7478: Moves invalid cloud size node's shutdown decision to proper method.

[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 dec8ef1434c5f5ca08290ab71f0ecac765dc6ab9..f7bb2cdbda19776d87f32f93bf73d64f8c214867 100644 (file)
--- a/services/nodemanager/arvnodeman/computenode/dispatch/__init__.py
+++ b/services/nodemanager/arvnodeman/computenode/dispatch/__init__.py
@@ -370,11 +370,6 @@ class ComputeNodeMonitorActor(config.actor_class):
     def get_state(self):
         """Get node state, one of ['unpaired', 'busy', 'idle', 'down']."""
 
-        # If this node's size is invalid (because it has a stale arvados_node_size
-        # tag), return 'down' so that it's properly shut down.
-        if self.cloud_node.size.id == 'invalid':
-            return 'down'
-
         # If this node is not associated with an Arvados node, return
         # 'unpaired' if we're in the boot grace period, and 'down' if not,
         # so it isn't counted towards usable nodes.
@@ -437,6 +432,11 @@ class ComputeNodeMonitorActor(config.actor_class):
         reason for the decision.
         """
 
+        # If this node's size is invalid (because it has a stale arvados_node_size
+        # tag), return True so that it's properly shut down.
+        if self.cloud_node.size.id == 'invalid':
+            return (True, "node's size tag '%s' not recognizable" % (self.cloud_node.extra['arvados_node_size'],))
+
         # Collect states and then consult state transition table whether we
         # should shut down.  Possible states are:
         # crunch_worker_state = ['unpaired', 'busy', 'idle', 'down']