7709: Merge branch 'master' into 7709-api-rails4

[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 95b6fa8e0ce962d67b43b26f8106469d9d692faf..29b04845b653190f88c306094486ac030add3af7 100644 (file)
--- a/services/nodemanager/arvnodeman/computenode/driver/__init__.py
+++ b/services/nodemanager/arvnodeman/computenode/driver/__init__.py
@@ -6,6 +6,7 @@ import logging
 from operator import attrgetter
 
 import libcloud.common.types as cloud_types
+from libcloud.common.exceptions import BaseHTTPError
 from libcloud.compute.base import NodeDriver, NodeAuthSSHKey
 
 from ...config import NETWORK_ERRORS
@@ -86,13 +87,17 @@ class BaseComputeNodeDriver(RetryMixin):
 
         Arguments:
         * term: The value that identifies a matching item.
-        * list_method: A string that names the method to call on this
-          instance's libcloud driver for a list of objects.
+        * list_method: A string that names the method to call for a
+          list of objects.
         * key: A function that accepts a cloud object and returns a
           value search for a `term` match on each item.  Returns the
           object's 'id' attribute by default.
         """
-        items = getattr(self.real, list_method)(**kwargs)
+        try:
+            list_func = getattr(self, list_method)
+        except AttributeError:
+            list_func = getattr(self.real, list_method)
+        items = list_func(**kwargs)
         results = [item for item in items if key(item) == term]
         count = len(results)
         if count != 1:
@@ -206,9 +211,31 @@ class BaseComputeNodeDriver(RetryMixin):
         # libcloud compute drivers typically raise bare Exceptions to
         # represent API errors.  Return True for any exception that is
         # exactly an Exception, or a better-known higher-level exception.
+        if (type(exception) is BaseHTTPError and
+            exception.message and
+            (exception.message.startswith("InvalidInstanceID.NotFound") or
+             exception.message.startswith("InstanceLimitExceeded"))):
+            return True
         return (isinstance(exception, cls.CLOUD_ERRORS) or
                 type(exception) is Exception)
 
+    def destroy_node(self, cloud_node):
+        try:
+            return self.real.destroy_node(cloud_node)
+        except self.CLOUD_ERRORS as destroy_error:
+            # 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
+            # on the node list.  If it is gone, we can declare victory.
+            try:
+                self.search_for_now(cloud_node.id, 'list_nodes')
+            except ValueError:
+                # If we catch ValueError, that means search_for_now didn't find
+                # it, which means destroy_node actually succeeded.
+                return True
+            # The node is still on the list.  Re-raise.
+            raise
+
     # Now that we've defined all our own methods, delegate generic, public
     # attributes of libcloud drivers that we haven't defined ourselves.
     def _delegate_to_real(attr_name):