X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/ea47e67af0e7528f0bcb23f3b34019b308eaa68a..8e31910034627dedd7259dd1e45a60768108c1e1:/services/nodemanager/arvnodeman/nodelist.py

diff --git a/services/nodemanager/arvnodeman/nodelist.py b/services/nodemanager/arvnodeman/nodelist.py
index 7bc3a5ebd2..66af7c32d1 100644
--- a/services/nodemanager/arvnodeman/nodelist.py
+++ b/services/nodemanager/arvnodeman/nodelist.py
@@ -1,4 +1,7 @@
 #!/usr/bin/env python
+# Copyright (C) The Arvados Authors. All rights reserved.
+#
+# SPDX-License-Identifier: AGPL-3.0
 
 from __future__ import absolute_import, print_function
 
@@ -12,8 +15,9 @@ import arvados.util
 class ArvadosNodeListMonitorActor(clientactor.RemotePollLoopActor):
     """Actor to poll the Arvados node list.
 
-    This actor regularly polls the list of Arvados node records, and
-    sends it to subscribers.
+    This actor regularly polls the list of Arvados node records,
+    augments it with the latest SLURM node info (`sinfo`), and sends
+    it to subscribers.
     """
 
     def is_common_error(self, exception):
@@ -26,28 +30,32 @@ class ArvadosNodeListMonitorActor(clientactor.RemotePollLoopActor):
         nodelist = arvados.util.list_all(self._client.nodes().list)
 
         # node hostname, state
-        sinfo_out = subprocess.check_output(["sinfo", "--noheader", "--format=%n %t"])
+        sinfo_out = subprocess.check_output(["sinfo", "--noheader", "--format=%n|%t|%f"])
         nodestates = {}
+        nodefeatures = {}
         for out in sinfo_out.splitlines():
             try:
-                nodename, state = out.split(" ", 2)
-                if state in ('alloc', 'alloc*',
-                             'comp',  'comp*',
-                             'mix',   'mix*',
-                             'drng',  'drng*'):
-                    nodestates[nodename] = 'busy'
-                elif state == 'idle':
-                    nodestates[nodename] = 'idle'
-                else:
-                    nodestates[nodename] = 'down'
+                nodename, state, features = out.split("|", 3)
             except ValueError:
-                pass
+                continue
+            if state in ('alloc', 'alloc*',
+                         'comp',  'comp*',
+                         'mix',   'mix*',
+                         'drng',  'drng*'):
+                nodestates[nodename] = 'busy'
+            elif state in ('idle', 'fail'):
+                nodestates[nodename] = state
+            else:
+                nodestates[nodename] = 'down'
+            if features != "(null)":
+                nodefeatures[nodename] = features
 
         for n in nodelist:
             if n["slot_number"] and n["hostname"] and n["hostname"] in nodestates:
                 n["crunch_worker_state"] = nodestates[n["hostname"]]
             else:
                 n["crunch_worker_state"] = 'down'
+            n["slurm_node_features"] = nodefeatures.get(n["hostname"], "")
 
         return nodelist
 
@@ -64,7 +72,7 @@ class CloudNodeListMonitorActor(clientactor.RemotePollLoopActor):
         self._calculator = server_calc
 
     def is_common_error(self, exception):
-        return self._client.is_cloud_exception(exception)
+        return isinstance(exception, config.CLOUD_ERRORS)
 
     def _item_key(self, node):
         return node.id
@@ -72,8 +80,8 @@ class CloudNodeListMonitorActor(clientactor.RemotePollLoopActor):
     def _send_request(self):
         nodes = self._client.list_nodes()
         for n in nodes:
-            # Replace with libcloud NodeSize object with compatible
+            # Replace the libcloud NodeSize object with compatible
             # CloudSizeWrapper object which merges the size info reported from
             # the cloud with size information from the configuration file.
-            n.size = self._calculator.find_size(n.size.id)
+            n.size = self._calculator.find_size(n.extra['arvados_node_size'])
         return nodes