#!/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
import libcloud.common.types as cloud_types
from libcloud.compute.base import NodeDriver, NodeAuthSSHKey
-from ...config import NETWORK_ERRORS
+from ...config import CLOUD_ERRORS
+from ...status import tracker
from .. import RetryMixin
class BaseComputeNodeDriver(RetryMixin):
Subclasses must implement arvados_create_kwargs, sync_node,
node_fqdn, and node_start_time.
"""
- CLOUD_ERRORS = NETWORK_ERRORS + (cloud_types.LibcloudError,)
+
@RetryMixin._retry()
def _create_driver(self, driver_class, **auth_kwargs):
value search for a `term` match on each item. Returns the
object's 'id' attribute by default.
"""
- items = getattr(self, 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:
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.
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 self.CLOUD_ERRORS as create_error:
+ except CLOUD_ERRORS as create_error:
# Workaround for bug #6702: sometimes the create node request
# succeeds but times out and raises an exception instead of
# returning a result. If this happens, we get stuck in a retry
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):
# seconds since the epoch UTC.
raise NotImplementedError("BaseComputeNodeDriver.node_start_time")
- @classmethod
- def is_cloud_exception(cls, exception):
- # 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.
- 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 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
+ # 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.
+ tracker.counter_add('destroy_node_errors')
+ raise
# Now that we've defined all our own methods, delegate generic, public
# attributes of libcloud drivers that we haven't defined ourselves.