Merge branch 'master' into 13822-nm-delayed-daemon

[arvados.git] / services / nodemanager / arvnodeman / computenode / driver / __init__.py

diff --git a/services/nodemanager/arvnodeman/computenode/driver/__init__.py b/services/nodemanager/arvnodeman/computenode/driver/__init__.py
index c8c54dc0bfb9384c3f79545309fff86433409133..48d19f592bbdb0b87d905bac377c849000b59ef1 100644 (file)
--- a/services/nodemanager/arvnodeman/computenode/driver/__init__.py
+++ b/services/nodemanager/arvnodeman/computenode/driver/__init__.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
 
@@ -9,6 +12,7 @@ import libcloud.common.types as cloud_types
 from libcloud.compute.base import NodeDriver, NodeAuthSSHKey
 
 from ...config import CLOUD_ERRORS
+from ...status import tracker
 from .. import RetryMixin
 
 class BaseComputeNodeDriver(RetryMixin):
@@ -31,8 +35,10 @@ class BaseComputeNodeDriver(RetryMixin):
         return driver_class(**auth_kwargs)
 
     @RetryMixin._retry()
-    def _set_sizes(self):
-        self.sizes = {sz.id: sz for sz in self.real.list_sizes()}
+    def sizes(self):
+        if self._sizes is None:
+            self._sizes = {sz.id: sz for sz in self.real.list_sizes()}
+        return self._sizes
 
     def __init__(self, auth_kwargs, list_kwargs, create_kwargs,
                  driver_class, retry_wait=1, max_retry_wait=180):
@@ -69,7 +75,7 @@ class BaseComputeNodeDriver(RetryMixin):
                 if new_pair is not None:
                     self.create_kwargs[new_pair[0]] = new_pair[1]
 
-        self._set_sizes()
+        self._sizes = None
 
     def _init_ping_host(self, ping_host):
         self.ping_host = ping_host
@@ -120,7 +126,11 @@ class BaseComputeNodeDriver(RetryMixin):
     def list_nodes(self, **kwargs):
         l = self.list_kwargs.copy()
         l.update(kwargs)
-        return self.real.list_nodes(**l)
+        try:
+            return self.real.list_nodes(**l)
+        except CLOUD_ERRORS:
+            tracker.counter_add('list_nodes_errors')
+            raise
 
     def create_cloud_name(self, arvados_node):
         """Return a cloud node name for the given Arvados node record.
@@ -166,7 +176,7 @@ class BaseComputeNodeDriver(RetryMixin):
         try:
             kwargs = self.create_kwargs.copy()
             kwargs.update(self.arvados_create_kwargs(size, arvados_node))
-            kwargs['size'] = size
+            kwargs['size'] = size.real
             return self.real.create_node(**kwargs)
         except CLOUD_ERRORS as create_error:
             # Workaround for bug #6702: sometimes the create node request
@@ -178,6 +188,7 @@ class BaseComputeNodeDriver(RetryMixin):
             try:
                 return self.search_for_now(kwargs['name'], 'list_nodes', self._name_key)
             except ValueError:
+                tracker.counter_add('create_node_errors')
                 raise create_error
 
     def post_create_node(self, cloud_node):
@@ -208,7 +219,7 @@ class BaseComputeNodeDriver(RetryMixin):
     def destroy_node(self, cloud_node):
         try:
             return self.real.destroy_node(cloud_node)
-        except CLOUD_ERRORS as destroy_error:
+        except CLOUD_ERRORS:
             # Sometimes the destroy node request succeeds but times out and
             # raises an exception instead of returning success.  If this
             # happens, we get a noisy stack trace.  Check if the node is still
@@ -220,6 +231,7 @@ class BaseComputeNodeDriver(RetryMixin):
                 # it, which means destroy_node actually succeeded.
                 return True
             # The node is still on the list.  Re-raise.
+            tracker.counter_add('destroy_node_errors')
             raise
 
     # Now that we've defined all our own methods, delegate generic, public