[ARVADOS] created: 29379beaa615d5a36032a05e71d7a2730e255c48

[Git user]

        at  29379beaa615d5a36032a05e71d7a2730e255c48 (commit)


commit 29379beaa615d5a36032a05e71d7a2730e255c48
Author: Peter Amstutz <peter.amstutz at curoverse.com>
Date:   Fri Apr 15 22:48:13 2016 -0400

    Don't double-count nodes that are shutting down.  refs #8953

diff --git a/services/nodemanager/arvnodeman/daemon.py b/services/nodemanager/arvnodeman/daemon.py
index 519213f..1120440 100644
--- a/services/nodemanager/arvnodeman/daemon.py
+++ b/services/nodemanager/arvnodeman/daemon.py
@@ -302,17 +302,17 @@ class NodeManagerDaemonActor(actor_class):
         booting_count = self._nodes_booting(size) + self._nodes_unpaired(size)
         shutdown_count = self._size_shutdowns(size)
         busy_count = self._nodes_busy(size)
-        up_count = self._nodes_up(size) - (shutdown_count + busy_count + self._nodes_missing(size))
+        idle_count = self._nodes_up(size) - (busy_count + self._nodes_missing(size))
 
         self._logger.info("%s: wishlist %i, up %i (booting %i, idle %i, busy %i), shutting down %i", size.name,
                           self._size_wishlist(size),
-                          up_count + busy_count,
+                          idle_count + busy_count,
                           booting_count,
-                          up_count - booting_count,
+                          idle_count - booting_count,
                           busy_count,
                           shutdown_count)
 
-        wanted = self._size_wishlist(size) - up_count
+        wanted = self._size_wishlist(size) - idle_count
         if wanted > 0 and self.max_total_price and ((total_price + (size.price*wanted)) > self.max_total_price):
             can_boot = int((self.max_total_price - total_price) / size.price)
             if can_boot == 0:
@@ -323,7 +323,7 @@ class NodeManagerDaemonActor(actor_class):
             return wanted
 
     def _nodes_excess(self, size):
-        up_count = self._nodes_up(size) - self._size_shutdowns(size)
+        up_count = (self._nodes_booting(size) + self._nodes_booted(size)) - self._size_shutdowns(size)
         if size.id == self.min_cloud_size.id:
             up_count -= self.min_nodes
         return up_count - self._nodes_busy(size) - self._size_wishlist(size)
diff --git a/services/nodemanager/tests/test_computenode_dispatch_slurm.py b/services/nodemanager/tests/test_computenode_dispatch_slurm.py
index 0f98afc..92858b6 100644
--- a/services/nodemanager/tests/test_computenode_dispatch_slurm.py
+++ b/services/nodemanager/tests/test_computenode_dispatch_slurm.py
@@ -61,20 +61,26 @@ class SLURMComputeNodeShutdownActorTestCase(ComputeNodeShutdownActorMixin,
         self.assertFalse(proc_mock.called)
 
     def test_node_undrained_when_shutdown_cancelled(self, proc_mock):
-        proc_mock.side_effect = iter(['drng\n', 'idle\n'])
-        self.make_mocks(arvados_node=testutil.arvados_node_mock(job_uuid=True))
-        self.make_actor()
-        self.shutdown_actor.cancel_shutdown("test")
-        self.check_success_flag(False, 2)
-        self.check_slurm_got_args(proc_mock, 'NodeName=compute99', 'State=RESUME')
+        try:
+            proc_mock.side_effect = iter(['drng\n', 'idle\n'])
+            self.make_mocks(arvados_node=testutil.arvados_node_mock(job_uuid=True))
+            self.make_actor()
+            self.shutdown_actor.cancel_shutdown("test")
+            self.check_success_flag(False, 2)
+            self.check_slurm_got_args(proc_mock, 'NodeName=compute99', 'State=RESUME')
+        finally:
+            self.shutdown_actor.actor_ref.stop()
 
     def test_alloc_node_undrained_when_shutdown_cancelled(self, proc_mock):
-        proc_mock.side_effect = iter(['alloc\n'])
-        self.make_mocks(arvados_node=testutil.arvados_node_mock(job_uuid=True))
-        self.make_actor()
-        self.shutdown_actor.cancel_shutdown("test")
-        self.check_success_flag(False, 2)
-        self.check_slurm_got_args(proc_mock, 'sinfo', '--noheader', '-o', '%t', '-n', 'compute99')
+        try:
+            proc_mock.side_effect = iter(['alloc\n'])
+            self.make_mocks(arvados_node=testutil.arvados_node_mock(job_uuid=True))
+            self.make_actor()
+            self.shutdown_actor.cancel_shutdown("test")
+            self.check_success_flag(False, 2)
+            self.check_slurm_got_args(proc_mock, 'sinfo', '--noheader', '-o', '%t', '-n', 'compute99')
+        finally:
+            self.shutdown_actor.actor_ref.stop()
 
     def test_cancel_shutdown_retry(self, proc_mock):
         proc_mock.side_effect = iter([OSError, 'drain\n', OSError, 'idle\n', 'idle\n'])
diff --git a/services/nodemanager/tests/test_daemon.py b/services/nodemanager/tests/test_daemon.py
index 3b5f721..7da250b 100644
--- a/services/nodemanager/tests/test_daemon.py
+++ b/services/nodemanager/tests/test_daemon.py
@@ -522,7 +522,7 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin,
     def test_nodes_shutting_down_replaced_below_max_nodes(self):
         size = testutil.MockSize(6)
         cloud_node = testutil.cloud_node_mock(6, size=size)
-        self.make_daemon([cloud_node], [testutil.arvados_node_mock(6)],
+        self.make_daemon([cloud_node], [testutil.arvados_node_mock(6, crunch_worker_state='down')],
                          avail_sizes=[(size, {"cores":1})])
         self.assertEqual(1, self.alive_monitor_count())
         monitor = self.monitor_list()[0].proxy()

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


hooks/post-receive
--