Merge branch '8697-ruby187-compat'

[arvados.git] / services / nodemanager / arvnodeman / daemon.py

diff --git a/services/nodemanager/arvnodeman/daemon.py b/services/nodemanager/arvnodeman/daemon.py
index a02da230d8a8b8765debcc18f6971bb1769aacb2..11204409c03cd1894ac4128f1bfc84e3a4767652 100644 (file)
--- a/services/nodemanager/arvnodeman/daemon.py
+++ b/services/nodemanager/arvnodeman/daemon.py
@@ -151,9 +151,6 @@ class NodeManagerDaemonActor(actor_class):
     def _update_poll_time(self, poll_key):
         self.last_polls[poll_key] = time.time()
 
-    def _resume_node(self, node_record):
-        node_record.actor.resume_node()
-
     def _pair_nodes(self, node_record, arvados_node):
         self._logger.info("Cloud node %s is now paired with Arvados node %s",
                           node_record.cloud_node.name, arvados_node['uuid'])
@@ -221,15 +218,6 @@ class NodeManagerDaemonActor(actor_class):
                 if cloud_rec.actor.offer_arvados_pair(arv_node).get():
                     self._pair_nodes(cloud_rec, arv_node)
                     break
-        for rec in self.cloud_nodes.nodes.itervalues():
-            # crunch-dispatch turns all slurm states that are not either "idle"
-            # or "alloc" into "down", but in case that behavior changes, assume
-            # any state that is not "idle" or "alloc" could be a state we want
-            # to try to resume from.
-            if (rec.arvados_node is not None and
-                rec.arvados_node["info"].get("slurm_state") not in ("idle", "alloc") and
-                rec.cloud_node.id not in self.shutdowns):
-                self._resume_node(rec)
 
     def _nodes_booting(self, size):
         s = sum(1
@@ -314,17 +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:
@@ -335,7 +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)