12055: Avoid RAM exhaustion on bootup by asking AWS only the AMI

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

diff --git a/services/nodemanager/arvnodeman/config.py b/services/nodemanager/arvnodeman/config.py
index f4ffff5fb60a23598d03fcea38a47d09af945ead..e47f9fcb1d036b78f94af0af25e8c37dc17b5ad0 100644 (file)
--- a/services/nodemanager/arvnodeman/config.py
+++ b/services/nodemanager/arvnodeman/config.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
 
@@ -14,11 +17,15 @@ from apiclient import errors as apierror
 
 from .baseactor import BaseNodeManagerActor
 
+from libcloud.common.types import LibcloudError
+from libcloud.common.exceptions import BaseHTTPError
+
 # IOError is the base class for socket.error, ssl.SSLError, and friends.
 # It seems like it hits the sweet spot for operations we want to retry:
 # it's low-level, but unlikely to catch code bugs.
 NETWORK_ERRORS = (IOError,)
 ARVADOS_ERRORS = NETWORK_ERRORS + (apierror.Error,)
+CLOUD_ERRORS = NETWORK_ERRORS + (LibcloudError, BaseHTTPError)
 
 actor_class = BaseNodeManagerActor
 
@@ -51,9 +58,10 @@ class NodeManagerConfig(ConfigParser.SafeConfigParser):
                        'watchdog': '600',
                        'node_mem_scaling': '0.95'},
             'Manage': {'address': '127.0.0.1',
-                       'port': '-1'},
+                       'port': '-1',
+                       'ManagementToken': ''},
             'Logging': {'file': '/dev/stderr',
-                        'level': 'WARNING'},
+                        'level': 'WARNING'}
         }.iteritems():
             if not self.has_section(sec_name):
                 self.add_section(sec_name)
@@ -107,7 +115,7 @@ class NodeManagerConfig(ConfigParser.SafeConfigParser):
         module = importlib.import_module('arvnodeman.computenode.driver.' +
                                          self.get('Cloud', 'provider'))
         driver_class = module.ComputeNodeDriver.DEFAULT_DRIVER
-        if self.get('Cloud', 'driver_class'):
+        if self.has_option('Cloud', 'driver_class'):
             d = self.get('Cloud', 'driver_class').split('.')
             mod = '.'.join(d[:-1])
             cls = d[-1]