13804: Node must report as "idle" two consecutive times

[arvados.git] / services / nodemanager / arvnodeman / computenode / dispatch / __init__.py

diff --git a/services/nodemanager/arvnodeman/computenode/dispatch/__init__.py b/services/nodemanager/arvnodeman/computenode/dispatch/__init__.py
index 597a011e72075975706a42a31ed03bc74ff42b36..d9b475b908795546d4506e23989631fd100e9777 100644 (file)
--- a/services/nodemanager/arvnodeman/computenode/dispatch/__init__.py
+++ b/services/nodemanager/arvnodeman/computenode/dispatch/__init__.py
@@ -20,6 +20,7 @@ from .. import \
     arvados_node_missing, RetryMixin
 from ...clientactor import _notify_subscribers
 from ... import config
+from ... import status
 from .transitions import transitions
 
 QuotaExceeded = "QuotaExceeded"
@@ -122,7 +123,7 @@ class ComputeNodeSetupActor(ComputeNodeStateChangeBase):
     def prepare_arvados_node(self, node):
         self._clean_arvados_node(node, "Prepared by Node Manager")
         self.arvados_node = self._arvados.nodes().update(
-            body={}, assign_slot=True).execute()
+            uuid=node['uuid'], body={}, assign_slot=True).execute()
         self._later.create_cloud_node()
 
     @ComputeNodeStateChangeBase._finish_on_exception
@@ -272,6 +273,9 @@ class ComputeNodeShutdownActor(ComputeNodeStateChangeBase):
                 self.cancel_shutdown("No longer eligible for shut down because %s" % reason,
                                      try_resume=True)
                 return
+        # If boot failed, count the event
+        if self._monitor.get_state().get() == 'unpaired':
+            status.tracker.counter_add('boot_failures')
         self._destroy_node()
 
     def _destroy_node(self):
@@ -346,6 +350,7 @@ class ComputeNodeMonitorActor(config.actor_class):
         self.boot_fail_after = boot_fail_after
         self.subscribers = set()
         self.arvados_node = None
+        self.consecutive_idle = 0
         self._later.update_arvados_node(arvados_node)
         self.last_shutdown_opening = None
         self._later.consider_shutdown()
@@ -409,6 +414,12 @@ class ComputeNodeMonitorActor(config.actor_class):
         #if state == 'idle' and self.arvados_node['job_uuid']:
         #    state = 'busy'
 
+        # Update idle node times tracker
+        if state == 'idle':
+            status.tracker.idle_in(self.arvados_node['hostname'])
+        else:
+            status.tracker.idle_out(self.arvados_node['hostname'])
+
         return state
 
     def in_state(self, *states):
@@ -441,8 +452,14 @@ class ComputeNodeMonitorActor(config.actor_class):
         else:
             boot_grace = "boot exceeded"
 
-        # API server side not implemented yet.
-        idle_grace = 'idle exceeded'
+        if crunch_worker_state == "idle":
+            # Must report as "idle" at least two consecutive times
+            if self.consecutive_idle < 2:
+                idle_grace = 'idle wait'
+            else:
+                idle_grace = 'idle exceeded'
+        else:
+            idle_grace = 'not idle'
 
         node_state = (crunch_worker_state, window, boot_grace, idle_grace)
         t = transitions[node_state]
@@ -502,4 +519,8 @@ class ComputeNodeMonitorActor(config.actor_class):
         if arvados_node is not None:
             self.arvados_node = arvados_node
             self._update.sync_node(self.cloud_node, self.arvados_node)
+            if self.arvados_node['crunch_worker_state'] == "idle":
+                self.consecutive_idle += 1
+            else:
+                self.consecutive_idle = 0
             self._later.consider_shutdown()