3 from __future__ import absolute_import, print_function
6 from operator import attrgetter
8 import libcloud.common.types as cloud_types
9 from libcloud.compute.base import NodeDriver, NodeAuthSSHKey
11 from ...config import NETWORK_ERRORS
14 class BaseComputeNodeDriver(object):
15 """Abstract base class for compute node drivers.
17 libcloud drivers abstract away many of the differences between
18 cloud providers, but managing compute nodes requires some
19 cloud-specific features (e.g., keeping track of node FQDNs and
20 boot times). Compute node drivers are responsible for translating
21 the node manager's cloud requests to a specific cloud's
24 Subclasses must implement arvados_create_kwargs, sync_node,
25 node_fqdn, and node_start_time.
27 CLOUD_ERRORS = NETWORK_ERRORS + (cloud_types.LibcloudError,)
30 def _create_driver(self, driver_class, **auth_kwargs):
31 return driver_class(**auth_kwargs)
33 def __init__(self, auth_kwargs, list_kwargs, create_kwargs,
34 driver_class, retry_wait=1, max_retry_wait=180):
35 """Base initializer for compute node drivers.
38 * auth_kwargs: A dictionary of arguments that are passed into the
39 driver_class constructor to instantiate a libcloud driver.
40 * list_kwargs: A dictionary of arguments that are passed to the
41 libcloud driver's list_nodes method to return the list of compute
43 * create_kwargs: A dictionary of arguments that are passed to the
44 libcloud driver's create_node method to create a new compute node.
45 * driver_class: The class of a libcloud driver to use.
47 self.min_retry_wait = retry_wait
48 self.max_retry_wait = max_retry_wait
49 self.retry_wait = retry_wait
50 self._cloud = type(self)
51 self._logger = logging.getLogger(str(self._cloud))
53 self.real = self._create_driver(driver_class, **auth_kwargs)
54 self.list_kwargs = list_kwargs
55 self.create_kwargs = create_kwargs
56 # Transform entries in create_kwargs. For each key K, if this class
57 # has an _init_K method, remove the entry and call _init_K with the
58 # corresponding value. If _init_K returns None, the entry stays out
59 # of the dictionary (we expect we're holding the value somewhere
60 # else, like an instance variable). Otherwise, _init_K returns a
61 # key-value tuple pair, and we add that entry to create_kwargs.
62 for key in self.create_kwargs.keys():
63 init_method = getattr(self, '_init_' + key, None)
64 if init_method is not None:
65 new_pair = init_method(self.create_kwargs.pop(key))
66 if new_pair is not None:
67 self.create_kwargs[new_pair[0]] = new_pair[1]
69 self.sizes = {sz.id: sz for sz in self.real.list_sizes()}
71 def _init_ping_host(self, ping_host):
72 self.ping_host = ping_host
74 def _init_ssh_key(self, filename):
75 with open(filename) as ssh_file:
76 key = NodeAuthSSHKey(ssh_file.read())
79 def search_for(self, term, list_method, key=attrgetter('id'), **kwargs):
80 """Return one matching item from a list of cloud objects.
82 Raises ValueError if the number of matching objects is not exactly 1.
85 * term: The value that identifies a matching item.
86 * list_method: A string that names the method to call on this
87 instance's libcloud driver for a list of objects.
88 * key: A function that accepts a cloud object and returns a
89 value search for a `term` match on each item. Returns the
90 object's 'id' attribute by default.
92 cache_key = (list_method, term)
93 if cache_key not in self.SEARCH_CACHE:
94 items = getattr(self.real, list_method)(**kwargs)
95 results = [item for item in items
99 raise ValueError("{} returned {} results for '{}'".format(
100 list_method, count, term))
101 self.SEARCH_CACHE[cache_key] = results[0]
102 return self.SEARCH_CACHE[cache_key]
104 def list_nodes(self):
105 return self.real.list_nodes(**self.list_kwargs)
107 def arvados_create_kwargs(self, size, arvados_node):
108 """Return dynamic keyword arguments for create_node.
110 Subclasses must override this method. It should return a dictionary
111 of keyword arguments to pass to the libcloud driver's create_node
112 method. These arguments will extend the static arguments in
116 * size: The node size that will be created (libcloud NodeSize object)
117 * arvados_node: The Arvados node record that will be associated
118 with this cloud node, as returned from the API server.
120 raise NotImplementedError("BaseComputeNodeDriver.arvados_create_kwargs")
122 def broken(self, cloud_node):
123 """Return true if libcloud has indicated the node is in a "broken" state."""
126 def _make_ping_url(self, arvados_node):
127 return 'https://{}/arvados/v1/nodes/{}/ping?ping_secret={}'.format(
128 self.ping_host, arvados_node['uuid'],
129 arvados_node['info']['ping_secret'])
131 def create_node(self, size, arvados_node):
132 kwargs = self.create_kwargs.copy()
133 kwargs.update(self.arvados_create_kwargs(size, arvados_node))
134 kwargs['size'] = size
135 return self.real.create_node(**kwargs)
137 def post_create_node(self, cloud_node):
138 # ComputeNodeSetupActor calls this method after the cloud node is
139 # created. Any setup tasks that need to happen afterward (e.g.,
140 # tagging) should be done in this method.
143 def sync_node(self, cloud_node, arvados_node):
144 # When a compute node first pings the API server, the API server
145 # will automatically assign some attributes on the corresponding
146 # node record, like hostname. This method should propagate that
147 # information back to the cloud node appropriately.
148 raise NotImplementedError("BaseComputeNodeDriver.sync_node")
151 def node_fqdn(cls, node):
152 # This method should return the FQDN of the node object argument.
153 # Different clouds store this in different places.
154 raise NotImplementedError("BaseComputeNodeDriver.node_fqdn")
157 def node_start_time(cls, node):
158 # This method should return the time the node was started, in
159 # seconds since the epoch UTC.
160 raise NotImplementedError("BaseComputeNodeDriver.node_start_time")
163 def is_cloud_exception(cls, exception):
164 # libcloud compute drivers typically raise bare Exceptions to
165 # represent API errors. Return True for any exception that is
166 # exactly an Exception, or a better-known higher-level exception.
167 return (isinstance(exception, cls.CLOUD_ERRORS) or
168 type(exception) is Exception)
170 # Now that we've defined all our own methods, delegate generic, public
171 # attributes of libcloud drivers that we haven't defined ourselves.
172 def _delegate_to_real(attr_name):
174 lambda self: getattr(self.real, attr_name),
175 lambda self, value: setattr(self.real, attr_name, value),
176 doc=getattr(getattr(NodeDriver, attr_name), '__doc__', None))
179 for _attr_name in dir(NodeDriver):
180 if (not _attr_name.startswith('_')) and (_attr_name not in _locals):
181 _locals[_attr_name] = _delegate_to_real(_attr_name)