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.
"""
super(BaseComputeNodeDriver, self).__init__(retry_wait, max_retry_wait,
- logging.getLogger(str(type(self))),
+ logging.getLogger(self.__class__.__name__),
type(self),
None)
self.real = self._create_driver(driver_class, **auth_kwargs)
if new_pair is not None:
self.create_kwargs[new_pair[0]] = new_pair[1]
- self.sizes = {sz.id: sz for sz in self.real.list_sizes()}
+ self._set_sizes()
def _init_ping_host(self, ping_host):
self.ping_host = ping_host
key = NodeAuthSSHKey(ssh_file.read())
return 'auth', key
- def search_for(self, term, list_method, key=attrgetter('id'), **kwargs):
+ 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 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, list_method)(**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:
- items = getattr(self.real, list_method)(**kwargs)
- results = [item for item in items
- 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.
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(size, 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 self.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
lambda self, value: setattr(self.real, attr_name, value),
doc=getattr(getattr(NodeDriver, attr_name), '__doc__', None))
+ # node id
+ @classmethod
+ 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):
projects / arvados.git / blobdiff