nova/nova/cells/scheduler.py

259 lines
11 KiB
Python

# Copyright (c) 2012 Rackspace Hosting
# All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License"); you may
# not use this file except in compliance with the License. You may obtain
# a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
# WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
# License for the specific language governing permissions and limitations
# under the License.
"""
Cells Scheduler
"""
import copy
import time
from oslo.config import cfg
from nova.cells import filters
from nova.cells import weights
from nova import compute
from nova.compute import flavors
from nova.compute import instance_actions
from nova.compute import vm_states
from nova import conductor
from nova.db import base
from nova import exception
from nova.i18n import _LE, _LI
from nova import objects
from nova.objects import base as obj_base
from nova.openstack.common import log as logging
from nova.scheduler import utils as scheduler_utils
from nova import utils
cell_scheduler_opts = [
cfg.ListOpt('scheduler_filter_classes',
default=['nova.cells.filters.all_filters'],
help='Filter classes the cells scheduler should use. '
'An entry of "nova.cells.filters.all_filters" '
'maps to all cells filters included with nova.'),
cfg.ListOpt('scheduler_weight_classes',
default=['nova.cells.weights.all_weighers'],
help='Weigher classes the cells scheduler should use. '
'An entry of "nova.cells.weights.all_weighers" '
'maps to all cell weighers included with nova.'),
cfg.IntOpt('scheduler_retries',
default=10,
help='How many retries when no cells are available.'),
cfg.IntOpt('scheduler_retry_delay',
default=2,
help='How often to retry in seconds when no cells are '
'available.')
]
LOG = logging.getLogger(__name__)
CONF = cfg.CONF
CONF.register_opts(cell_scheduler_opts, group='cells')
class CellsScheduler(base.Base):
"""The cells scheduler."""
def __init__(self, msg_runner):
super(CellsScheduler, self).__init__()
self.msg_runner = msg_runner
self.state_manager = msg_runner.state_manager
self.compute_api = compute.API()
self.compute_task_api = conductor.ComputeTaskAPI()
self.filter_handler = filters.CellFilterHandler()
filter_classes = self.filter_handler.get_matching_classes(
CONF.cells.scheduler_filter_classes)
self.filters = [cls() for cls in filter_classes]
self.weight_handler = weights.CellWeightHandler()
weigher_classes = self.weight_handler.get_matching_classes(
CONF.cells.scheduler_weight_classes)
self.weighers = [cls() for cls in weigher_classes]
def _create_instances_here(self, ctxt, instance_uuids, instance_properties,
instance_type, image, security_groups, block_device_mapping):
instance_values = copy.copy(instance_properties)
# The parent may pass these metadata values as lists, and the
# create call expects it to be a dict.
instance_values['metadata'] = utils.instance_meta(instance_values)
sys_metadata = utils.instance_sys_meta(instance_values)
# Make sure the flavor info is set. It may not have been passed
# down.
sys_metadata = flavors.save_flavor_info(sys_metadata, instance_type)
instance_values['system_metadata'] = sys_metadata
# Pop out things that will get set properly when re-creating the
# instance record.
instance_values.pop('id')
instance_values.pop('name')
instance_values.pop('info_cache')
instance_values.pop('security_groups')
# FIXME(danms): The instance was brutally serialized before being
# sent over RPC to us. Thus, the pci_requests value wasn't really
# sent in a useful form. Since it was getting ignored for cells
# before it was part of the Instance, skip it now until cells RPC
# is sending proper instance objects.
instance_values.pop('pci_requests', None)
instances = []
num_instances = len(instance_uuids)
for i, instance_uuid in enumerate(instance_uuids):
instance = objects.Instance(context=ctxt)
instance.update(instance_values)
instance.uuid = instance_uuid
instance = self.compute_api.create_db_entry_for_new_instance(
ctxt,
instance_type,
image,
instance,
security_groups,
block_device_mapping,
num_instances, i)
instances.append(instance)
instance_p = obj_base.obj_to_primitive(instance)
self.msg_runner.instance_update_at_top(ctxt, instance_p)
return instances
def _create_action_here(self, ctxt, instance_uuids):
for instance_uuid in instance_uuids:
objects.InstanceAction.action_start(
ctxt,
instance_uuid,
instance_actions.CREATE,
want_result=False)
def _get_possible_cells(self):
cells = self.state_manager.get_child_cells()
our_cell = self.state_manager.get_my_state()
# Include our cell in the list, if we have any capacity info
if not cells or our_cell.capacities:
cells.append(our_cell)
return cells
def _grab_target_cells(self, filter_properties):
cells = self._get_possible_cells()
cells = self.filter_handler.get_filtered_objects(self.filters, cells,
filter_properties)
# NOTE(comstud): I know this reads weird, but the 'if's are nested
# this way to optimize for the common case where 'cells' is a list
# containing at least 1 entry.
if not cells:
if cells is None:
# None means to bypass further scheduling as a filter
# took care of everything.
return
raise exception.NoCellsAvailable()
weighted_cells = self.weight_handler.get_weighed_objects(
self.weighers, cells, filter_properties)
LOG.debug("Weighted cells: %(weighted_cells)s",
{'weighted_cells': weighted_cells})
target_cells = [cell.obj for cell in weighted_cells]
return target_cells
def _build_instances(self, message, target_cells, instance_uuids,
build_inst_kwargs):
"""Attempt to build instance(s) or send msg to child cell."""
ctxt = message.ctxt
instance_properties = build_inst_kwargs['instances'][0]
filter_properties = build_inst_kwargs['filter_properties']
instance_type = filter_properties['instance_type']
image = build_inst_kwargs['image']
security_groups = build_inst_kwargs['security_groups']
block_device_mapping = build_inst_kwargs['block_device_mapping']
LOG.debug("Building instances with routing_path=%(routing_path)s",
{'routing_path': message.routing_path})
for target_cell in target_cells:
try:
if target_cell.is_me:
# Need to create instance DB entries as the conductor
# expects that the instance(s) already exists.
instances = self._create_instances_here(ctxt,
instance_uuids, instance_properties, instance_type,
image, security_groups, block_device_mapping)
build_inst_kwargs['instances'] = instances
# Need to record the create action in the db as the
# conductor expects it to already exist.
self._create_action_here(ctxt, instance_uuids)
self.compute_task_api.build_instances(ctxt,
**build_inst_kwargs)
return
self.msg_runner.build_instances(ctxt, target_cell,
build_inst_kwargs)
return
except Exception:
LOG.exception(_LE("Couldn't communicate with cell '%s'"),
target_cell.name)
# FIXME(comstud): Would be nice to kick this back up so that
# the parent cell could retry, if we had a parent.
LOG.error(_LE("Couldn't communicate with any cells"))
raise exception.NoCellsAvailable()
def build_instances(self, message, build_inst_kwargs):
image = build_inst_kwargs['image']
instance_uuids = [inst['uuid'] for inst in
build_inst_kwargs['instances']]
instances = build_inst_kwargs['instances']
request_spec = scheduler_utils.build_request_spec(message.ctxt,
image, instances)
filter_properties = copy.copy(build_inst_kwargs['filter_properties'])
filter_properties.update({'context': message.ctxt,
'scheduler': self,
'routing_path': message.routing_path,
'host_sched_kwargs': build_inst_kwargs,
'request_spec': request_spec})
self._schedule_build_to_cells(message, instance_uuids,
filter_properties, self._build_instances, build_inst_kwargs)
def _schedule_build_to_cells(self, message, instance_uuids,
filter_properties, method, method_kwargs):
"""Pick a cell where we should create a new instance(s)."""
try:
for i in xrange(max(0, CONF.cells.scheduler_retries) + 1):
try:
target_cells = self._grab_target_cells(filter_properties)
if target_cells is None:
# a filter took care of scheduling. skip.
return
return method(message, target_cells, instance_uuids,
method_kwargs)
except exception.NoCellsAvailable:
if i == max(0, CONF.cells.scheduler_retries):
raise
sleep_time = max(1, CONF.cells.scheduler_retry_delay)
LOG.info(_LI("No cells available when scheduling. Will "
"retry in %(sleep_time)s second(s)"),
{'sleep_time': sleep_time})
time.sleep(sleep_time)
continue
except Exception:
LOG.exception(_LE("Error scheduling instances %(instance_uuids)s"),
{'instance_uuids': instance_uuids})
ctxt = message.ctxt
for instance_uuid in instance_uuids:
self.msg_runner.instance_update_at_top(ctxt,
{'uuid': instance_uuid,
'vm_state': vm_states.ERROR})
try:
self.db.instance_update(ctxt,
instance_uuid,
{'vm_state': vm_states.ERROR})
except Exception:
pass