コード例 #1
0
ファイル: manager.py プロジェクト: renzok/ferry
class DockerManager(object):
    SSH_PORT = 22

    def __init__(self):
        # Image names
        self.DOCKER_GLUSTER = DEFAULT_DOCKER_REPO + '/gluster'
        self.DOCKER_HADOOP = DEFAULT_DOCKER_REPO + '/hadoop'
        self.DOCKER_HADOOP_CLIENT = DEFAULT_DOCKER_REPO + '/hadoop-client'
        self.DOCKER_HIVE = DEFAULT_DOCKER_REPO + '/hive-metastore'
        self.DOCKER_CASSANDRA = DEFAULT_DOCKER_REPO + '/cassandra'
        self.DOCKER_CASSANDRA_CLIENT = DEFAULT_DOCKER_REPO + '/cassandra-client'
        self.DOCKER_TITAN = DEFAULT_DOCKER_REPO + '/titan'
        self.DOCKER_MPI = DEFAULT_DOCKER_REPO + '/openmpi'
        self.DOCKER_MPI_CLIENT = DEFAULT_DOCKER_REPO + '/openmpi'

        # Generate configuration.
        self.config = ConfigFactory()

        # Docker tools
        self.docker = DockerFabric()
        self.deploy = DeployEngine(self.docker)

        # Initialize the state. 
        self._init_state_db()

    """
    Contact the state database. 
    """
    def _init_state_db(self):
        self.mongo = MongoClient(os.environ['MONGODB'], 27017, connectTimeoutMS=6000)

        self.cluster_collection = self.mongo['state']['clusters']
        self.service_collection = self.mongo['state']['services']
        self.snapshot_collection = self.mongo['state']['snapshots']

    def _serialize_containers(self, containers):
        info = []
        for c in containers:
            info.append(c.json())
        return info

    """
    Update the service configuration. 
    """
    def _update_service_configuration(self, service_uuid, service_info):
        service = self.service_collection.find_one( {'uuid':service_uuid} )
        if not service:
            self.service_collection.insert( service_info )
        else:
            self.service_collection.update( {'uuid' : service_uuid},
                                            {'$set': service_info} )
    """
    Get the storage information. 
    """
    def _get_service_configuration(self, service_uuid, detailed=False):
        info = self.service_collection.find_one( {'uuid':service_uuid}, {'_id':False} )
        if detailed:
            return info
        else:
            return info['entry']

    def _get_inspect_info(self, service_uuid):
        raw_info = self._get_service_configuration(service_uuid, detailed=True)
        json_reply = {'uuid' : service_uuid}

        # Get individual container information
        json_reply['containers'] = []
        for c in raw_info['containers']:
            json_reply['containers'].append(c)

        # Now get the entry information
        json_reply['entry'] = raw_info['entry']

        # Check if this service has a user-defined
        # unique name.
        if 'uniq' in raw_info:
            json_reply['uniq'] = raw_info['uniq']
        else:
            json_reply['uniq'] = None

        return json_reply

    def _get_snapshot_info(self, stack_uuid):
        v = self.cluster_collection.find_one( {'uuid' : stack_uuid} )
        s = self.snapshot_collection.find_one( {'snapshot_uuid':v['snapshot_uuid']} )
        if s:
            time = s['snapshot_ts'].strftime("%m/%w/%Y (%I:%M %p)")
            return { 'snapshot_ts' : time,
                     'snapshot_uuid' : v['snapshot_uuid'] }

    def _get_service(self, service_type):
        service = None
        if service_type == 'mpi':
            service = self.config.mpi
        elif service_type == 'yarn':
            service = self.config.yarn
        elif service_type == 'gluster':
            service = self.config.gluster
        elif service_type == 'cassandra':
            service = self.config.cassandra
        elif service_type == 'titan':
            service = self.config.titan
        elif service_type == 'hadoop':
            service = self.config.hadoop
        elif service_type == 'hive':
            service = self.config.hadoop
        elif service_type == 'hadoop-client':
            service = self.config.hadoop_client
        elif service_type == 'cassandra-client':
            service = self.config.cass_client
        elif service_type == 'mpi-client':
            service = self.config.mpi_client
        else:
            logging.error("unknown service " + service_type)
        return service

    """
    Helper method to copy directories. shutil fails if the 
    destination already exists. 
    """
    def _copytree(src, dst):
        for item in os.listdir(src):
            s = os.path.join(src, item)
            d = os.path.join(dst, item)
            if os.path.isdir(s):
                shutil.copytree(s, d)
            else:
                shutil.copy2(s, d)

    def _copy_instance_logs(self, instance, to_dir):
        service = self._get_service(instance.service_type)
        log_dir = service.container_log_dir

        # We're performing a reverse lookup. 
        for d in instance.volumes.keys():
            if instance.volumes[d] == log_dir:
                self._copytree(d, to_dir)
                return

    def copy_logs(self, stack_uuid, to_dir):
        connectors, compute, storage = self._get_cluster_instances(stack_uuid)
        storage_dir = to_dir + '/' + stack_uuid + '/storage'
        compute_dir = to_dir + '/' + stack_uuid + '/compute'
        connector_dir = to_dir + '/' + stack_uuid + '/connectors'

        for c in connectors:
            for i in c['instances']:
                self._copy_instance_logs(i, connector_dir)
        for c in compute:
            for i in c['instances']:
                self._copy_instance_logs(i, compute_dir)
        for c in storage:
            for i in c['instances']:
                self._copy_instance_logs(i, storage_dir)

    """
    Inspect a deployed stack. 
    """
    def inspect_deployed(self, uuid, registry):
        json_reply = {}

        # Need to inspect the registry to make sure
        # that all the images are available. 

        return json.dumps(json_reply, 
                          sort_keys=True,
                          indent=2,
                          separators=(',',':'))

    """
    Inspect a running stack. 
    """
    def inspect_stack(self, stack_uuid):
        json_reply = {}

        # Get the collection of all backends and connector UUIDS.
        cluster = self.cluster_collection.find_one( {'uuid': stack_uuid} )
        connector_uuids = cluster['connectors']
        storage_uuids = []
        compute_uuids = []
        for b in cluster['backends']['uuids']:
            if b['storage'] != None:
                storage_uuids.append(b['storage'])

            if b['compute'] != None:
                for c in b['compute']:
                    compute_uuids.append(c)

        # For each UUID, collect the detailed service information. 
        json_reply['connectors'] = []            
        for uuid in connector_uuids:
            json_reply['connectors'].append(self._get_inspect_info(uuid))

        json_reply['storage'] = []
        for uuid in storage_uuids:
            json_reply['storage'].append(self._get_inspect_info(uuid))

        json_reply['compute'] = []
        for uuid in compute_uuids:
            json_reply['compute'].append(self._get_inspect_info(uuid))

        # Now append some snapshot info. 
        json_reply['snapshots'] = self._get_snapshot_info(stack_uuid)

        return json.dumps(json_reply, 
                          sort_keys=True,
                          indent=2,
                          separators=(',',':'))

    """
    Query the available snapshots. 
    """
    def query_snapshots(self, constraints=None):
        json_reply = {}

        values = self.snapshot_collection.find()
        for v in values:
            c = self.cluster_collection.find_one( {'uuid':v['cluster_uuid']} )
            time = v['snapshot_ts'].strftime("%m/%w/%Y (%I:%M %p)")
            json_reply[v['snapshot_uuid']] = { 'uuid' : v['snapshot_uuid'],
                                               'base' : c['base'], 
                                               'snapshot_ts' : time }
        return json.dumps(json_reply, 
                          sort_keys=True,
                          indent=2,
                          separators=(',',':'))
    
    """
    Query the available stacks. 
    """
    def query_stacks(self, constraints=None):
        json_reply = {}

        if constraints:
            values = self.cluster_collection.find(constraints)
        else:
            values = self.cluster_collection.find()

        for v in values:
            time = ''
            s = self.snapshot_collection.find_one( {'snapshot_uuid':v['snapshot_uuid']} )
            if s:
                time = v['ts'].strftime("%m/%w/%Y (%I:%M %p)")
            json_reply[v['uuid']] = { 'uuid' : v['uuid'],
                                      'base' : v['base'], 
                                      'ts' : time,
                                      'backends' : v['backends']['uuids'],
                                      'connectors': v['connectors'],
                                      'status' : v['status']}
        return json.dumps(json_reply, 
                          sort_keys=True,
                          indent=2,
                          separators=(',',':'))

    """
    Query the deployed applications. 
    """
    def query_deployed(self, conf=None):
        json_reply = {}

        cursors = self.deploy.find(conf=conf)
        for c in cursors:
            for v in c:
                time = v['ts'].strftime("%m/%w/%Y (%I:%M %p)")
                c = self.cluster_collection.find_one( {'uuid':v['cluster_uuid']} )
                
                json_reply[v['uuid']] = { 'uuid' : v['uuid'],
                                          'base' : c['base'], 
                                          'ts' : time,
                                          'backends' : c['backends']['uuids'],
                                          'connectors': c['connectors'],
                                          'status': 'deployed' }
        return json.dumps(json_reply, 
                          sort_keys=True,
                          indent=2,
                          separators=(',',':'))

    """
    Allocate new UUIDs. 
    """
    def _new_service_uuid(self):
        services = self.service_collection.find()
        return "se-" + str(services.count())

    def _new_stack_uuid(self):
        clusters = self.cluster_collection.find()
        return "sa-" + str(clusters.count())

    def _new_snapshot_uuid(self, cluster_uuid):
        return "sn-%s-%s" % (cluster_uuid, str(uuid.uuid4()))

    """
    Determine if the supplied UUID is a valid snapshot. 
    """
    def is_snapshot(self, snapshot_uuid):
        v = self.snapshot_collection.find_one( {'snapshot_uuid':snapshot_uuid} )
        if v:
            return True
        else:
            return False

    """
    Check if the UUID is of a stopped application. 
    """
    def is_stopped(self, uuid, conf=None):
        cluster = self.cluster_collection.find_one( {'uuid':uuid} )
        if cluster:
            return cluster['status'] == 'stopped'
        return False

    """
    Check if the UUID is of a stopped application. 
    """
    def is_removed(self, uuid, conf=None):
        cluster = self.cluster_collection.find_one( {'uuid':uuid} )
        if cluster:
            return cluster['status'] == 'removed'
        return False

    """
    Check if the UUID is of a deployed application. 
    """
    def is_deployed(self, uuid, conf=None):
        v = self.deploy.find( one=True, 
                              spec = {'uuid':uuid},
                              conf = conf )
        if v:
            return True
        else:
            return False

    """
    Get the base image of this cluster. 
    """
    def get_base_image(self, uuid):
        cluster = self.cluster_collection.find_one( {'uuid':uuid} )
        if cluster:
            return cluster['base']
        return None

    """
    Create a new data directory
    """
    def _new_data_dir(self, service_uuid, storage_type, storage_id):
        # First check if this data directory already exists. If so,
        # go ahead and delete it (this will hopefully get rid of all xattr stuff)
        new_dir = 'tmp/%s/data_%s' % (service_uuid, storage_type + '_' + str(storage_id))
        return self._create_dir(new_dir, replace=True)

    """
    Create a new log directory
    """
    def _new_log_dir(self, service_uuid, storage_type, storage_id, replace = False):
        # First check if this data directory already exists. If so,
        # go ahead and delete it (this will hopefully get rid of all xattr stuff)
        new_dir = 'tmp/%s/log_%s' % (service_uuid, storage_type + '_' + str(storage_id))
        return self._create_dir(new_dir, replace=replace)

    def _create_dir(self, new_dir, replace=False):
        # See if we need to delete an existing data dir.
        if os.path.exists(new_dir) and replace:
            logging.warning("deleting dir " + new_dir)
            shutil.rmtree(new_dir)

        try:
            # Now create the new directory and assign
            # the right permissions. 
            sh.mkdir('-p', new_dir)
        except:
            logging.warning(new_dir + " already exists")

        try:
            uid, gid = ferry.install._get_ferry_user()
            os.chown(new_dir, uid, gid)
            os.chmod(new_dir, 0774)
        except OSError as e:
            logging.warning("could not change permissions for " + new_dir)

        return os.path.abspath(new_dir)

    def _get_service_environment(self,
                                 service, 
                                 instance, 
                                 num_instances):
        container_dir = service.container_data_dir
        log_dir = service.container_log_dir
        host_name = service.new_host_name(instance)
        ports = service.get_necessary_ports(num_instances)
        exposed = service.get_exposed_ports(num_instances)
        
        # Add SSH port for management purposes. 
        exposed.append(DockerManager.SSH_PORT)

        return container_dir, log_dir, host_name, ports, exposed
    """
    Prepare the environment for storage containers.
    """
    def _prepare_storage_environment(self, 
                                     service_uuid, 
                                     num_instances, 
                                     storage_type, 
                                     layers,
                                     args = None,
                                     replace = False):
        # Generate the data volumes. This basically defines which
        # directories on the host get mounted in the container. 
        ports = []
        exposed = []
        plan = {'localhost':{'containers':[]}}

        # Get the actual number of containers needed. 
        if storage_type == 'gluster':
            instances = self.config.gluster.get_total_instances(num_instances, layers)
        elif storage_type == 'cassandra':
            instances = self.config.cassandra.get_total_instances(num_instances, layers)
        elif storage_type == 'hadoop':
            instances = self.config.hadoop.get_total_instances(num_instances, layers)

        # Now get the new container-specific information. 
        i = 0
        for t in instances:
            instance_type = self._get_instance_image(t)
            service = self._get_service(t)
            container_dir, log_dir, host_name, ports, exposed = self._get_service_environment(service, i, num_instances)
            new_log_dir = self._new_log_dir(service_uuid, t, i, replace=replace)
            dir_info = { new_log_dir : log_dir }

            # Only use a data directory mapping if we're not
            # using BTRFS (this is to get around the xattr problem). 
            if self.docker.get_fs_type() != "btrfs":
                new_data_dir = self._new_data_dir(service_uuid, t, i)
                dir_info[new_data_dir] = container_dir

            container_info = {'image':instance_type,
                              'type':t, 
                              'volumes':dir_info,
                              'volume_user':DEFAULT_FERRY_OWNER, 
                              'ports':ports,
                              'exposed':exposed, 
                              'hostname':host_name,
                              'args':args}
            plan['localhost']['containers'].append(container_info)
            i += 1

        return plan

    """
    Prepare the environment for compute containers.
    """
    def _prepare_compute_environment(self, 
                                     service_uuid, 
                                     num_instances, 
                                     compute_type,
                                     layers, 
                                     args = None):
        # Generate the data volumes. This basically defines which
        # directories on the host get mounted in the container. 
        ports = []
        exposed = []
        instance_type = ''
        plan = {'localhost':{'containers':[]}}

        # Get the actual number of containers needed. 
        if compute_type == 'yarn':
            instances = self.config.yarn.get_total_instances(num_instances, layers)
        elif compute_type == 'mpi':
            instances = self.config.mpi.get_total_instances(num_instances, layers)

        i = 0
        for t in instances:
            instance_type = self._get_instance_image(t)
            service = self._get_service(t)
            container_dir, log_dir, host_name, ports, exposed = self._get_service_environment(service, i, num_instances)
            new_log_dir = self._new_log_dir(service_uuid, t, i)
            dir_info = { new_log_dir : log_dir }
            container_info = {'image':instance_type,
                              'volumes':dir_info,
                              'volume_user':DEFAULT_FERRY_OWNER, 
                              'type':t, 
                              'ports':ports,
                              'exposed':exposed, 
                              'hostname':host_name,
                              'args':args}
            plan['localhost']['containers'].append(container_info)
            i += 1
        return plan

    """
    Fetch the instance type. If the UUID is not associated with a running
    service, then just use the raw image. Otherwise, look for a snapshot image. 
    """
    def _get_instance_image(self, instance_type, uuid=None):
        image = None
        if instance_type == 'hadoop-client':
            image = self.DOCKER_HADOOP_CLIENT
        elif instance_type == 'cassandra-client':
            image = self.DOCKER_CASSANDRA_CLIENT
        elif instance_type == 'mpi-client':
            image = self.DOCKER_MPI_CLIENT
        elif instance_type == 'gluster':
            image = self.DOCKER_GLUSTER
        elif instance_type == 'cassandra':
            image = self.DOCKER_CASSANDRA
        elif instance_type == 'titan':
            image = self.DOCKER_TITAN
        elif instance_type == 'hadoop':
            image = self.DOCKER_HADOOP
        elif instance_type == 'hive':
            image = self.DOCKER_HIVE
        elif instance_type == 'mpi':
            image = self.DOCKER_MPI
        elif instance_type == 'yarn':
            image = self.DOCKER_HADOOP

        return image

    """
    Prepare the environment for connector containers.
    """
    def _prepare_connector_environment(self, 
                                       service_uuid, 
                                       connector_type, 
                                       instance_type=None,
                                       name=None, 
                                       args=None):
        ports = []
        exposed = []
        plan = {'localhost':{'containers':[]}}

        # Determine the instance type from the connector type. 
        if not instance_type:
            instance_type = self._get_instance_image(connector_type)

        service = self._get_service(connector_type)
        container_dir, log_dir, host_name, ports, exposed = self._get_service_environment(service, 0, 1)
        new_log_dir = self._new_log_dir(service_uuid, connector_type, 0)
        dir_info = { new_log_dir : log_dir }
        container_info = { 'image':instance_type,
                           'volumes':dir_info,
                           'volume_user':DEFAULT_FERRY_OWNER, 
                           'type':connector_type, 
                           'ports':ports,
                           'exposed':exposed, 
                           'hostname':host_name,
                           'name':name, 
                           'args':args}
        plan['localhost']['containers'].append(container_info)
        return plan

    """
    Transfer the configuration to the containers. 
    """
    def _transfer_config(self, config_dirs):
        for c in config_dirs:
            container = c[0]
            from_dir = c[1]
            to_dir = c[2]
            self.docker.copy([container], from_dir, to_dir)

    """
    Transfer these environment variables to the containers.
    Since the user normally interacts with these containers by 
    logging in (via ssh), we must place these variables in the profile. 
    """
    def _transfer_env_vars(self, containers, env_vars):
        for k in env_vars.keys():
            self.docker.cmd(containers, 
                            "echo export %s=%s >> /etc/profile" % (k, env_vars[k]))
    """
    Start the containers on the specified environment
    """
    def _start_containers(self, plan):
        return self.docker.alloc(plan['localhost']['containers']);

    """
    Restart the stopped containers. 
    """
    def _restart_containers(self, container_info):
        return self.docker.restart(container_info)

    """
    Register the set of services under a single cluster identifier. 
    """
    def register_stack(self, backends, connectors, base, uuid=None):
        if not uuid:
            cluster_uuid = self._new_stack_uuid()
        else:
            cluster_uuid = uuid

        ts = datetime.datetime.now()
        cluster = { 'uuid' : cluster_uuid,
                    'backends':backends,
                    'connectors':connectors,
                    'num_snapshots':0,
                    'snapshot_ts':'', 
                    'snapshot_uuid':base, 
                    'base':base,
                    'status': 'running',
                    'ts':ts }

        if not uuid:
            self.cluster_collection.insert( cluster )
        else:
            self._update_stack(uuid, cluster)

        return cluster_uuid

    """
    Helper method to update a cluster's status. 
    """
    def _update_stack(self, cluster_uuid, state):
        self.cluster_collection.update( {'uuid' : cluster_uuid},
                                        {'$set' : state} )

    def _get_cluster_instances(self, cluster_uuid):
        all_connectors = []
        all_storage = []
        all_compute = []
        cluster = self.cluster_collection.find_one( {'uuid':cluster_uuid} )
        if cluster:
            backends = cluster['backends']
            connector_uuids = cluster['connectors']
            for c in connector_uuids:
                connectors = {'uuid' : c,
                              'instances' : []}
                connector_info = self._get_service_configuration(c, detailed=True)
                for connector in connector_info['containers']:
                    connector_instance = DockerInstance(connector)
                    connectors['instances'].append(connector_instance)
                    connectors['type'] = connector_instance.service_type
                all_connectors.append(connectors)

            # Collect all the UUIDs of the backend containers. 
            # and stop them. The backend is considered ephemeral!
            for b in backends['uuids']:
                if b['storage'] != None:
                    storage = {'uuid' : b['storage'],
                               'instances' : []}
                    storage_info = self._get_service_configuration(b['storage'], detailed=True)
                    for s in storage_info['containers']:
                        storage_instance = DockerInstance(s)
                        storage['instances'].append(storage_instance)
                        storage['type'] = storage_instance.service_type
                    all_storage.append(storage)

                if b['compute'] != None:
                    for c in b['compute']:
                        compute = {'uuid' : c,
                                   'instances' : []}
                        compute_info = self._get_service_configuration(c, detailed=True)
                        for container in compute_info['containers']:
                            compute_instance = DockerInstance(container)
                            compute['instances'].append(compute_instance)
                            compute['type'] = compute_instance.service_type
                        all_compute.append(compute)
            return all_connectors, all_compute, all_storage

    """
    Stop a running cluster.
    """
    def _stop_stack(self, cluster_uuid):
        connectors, compute, storage = self._get_cluster_instances(cluster_uuid)
        for c in connectors:
            self._stop_service(c['uuid'], c['instances'], c['type'])
        for c in compute:
            self._stop_service(c['uuid'], c['instances'], c['type'])
        for s in storage:
            self._stop_service(s['uuid'], s['instances'], s['type'])

    def _purge_stack(self, cluster_uuid):
        volumes = []
        connectors, compute, storage = self._get_cluster_instances(cluster_uuid)
        for c in connectors:
            self.docker.remove(c['instances'])
        for c in compute:
            self.docker.remove(c['instances'])
        for s in storage:
            for i in s['instances']:
                for v in i.volumes.keys():
                    volumes.append(v)
            self.docker.remove(s['instances'])

        # Now remove the data directories. 
        for v in volumes:
            shutil.rmtree(v)
    
    """
    Take a snapshot of an existing stack. 
    """
    def _snapshot_stack(self, cluster_uuid):
        cluster = self.cluster_collection.find_one( {'uuid':cluster_uuid} )
        if cluster:
            # We need to deserialize the docker containers from the cluster/service
            # description so that the snapshot code has access to certain pieces
            # of information (service type, etc.). 
            connectors = []
            connector_uuids = cluster['connectors']
            for c in connector_uuids:
                connector_info = self._get_service_configuration(c, detailed=True)
                connectors.append(DockerInstance(connector_info['containers'][0]))
            cs_snapshots = self.docker.snapshot(connectors, 
                                                cluster_uuid, 
                                                cluster['num_snapshots'])

            # Register the snapshot in the snapshot state. 
            snapshot_uuid = self._new_snapshot_uuid(cluster_uuid)
            snapshot_ts = datetime.datetime.now()
            snapshot_state = { 'snapshot_ts' : snapshot_ts, 
                               'snapshot_uuid' : snapshot_uuid,
                               'snapshot_cs' : cs_snapshots,
                               'cluster_uuid' : cluster_uuid}
            self.snapshot_collection.insert( snapshot_state )

            # Now update the cluster state. 
            cluster_state = { 'num_snapshots' : cluster['num_snapshots'] + 1,
                              'snapshot_uuid' : snapshot_uuid }
            self.cluster_collection.update( {'uuid':cluster_uuid}, 
                                             {"$set": cluster_state } )

    """
    Allocate a new compute cluster.
    """
    def allocate_compute(self,
                         compute_type, 
                         storage_uuid,
                         args, 
                         num_instances=1,
                         layers=[]):
        # Allocate a UUID.
        service_uuid = self._new_service_uuid()
        service = self._get_service(compute_type)

        # Generate the data volumes. This basically defines which
        # directories on the host get mounted in the container. 
        plan = self._prepare_compute_environment(service_uuid, num_instances, compute_type, layers, args)

        # Get the entry point for the storage layer. 
        storage_entry = self._get_service_configuration(storage_uuid)

        # Allocate all the containers. 
        containers = self._start_containers(plan)

        # Generate a configuration dir.
        config_dirs, entry_point = self.config.generate_compute_configuration(service_uuid, 
                                                                              containers, 
                                                                              service, 
                                                                              args, 
                                                                              [storage_entry])

        # Now copy over the configuration.
        self._transfer_config(config_dirs)

        container_info = self._serialize_containers(containers)
        service = {'uuid':service_uuid, 
                   'containers':container_info, 
                   'class':'compute',
                   'type':compute_type,
                   'entry':entry_point,
                   'storage':storage_uuid, 
                   'status':'running'}
        self._update_service_configuration(service_uuid, service)

        # After the docker instance start, we need to start the
        # actual storage service (gluster, etc.). 
        self._start_service(service_uuid, containers, compute_type)
        return service_uuid
        
    def _start_service(self,
                       uuid,
                       containers,
                       service_type):
        entry_point = self._get_service_configuration(uuid)
        service = self._get_service(service_type)
        service.start_service(containers, entry_point, self.docker)

    def _stop_service(self,
                      uuid,
                      containers,
                      service_type):
        entry_point = self._get_service_configuration(uuid)
        service = self._get_service(service_type)
        service.stop_service(containers, entry_point, self.docker)

    """
    Create a storage cluster and start a particular
    personality on that cluster. 
    """
    def allocate_storage(self, 
                         storage_type, 
                         num_instances=1,
                         layers=[], 
                         args=None,
                         replace=False):
        # Allocate a UUID.
        service_uuid = self._new_service_uuid()
        service = self._get_service(storage_type)

        # Generate the data volumes. This basically defines which
        # directories on the host get mounted in the container. 
        plan = self._prepare_storage_environment(service_uuid, num_instances, storage_type, layers, args, replace)

        # Allocate all the containers. 
        containers = self._start_containers(plan)

        # Generate a configuration dir.
        config_dirs, entry_point = self.config.generate_storage_configuration(service_uuid, 
                                                                              containers, 
                                                                              service, 
                                                                              args)

        # Now copy over the configuration.
        self._transfer_config(config_dirs)

        container_info = self._serialize_containers(containers)
        service = {'uuid':service_uuid, 
                   'containers':container_info, 
                   'class':'storage',
                   'type':storage_type,
                   'entry':entry_point,
                   'status':'running'}
        self._update_service_configuration(service_uuid, service)

        # After the docker instance start, we need to start the
        # actual storage service (gluster, etc.). 
        self._start_service(service_uuid, containers, storage_type)
        return service_uuid

    """
    Get the default deployment conf file. 
    """
    def _get_default_conf(self):        
        return FERRY_HOME + '/data/conf/deploy_default.json'

    """
    Get the deployment configuration parameters. 
    """
    def _get_deploy_params(self, mode, conf):
        # First just find and read the configuration file. 
        if conf == 'default':
            conf = self._get_default_conf()

        # Read the configuration file.
        if os.path.exists(conf):
            f = open(conf, 'r').read()
            j = json.loads(f)

            # Now find the right configuration.
            if mode in j:
                j[mode]['_mode'] = mode
                return j[mode]

        return None

    """
    Deploy an existing stack. 
    """
    def deploy_stack(self, cluster_uuid, params=None):
        containers = []
        cluster = self.cluster_collection.find_one( {'uuid':cluster_uuid} )
        if cluster:
            connector_uuids = cluster['connectors']
            for c in connector_uuids:
                connector_info = self._get_service_configuration(c, detailed=True)
                containers.append(DockerInstance(connector_info['containers'][0]))

            self.deploy.deploy(cluster_uuid, containers, params)

            # Check if we need to try starting the
            # stack right away. 
            if params and 'start-on-create' in params:
                return True
        return False
    """
    Manage the stack.
    """
    def manage_stack(self,
                     stack_uuid,
                     action):
        status = 'running'
        if(action == 'snapshot'):        
            # The user wants to take a snapshot of the current stack. This
            # doesn't actually stop anything.
            self._snapshot_stack(stack_uuid)
        elif(action == 'stop'):
            self._stop_stack(stack_uuid)

            status = 'stopped'
            service_status = { 'uuid':stack_uuid, 'status':status }
            self._update_stack(stack_uuid, service_status)
        elif(action == 'rm'):
            # First need to check if the stack is stopped.
            if self.is_stopped(stack_uuid):
                self._purge_stack(stack_uuid)

                status = 'removed'
                service_status = { 'uuid':stack_uuid, 'status':status }
                self._update_stack(stack_uuid, service_status)
            else:
                return { 'uuid' : stack_uuid,
                         'status' : False,
                         'msg': 'Stack is running. Please stop first' }

        return { 'uuid' : stack_uuid,
                 'status' : True,
                 'msg': status }

    """
    Lookup the stopped backend info. 
    """
    def fetch_stopped_backend(self, uuid):
        service = self.cluster_collection.find_one( {'uuid':uuid} )
        if service:
            return service['backends']['backend']

    """
    Lookup the snapshot backend info. 
    """
    def fetch_snapshot_backend(self, snapshot_uuid):
        snapshot = self.cluster_collection.find_one( {'snapshot_uuid':snapshot_uuid} )
        if snapshot:
            return snapshot['backends']['backend']

    """
    Lookup the deployed backend info. 
    """
    def fetch_deployed_backend(self, app_uuid, conf=None):
        app = self.deploy.find( one = True,
                                spec = { 'uuid' : app_uuid },
                                conf = conf )
        stack = self.cluster_collection.find_one( {'uuid':app['cluster_uuid'] } )
        if stack:
            return stack['backends']['backend']

    """
    Lookup the deployed application connector info and instantiate. 
    """
    def allocate_stopped_connectors(self, 
                                     app_uuid, 
                                     backend_info,
                                     conf = None):
        connector_info = []
        cluster = self.cluster_collection.find_one( {'uuid':app_uuid} )
        if cluster:
            for cuid in cluster['connectors']:
                # Retrieve the actual service information. This will
                # contain the container ID. 
                s = self._get_service_configuration(cuid, detailed=True)
                for c in s['containers']:
                    connector_info.append(self.restart_connector(service_uuid = app_uuid,
                                                                 connector_type = c['type'],
                                                                 backend = backend_info,
                                                                 name = c['name'], 
                                                                 args = c['args'],
                                                                 container = c['container']))
        return connector_info

    """
    Lookup the deployed application connector info and instantiate. 
    """
    def allocate_deployed_connectors(self, 
                                     app_uuid, 
                                     backend_info,
                                     conf = None):
        connector_info = []
        app = self.deploy.find( one = True,
                                spec = { 'uuid' : app_uuid },
                                conf = conf)
        if app:
            for c in app['connectors']:
                connector_info.append(self.allocate_connector(connector_type = c['type'],
                                                              backend = backend_info,
                                                              name = c['name'], 
                                                              args = c['args'],
                                                              image = c['image']))
        return connector_info
                
    """
    Lookup the snapshot connector info and instantiate. 
    """
    def allocate_snapshot_connectors(self, 
                                     snapshot_uuid, 
                                     backend_info):
        connector_info = []
        snapshot = self.snapshot_collection.find_one( {'snapshot_uuid':snapshot_uuid} )
        if snapshot:
            for s in snapshot['snapshot_cs']:
                connector_info.append(self.allocate_connector(connector_type = s['type'],
                                                              backend = backend_info,
                                                              name = s['name'], 
                                                              args = s['args'],
                                                              image = s['image']))
        return connector_info
                

    """
    Restart a stopped connector with an existing storage service. 
    """
    def restart_connector(self,
                          service_uuid, 
                          connector_type, 
                          backend=None,
                          name=None, 
                          args=None,
                          container=None):
        service = self._get_service(connector_type)

        # Allocate all the containers. 
        container = self._restart_containers({ 'container': container,
                                               'type' : connector_type,
                                               'args' : args})

        # Initialize the connector and connect to the storage. 
        storage_entry = []
        compute_entry = []
        for b in backend:
            if b['storage']:
                storage_entry.append(self._get_service_configuration(b['storage']))
            if b['compute']:
                for c in b['compute']:
                    compute_entry.append(self._get_service_configuration(c))

        # Generate the environment variables that will be 
        # injected into the containers. 
        env_vars = self.config.generate_env_vars(storage_entry,
                                                 compute_entry)

        # Now generate the configuration files that will be
        # transferred to the containers. 
        config_dirs, entry_point = self.config.generate_connector_configuration(service_uuid, 
                                                                                [container], 
                                                                                service,
                                                                                storage_entry,
                                                                                compute_entry,
                                                                                args)
        # Now copy over the configuration.
        self._transfer_config(config_dirs)
        self._transfer_env_vars([container], env_vars)

        # Update the connector state. 
        container_info = self._serialize_containers([container])
        service = {'uuid':service_uuid, 
                   'containers':container_info, 
                   'class':'connector',
                   'type':connector_type,
                   'entry':entry_point,
                   'uniq': name, 
                   'status':'running'}
        self._update_service_configuration(service_uuid, service)

        # Start the connector personality. 
        self._start_service(service_uuid, [container], connector_type)
        return service_uuid
                
    """
    Allocate a new connector and associate with an existing storage service. 
    """
    def allocate_connector(self,
                           connector_type, 
                           backend=None,
                           name=None, 
                           args=None,
                           image=None):
        # Allocate a UUID.
        service_uuid = self._new_service_uuid()
        service = self._get_service(connector_type)

        # Generate the data volumes. This basically defines which
        # directories on the host get mounted in the container. 
        plan = self._prepare_connector_environment(service_uuid = service_uuid, 
                                                   connector_type = connector_type, 
                                                   instance_type = image,
                                                   name = name,
                                                   args = args)


        # Allocate all the containers. 
        containers = self._start_containers(plan)

        # Initialize the connector and connect to the storage. 
        storage_entry = []
        compute_entry = []
        if backend:
            for b in backend:
                if b['storage']:
                    storage_entry.append(self._get_service_configuration(b['storage']))
                if b['compute']:
                    for c in b['compute']:
                        compute_entry.append(self._get_service_configuration(c))

        # Generate the environment variables that will be 
        # injected into the containers. 
        env_vars = self.config.generate_env_vars(storage_entry,
                                                 compute_entry)

        # Now generate the configuration files that will be
        # transferred to the containers. 
        config_dirs, entry_point = self.config.generate_connector_configuration(service_uuid, 
                                                                                containers, 
                                                                                service,
                                                                                storage_entry,
                                                                                compute_entry,
                                                                                args)
        # Now copy over the configuration.
        self._transfer_config(config_dirs)
        self._transfer_env_vars(containers, env_vars)

        # Update the connector state. 
        container_info = self._serialize_containers(containers)
        service = {'uuid':service_uuid, 
                   'containers':container_info, 
                   'class':'connector',
                   'type':connector_type,
                   'entry':entry_point,
                   'uniq': name, 
                   'status':'running'}
        self._update_service_configuration(service_uuid, service)

        # Start the connector personality. 
        self._start_service(service_uuid, containers, connector_type)
        return service_uuid

    """
    Fetch the current docker version.
    """
    def version(self):
        return self.docker.version()