenode/__init__.py
@@ -31,7 +31,10 @@ def timestamp_fresh(timestamp, fresh_time):
     return (time.time() - timestamp) < fresh_time
 
 def arvados_node_missing(arvados_node, fresh_time):
-    return not timestamp_fresh(arvados_timestamp(arvados_node["last_ping_at"]), fresh_time)
+    if arvados_node["last_ping_at"] is None:
+        return None
+    else:
+        return not timestamp_fresh(arvados_timestamp(arvados_node["last_ping_at"]), fresh_time)
 
 class ShutdownTimer(object):
     """Keep track of a cloud node's shutdown windows.
diff --git a/services/nodemanager/arvnodeman/daemon.py b/services/nodemanager/arvnodeman/daemon.py
index 6f33a3b..ddddd41 100644
--- a/services/nodemanager/arvnodeman/daemon.py
+++ b/services/nodemanager/arvnodeman/daemon.py
@@ -212,7 +212,8 @@ class NodeManagerDaemonActor(actor_class):
     def _nodes_missing(self):
         return sum(1 for arv_node in
                    pykka.get_all(rec.actor.arvados_node for rec in
-                                 self.cloud_nodes.nodes.itervalues())
+                                 self.cloud_nodes.nodes.itervalues()
+                                 if rec.actor.cloud_node.get().id not in self.shutdowns)
                    if arv_node and cnode.arvados_node_missing(arv_node, self.node_stale_after))
 
     def _nodes_wanted(self):
diff --git a/services/nodemanager/tests/test_daemon.py b/services/nodemanager/tests/test_daemon.py
index 57ea46c..0206f4c 100644
--- a/services/nodemanager/tests/test_daemon.py
+++ b/services/nodemanager/tests/test_daemon.py
@@ -148,8 +148,8 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin,
 
     def test_excess_counts_missing(self):
         size = testutil.MockSize(1)
-        self.make_daemon(cloud_nodes=[testutil.cloud_node_mock(1),
-                                      testutil.cloud_node_mock(2)],
+        cloud_nodes = [testutil.cloud_node_mock(1), testutil.cloud_node_mock(2)]
+        self.make_daemon(cloud_nodes=cloud_nodes,
                          arvados_nodes=[testutil.arvados_node_mock(1),
                                         testutil.arvados_node_mock(2, last_ping_at='1970-01-01T01:02:03.04050607Z')],
                          want_sizes=[size])
@@ -158,6 +158,19 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin,
             self.daemon.node_can_shutdown(mon_ref.proxy()).get(self.TIMEOUT)
         self.assertEqual(1, self.node_shutdown.start.call_count)
 
+    def test_missing_shutdown_not_excess(self):
+        size = testutil.MockSize(1)
+        cloud_nodes = [testutil.cloud_node_mock(1), testutil.cloud_node_mock(2)]
+        self.make_daemon(cloud_nodes=cloud_nodes,
+                         arvados_nodes=[testutil.arvados_node_mock(1),
+                                        testutil.arvados_node_mock(2, last_ping_at='1970-01-01T01:02:03.04050607Z')],
+                         want_sizes=[size])
+        self.daemon.shutdowns.get()[cloud_nodes[1].id] = True
+        self.assertEqual(2, self.alive_monitor_count())
+        for mon_ref in self.monitor_list():
+            self.daemon.node_can_shutdown(mon_ref.proxy()).get(self.TIMEOUT)
+        self.assertEqual(0, self.node_shutdown.start.call_count)
+
     def test_booting_nodes_counted(self):
         cloud_node = testutil.cloud_node_mock(1)
         arv_node = testutil.arvados_node_mock(1)

-----------------------------------------------------------------------


hooks/post-receive
--