e570436d1c
The RBD driver and almost all backup drivers rely heavily on eventlet's tpool, which has a default of 20 threads, which will be too low. Currently the only way to change this is using the environmental variable EVENTLET_THREADPOOL_SIZE, which isn't very clean for the openstack services. This patch adds the possibility of setting specific values for each backend and for the backup service, and increases the default for the backup service from 20 threads to 60. The backup service can be configured under [DEFAULT] section with option backup_native_threads_pool_size, and the backends under their specific sections with backend_native_threads_pool_size. Change-Id: I5d7c1e8d7f1c6592ded1f74eea42d76ab523df92 Closes-Bug: #1754354
316 lines
13 KiB
Python
316 lines
13 KiB
Python
# Copyright 2010 United States Government as represented by the
|
|
# Administrator of the National Aeronautics and Space Administration.
|
|
# 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.
|
|
|
|
"""Base Manager class.
|
|
|
|
Managers are responsible for a certain aspect of the system. It is a logical
|
|
grouping of code relating to a portion of the system. In general other
|
|
components should be using the manager to make changes to the components that
|
|
it is responsible for.
|
|
|
|
For example, other components that need to deal with volumes in some way,
|
|
should do so by calling methods on the VolumeManager instead of directly
|
|
changing fields in the database. This allows us to keep all of the code
|
|
relating to volumes in the same place.
|
|
|
|
We have adopted a basic strategy of Smart managers and dumb data, which means
|
|
rather than attaching methods to data objects, components should call manager
|
|
methods that act on the data.
|
|
|
|
Methods on managers that can be executed locally should be called directly. If
|
|
a particular method must execute on a remote host, this should be done via rpc
|
|
to the service that wraps the manager
|
|
|
|
Managers should be responsible for most of the db access, and
|
|
non-implementation specific data. Anything implementation specific that can't
|
|
be generalized should be done by the Driver.
|
|
|
|
In general, we prefer to have one manager with multiple drivers for different
|
|
implementations, but sometimes it makes sense to have multiple managers. You
|
|
can think of it this way: Abstract different overall strategies at the manager
|
|
level(FlatNetwork vs VlanNetwork), and different implementations at the driver
|
|
level(LinuxNetDriver vs CiscoNetDriver).
|
|
|
|
Managers will often provide methods for initial setup of a host or periodic
|
|
tasks to a wrapping service.
|
|
|
|
This module provides Manager, a base class for managers.
|
|
|
|
"""
|
|
|
|
|
|
from oslo_config import cfg
|
|
from oslo_log import log as logging
|
|
import oslo_messaging as messaging
|
|
from oslo_service import periodic_task
|
|
from oslo_utils import timeutils
|
|
|
|
from cinder import context
|
|
from cinder import db
|
|
from cinder.db import base
|
|
from cinder import exception
|
|
from cinder import objects
|
|
from cinder import rpc
|
|
from cinder.scheduler import rpcapi as scheduler_rpcapi
|
|
from cinder import utils
|
|
|
|
from eventlet import greenpool
|
|
from eventlet import tpool
|
|
|
|
|
|
CONF = cfg.CONF
|
|
LOG = logging.getLogger(__name__)
|
|
|
|
|
|
class PeriodicTasks(periodic_task.PeriodicTasks):
|
|
def __init__(self):
|
|
super(PeriodicTasks, self).__init__(CONF)
|
|
|
|
|
|
class Manager(base.Base, PeriodicTasks):
|
|
# Set RPC API version to 1.0 by default.
|
|
RPC_API_VERSION = '1.0'
|
|
|
|
target = messaging.Target(version=RPC_API_VERSION)
|
|
|
|
def __init__(self, host=None, db_driver=None, cluster=None, **kwargs):
|
|
if not host:
|
|
host = CONF.host
|
|
self.host = host
|
|
self.cluster = cluster
|
|
self.additional_endpoints = []
|
|
self.availability_zone = CONF.storage_availability_zone
|
|
super(Manager, self).__init__(db_driver)
|
|
|
|
def _set_tpool_size(self, nthreads):
|
|
# NOTE(geguileo): Until PR #472 is merged we have to be very careful
|
|
# not to call "tpool.execute" before calling this method.
|
|
tpool.set_num_threads(nthreads)
|
|
|
|
@property
|
|
def service_topic_queue(self):
|
|
return self.cluster or self.host
|
|
|
|
def init_host(self, service_id=None, added_to_cluster=None):
|
|
"""Handle initialization if this is a standalone service.
|
|
|
|
A hook point for services to execute tasks before the services are made
|
|
available (i.e. showing up on RPC and starting to accept RPC calls) to
|
|
other components. Child classes should override this method.
|
|
|
|
:param service_id: ID of the service where the manager is running.
|
|
:param added_to_cluster: True when a host's cluster configuration has
|
|
changed from not being defined or being '' to
|
|
any other value and the DB service record
|
|
reflects this new value.
|
|
"""
|
|
pass
|
|
|
|
def init_host_with_rpc(self):
|
|
"""A hook for service to do jobs after RPC is ready.
|
|
|
|
Like init_host(), this method is a hook where services get a chance
|
|
to execute tasks that *need* RPC. Child classes should override
|
|
this method.
|
|
|
|
"""
|
|
pass
|
|
|
|
def is_working(self):
|
|
"""Method indicating if service is working correctly.
|
|
|
|
This method is supposed to be overridden by subclasses and return if
|
|
manager is working correctly.
|
|
"""
|
|
return True
|
|
|
|
def reset(self):
|
|
"""Method executed when SIGHUP is caught by the process.
|
|
|
|
We're utilizing it to reset RPC API version pins to avoid restart of
|
|
the service when rolling upgrade is completed.
|
|
"""
|
|
LOG.info('Resetting cached RPC version pins.')
|
|
rpc.LAST_OBJ_VERSIONS = {}
|
|
rpc.LAST_RPC_VERSIONS = {}
|
|
|
|
def set_log_levels(self, context, log_request):
|
|
utils.set_log_levels(log_request.prefix, log_request.level)
|
|
|
|
def get_log_levels(self, context, log_request):
|
|
levels = utils.get_log_levels(log_request.prefix)
|
|
log_levels = [objects.LogLevel(context, prefix=prefix, level=level)
|
|
for prefix, level in levels.items()]
|
|
return objects.LogLevelList(context, objects=log_levels)
|
|
|
|
|
|
class ThreadPoolManager(Manager):
|
|
def __init__(self, *args, **kwargs):
|
|
self._tp = greenpool.GreenPool()
|
|
super(ThreadPoolManager, self).__init__(*args, **kwargs)
|
|
|
|
def _add_to_threadpool(self, func, *args, **kwargs):
|
|
self._tp.spawn_n(func, *args, **kwargs)
|
|
|
|
|
|
class SchedulerDependentManager(ThreadPoolManager):
|
|
"""Periodically send capability updates to the Scheduler services.
|
|
|
|
Services that need to update the Scheduler of their capabilities
|
|
should derive from this class. Otherwise they can derive from
|
|
manager.Manager directly. Updates are only sent after
|
|
update_service_capabilities is called with non-None values.
|
|
|
|
"""
|
|
|
|
def __init__(self, host=None, db_driver=None, service_name='undefined',
|
|
cluster=None):
|
|
self.last_capabilities = None
|
|
self.service_name = service_name
|
|
self.scheduler_rpcapi = scheduler_rpcapi.SchedulerAPI()
|
|
super(SchedulerDependentManager, self).__init__(host, db_driver,
|
|
cluster=cluster)
|
|
|
|
def update_service_capabilities(self, capabilities):
|
|
"""Remember these capabilities to send on next periodic update."""
|
|
self.last_capabilities = capabilities
|
|
|
|
def _publish_service_capabilities(self, context):
|
|
"""Pass data back to the scheduler at a periodic interval."""
|
|
if self.last_capabilities:
|
|
LOG.debug('Notifying Schedulers of capabilities ...')
|
|
self.scheduler_rpcapi.update_service_capabilities(
|
|
context,
|
|
self.service_name,
|
|
self.host,
|
|
self.last_capabilities,
|
|
self.cluster)
|
|
try:
|
|
self.scheduler_rpcapi.notify_service_capabilities(
|
|
context,
|
|
self.service_name,
|
|
self.service_topic_queue,
|
|
self.last_capabilities)
|
|
except exception.ServiceTooOld as e:
|
|
# This means we have Newton's c-sch in the deployment, so
|
|
# rpcapi cannot send the message. We can safely ignore the
|
|
# error. Log it because it shouldn't happen after upgrade.
|
|
msg = ("Failed to notify about cinder-volume service "
|
|
"capabilities for host %(host)s. This is normal "
|
|
"during a live upgrade. Error: %(e)s")
|
|
LOG.warning(msg, {'host': self.host, 'e': e})
|
|
|
|
def reset(self):
|
|
super(SchedulerDependentManager, self).reset()
|
|
self.scheduler_rpcapi = scheduler_rpcapi.SchedulerAPI()
|
|
|
|
|
|
class CleanableManager(object):
|
|
def do_cleanup(self, context, cleanup_request):
|
|
LOG.info('Initiating service %s cleanup',
|
|
cleanup_request.service_id)
|
|
|
|
# If the 'until' field in the cleanup request is not set, we default to
|
|
# this very moment.
|
|
until = cleanup_request.until or timeutils.utcnow()
|
|
keep_entry = False
|
|
|
|
to_clean = db.worker_get_all(
|
|
context,
|
|
resource_type=cleanup_request.resource_type,
|
|
resource_id=cleanup_request.resource_id,
|
|
service_id=cleanup_request.service_id,
|
|
until=until)
|
|
|
|
for clean in to_clean:
|
|
original_service_id = clean.service_id
|
|
original_time = clean.updated_at
|
|
# Try to do a soft delete to mark the entry as being cleaned up
|
|
# by us (setting service id to our service id).
|
|
res = db.worker_claim_for_cleanup(context,
|
|
claimer_id=self.service_id,
|
|
orm_worker=clean)
|
|
|
|
# Claim may fail if entry is being cleaned by another service, has
|
|
# been removed (finished cleaning) by another service or the user
|
|
# started a new cleanable operation.
|
|
# In any of these cases we don't have to do cleanup or remove the
|
|
# worker entry.
|
|
if not res:
|
|
continue
|
|
|
|
# Try to get versioned object for resource we have to cleanup
|
|
try:
|
|
vo_cls = getattr(objects, clean.resource_type)
|
|
vo = vo_cls.get_by_id(context, clean.resource_id)
|
|
# Set the worker DB entry in the VO and mark it as being a
|
|
# clean operation
|
|
clean.cleaning = True
|
|
vo.worker = clean
|
|
except exception.NotFound:
|
|
LOG.debug('Skipping cleanup for non existent %(type)s %(id)s.',
|
|
{'type': clean.resource_type,
|
|
'id': clean.resource_id})
|
|
else:
|
|
# Resource status should match
|
|
if vo.status != clean.status:
|
|
LOG.debug('Skipping cleanup for mismatching work on '
|
|
'%(type)s %(id)s: %(exp_sts)s <> %(found_sts)s.',
|
|
{'type': clean.resource_type,
|
|
'id': clean.resource_id,
|
|
'exp_sts': clean.status,
|
|
'found_sts': vo.status})
|
|
else:
|
|
LOG.info('Cleaning %(type)s with id %(id)s and status '
|
|
'%(status)s',
|
|
{'type': clean.resource_type,
|
|
'id': clean.resource_id,
|
|
'status': clean.status},
|
|
resource=vo)
|
|
try:
|
|
# Some cleanup jobs are performed asynchronously, so
|
|
# we don't delete the worker entry, they'll take care
|
|
# of it
|
|
keep_entry = self._do_cleanup(context, vo)
|
|
except Exception:
|
|
LOG.exception('Could not perform cleanup.')
|
|
# Return the worker DB entry to the original service
|
|
db.worker_update(context, clean.id,
|
|
service_id=original_service_id,
|
|
updated_at=original_time)
|
|
continue
|
|
|
|
# The resource either didn't exist or was properly cleaned, either
|
|
# way we can remove the entry from the worker table if the cleanup
|
|
# method doesn't want to keep the entry (for example for delayed
|
|
# deletion).
|
|
if not keep_entry and not db.worker_destroy(context, id=clean.id):
|
|
LOG.warning('Could not remove worker entry %s.', clean.id)
|
|
|
|
LOG.info('Service %s cleanup completed.', cleanup_request.service_id)
|
|
|
|
def _do_cleanup(self, ctxt, vo_resource):
|
|
return False
|
|
|
|
def init_host(self, service_id, **kwargs):
|
|
ctxt = context.get_admin_context()
|
|
self.service_id = service_id
|
|
# TODO(geguileo): Once we don't support MySQL 5.5 anymore we can remove
|
|
# call to workers_init.
|
|
db.workers_init()
|
|
cleanup_request = objects.CleanupRequest(service_id=service_id)
|
|
self.do_cleanup(ctxt, cleanup_request)
|