13804: Try to cancel pending shutdowns if nodes are needed

[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 9106ea67ccc8ffac7813d64baa5ebc537548fa21..bb83a193fa30bae46dc7a7f6979843eb65fd42d1 100644 (file)
--- a/services/nodemanager/arvnodeman/computenode/dispatch/__init__.py
+++ b/services/nodemanager/arvnodeman/computenode/dispatch/__init__.py
@@ -243,12 +243,15 @@ class ComputeNodeShutdownActor(ComputeNodeStateChangeBase):
         return super(ComputeNodeShutdownActor, self)._finished()
 
     def cancel_shutdown(self, reason, **kwargs):
+        if not self.cancellable:
+            return False
         if self.cancel_reason is not None:
             # already cancelled
-            return
+            return False
         self.cancel_reason = reason
         self._logger.info("Shutdown cancelled: %s.", reason)
         self._finished(success_flag=False)
+        return True
 
     def _cancel_on_exception(orig_func):
         @functools.wraps(orig_func)
@@ -282,6 +285,7 @@ class ComputeNodeShutdownActor(ComputeNodeStateChangeBase):
         self._logger.info("Starting shutdown")
         arv_node = self._arvados_node()
         if self._cloud.destroy_node(self.cloud_node):
+            self.cancellable = False
             self._logger.info("Shutdown success")
             if arv_node:
                 self._later.clean_arvados_node(arv_node)
@@ -350,6 +354,7 @@ 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()
@@ -451,8 +456,14 @@ 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]
@@ -512,4 +523,8 @@ 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()