#!/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 logging
+from operator import attrgetter
+
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 .. import RetryMixin
-class BaseComputeNodeDriver(object):
+class BaseComputeNodeDriver(RetryMixin):
"""Abstract base class for compute node drivers.
- libcloud abstracts away many of the differences between cloud providers,
- but managing compute nodes requires some cloud-specific features (e.g.,
- on EC2 we use tags to identify compute nodes). Compute node drivers
- are responsible for translating the node manager's cloud requests to a
- specific cloud's vocabulary.
+ libcloud drivers abstract away many of the differences between
+ cloud providers, but managing compute nodes requires some
+ cloud-specific features (e.g., keeping track of node FQDNs and
+ boot times). Compute node drivers are responsible for translating
+ the node manager's cloud requests to a specific cloud's
+ vocabulary.
- Subclasses must implement arvados_create_kwargs (to update node
- creation kwargs with information about the specific Arvados node
- record), sync_node, and node_start_time.
+ Subclasses must implement arvados_create_kwargs, sync_node,
+ node_fqdn, and node_start_time.
"""
- CLOUD_ERRORS = NETWORK_ERRORS + (cloud_types.LibcloudError,)
- def __init__(self, auth_kwargs, list_kwargs, create_kwargs, driver_class):
- self.real = driver_class(**auth_kwargs)
+
+ @RetryMixin._retry()
+ def _create_driver(self, driver_class, **auth_kwargs):
+ return driver_class(**auth_kwargs)
+
+ @RetryMixin._retry()
+ def _set_sizes(self):
+ self.sizes = {sz.id: sz for sz in self.real.list_sizes()}
+
+ def __init__(self, auth_kwargs, list_kwargs, create_kwargs,
+ driver_class, retry_wait=1, max_retry_wait=180):
+ """Base initializer for compute node drivers.
+
+ Arguments:
+ * auth_kwargs: A dictionary of arguments that are passed into the
+ driver_class constructor to instantiate a libcloud driver.
+ * list_kwargs: A dictionary of arguments that are passed to the
+ libcloud driver's list_nodes method to return the list of compute
+ nodes.
+ * create_kwargs: A dictionary of arguments that are passed to the
+ libcloud driver's create_node method to create a new compute node.
+ * driver_class: The class of a libcloud driver to use.
+ """
+
+ super(BaseComputeNodeDriver, self).__init__(retry_wait, max_retry_wait,
+ logging.getLogger(self.__class__.__name__),
+ type(self),
+ None)
+ self.real = self._create_driver(driver_class, **auth_kwargs)
self.list_kwargs = list_kwargs
self.create_kwargs = create_kwargs
-
- def __getattr__(self, name):
- # Proxy non-extension methods to the real driver.
- if (not name.startswith('_') and not name.startswith('ex_')
- and hasattr(self.real, name)):
- return getattr(self.real, name)
- else:
- return super(BaseComputeNodeDriver, self).__getattr__(name)
-
- def search_for(self, term, list_method, key=lambda item: item.id):
+ # Transform entries in create_kwargs. For each key K, if this class
+ # has an _init_K method, remove the entry and call _init_K with the
+ # corresponding value. If _init_K returns None, the entry stays out
+ # of the dictionary (we expect we're holding the value somewhere
+ # else, like an instance variable). Otherwise, _init_K returns a
+ # key-value tuple pair, and we add that entry to create_kwargs.
+ for key in self.create_kwargs.keys():
+ init_method = getattr(self, '_init_' + key, None)
+ if init_method is not None:
+ new_pair = init_method(self.create_kwargs.pop(key))
+ if new_pair is not None:
+ self.create_kwargs[new_pair[0]] = new_pair[1]
+
+ self._set_sizes()
+
+ def _init_ping_host(self, ping_host):
+ self.ping_host = ping_host
+
+ def _init_ssh_key(self, filename):
+ with open(filename) as ssh_file:
+ key = NodeAuthSSHKey(ssh_file.read())
+ return 'auth', key
+
+ def search_for_now(self, term, list_method, key=attrgetter('id'), **kwargs):
+ """Return one matching item from a list of cloud objects.
+
+ Raises ValueError if the number of matching objects is not exactly 1.
+
+ Arguments:
+ * term: The value that identifies a matching item.
+ * 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.
+ """
+ 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:
+ raise ValueError("{} returned {} results for {!r}".format(
+ list_method, count, term))
+ return results[0]
+
+ def search_for(self, term, list_method, key=attrgetter('id'), **kwargs):
+ """Return one cached matching item from a list of cloud objects.
+
+ See search_for_now() for details of arguments and exceptions.
+ This method caches results, so it's good to find static cloud objects
+ like node sizes, regions, etc.
+ """
cache_key = (list_method, term)
if cache_key not in self.SEARCH_CACHE:
- results = [item for item in getattr(self.real, list_method)()
- if key(item) == term]
- count = len(results)
- if count != 1:
- raise ValueError("{} returned {} results for '{}'".format(
- list_method, count, term))
- self.SEARCH_CACHE[cache_key] = results[0]
+ self.SEARCH_CACHE[cache_key] = self.search_for_now(
+ term, list_method, key, **kwargs)
return self.SEARCH_CACHE[cache_key]
- def list_nodes(self):
- return self.real.list_nodes(**self.list_kwargs)
+ def list_nodes(self, **kwargs):
+ l = self.list_kwargs.copy()
+ l.update(kwargs)
+ return self.real.list_nodes(**l)
+
+ def create_cloud_name(self, arvados_node):
+ """Return a cloud node name for the given Arvados node record.
+
+ Subclasses must override this method. It should return a string
+ that can be used as the name for a newly-created cloud node,
+ based on identifying information in the Arvados node record.
+
+ Arguments:
+ * arvados_node: This Arvados node record to seed the new cloud node.
+ """
+ raise NotImplementedError("BaseComputeNodeDriver.create_cloud_name")
+
+ def arvados_create_kwargs(self, size, arvados_node):
+ """Return dynamic keyword arguments for create_node.
- def arvados_create_kwargs(self, arvados_node):
+ Subclasses must override this method. It should return a dictionary
+ of keyword arguments to pass to the libcloud driver's create_node
+ method. These arguments will extend the static arguments in
+ create_kwargs.
+
+ Arguments:
+ * size: The node size that will be created (libcloud NodeSize object)
+ * arvados_node: The Arvados node record that will be associated
+ with this cloud node, as returned from the API server.
+ """
raise NotImplementedError("BaseComputeNodeDriver.arvados_create_kwargs")
+ def broken(self, cloud_node):
+ """Return true if libcloud has indicated the node is in a "broken" state."""
+ return False
+
+ def _make_ping_url(self, arvados_node):
+ return 'https://{}/arvados/v1/nodes/{}/ping?ping_secret={}'.format(
+ self.ping_host, arvados_node['uuid'],
+ arvados_node['info']['ping_secret'])
+
+ @staticmethod
+ def _name_key(cloud_object):
+ return cloud_object.name
+
def create_node(self, size, arvados_node):
- kwargs = self.create_kwargs.copy()
- kwargs.update(self.arvados_create_kwargs(arvados_node))
- kwargs['size'] = size
- return self.real.create_node(**kwargs)
+ try:
+ kwargs = self.create_kwargs.copy()
+ kwargs.update(self.arvados_create_kwargs(size, arvados_node))
+ kwargs['size'] = size
+ return self.real.create_node(**kwargs)
+ 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
+ # loop forever because subsequent create_node attempts will fail
+ # due to node name collision. So check if the node we intended to
+ # create shows up in the cloud node list and return it if found.
+ try:
+ return self.search_for_now(kwargs['name'], 'list_nodes', self._name_key)
+ except ValueError:
+ raise create_error
def post_create_node(self, cloud_node):
# ComputeNodeSetupActor calls this method after the cloud node is
# information back to the cloud node appropriately.
raise NotImplementedError("BaseComputeNodeDriver.sync_node")
+ @classmethod
+ def node_fqdn(cls, node):
+ # This method should return the FQDN of the node object argument.
+ # Different clouds store this in different places.
+ raise NotImplementedError("BaseComputeNodeDriver.node_fqdn")
+
@classmethod
def node_start_time(cls, node):
+ # This method should return the time the node was started, in
+ # seconds since the epoch UTC.
raise NotImplementedError("BaseComputeNodeDriver.node_start_time")
+ def destroy_node(self, cloud_node):
+ try:
+ return self.real.destroy_node(cloud_node)
+ except 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):
+ return property(
+ lambda self: getattr(self.real, attr_name),
+ lambda self, value: setattr(self.real, attr_name, value),
+ doc=getattr(getattr(NodeDriver, attr_name), '__doc__', None))
+
+ # node id
@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
- getattr(exception, '__class__', None) is Exception)
+ def node_id(cls):
+ raise NotImplementedError("BaseComputeNodeDriver.node_id")
+
+ _locals = locals()
+ for _attr_name in dir(NodeDriver):
+ if (not _attr_name.startswith('_')) and (_attr_name not in _locals):
+ _locals[_attr_name] = _delegate_to_real(_attr_name)
projects / arvados.git / blobdiff