deb-designate/designate/pool_manager/service.py
Kiall Mac Innes f3dd7a8f42 PM: Ensure only one periodic task runs per host
Ensure multiple periodic sync/recovery tasks do not run concurrently on
a single host by using an inter process lock inside the periodic task.

This has the effect of serializing the tasks, rather than the more correct
deduplication of tasks. Bug 1445127 tracks the longer term fix.

Change-Id: Ib0041f127aacfd0f49ca84deecc2a6950594dd58
Closes-Bug: 1445125
2015-04-21 17:15:18 +01:00

681 lines
26 KiB
Python

# Copyright 2014 eBay Inc.
#
# Author: Ron Rickard <rrickard@ebaysf.com>
#
# 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.
from contextlib import contextmanager
from decimal import Decimal
from oslo.config import cfg
from oslo import messaging
from oslo_log import log as logging
from oslo_concurrency import lockutils
from designate import backend
from designate import exceptions
from designate import objects
from designate import utils
from designate.central import rpcapi as central_api
from designate.mdns import rpcapi as mdns_api
from designate import service
from designate.context import DesignateContext
from designate.i18n import _LE
from designate.i18n import _LI
from designate.i18n import _LW
from designate.pool_manager import cache
LOG = logging.getLogger(__name__)
CONF = cfg.CONF
SUCCESS_STATUS = 'SUCCESS'
ERROR_STATUS = 'ERROR'
NO_DOMAIN_STATUS = 'NO_DOMAIN'
CREATE_ACTION = 'CREATE'
DELETE_ACTION = 'DELETE'
UPDATE_ACTION = 'UPDATE'
MAXIMUM_THRESHOLD = 100
@contextmanager
def wrap_backend_call():
"""
Wraps backend calls, ensuring any exception raised is a Backend exception.
"""
try:
yield
except exceptions.Backend:
raise
except Exception as e:
raise exceptions.Backend('Unknown backend failure: %r' % e)
class Service(service.RPCService, service.Service):
"""
Service side of the Pool Manager RPC API.
API version history:
1.0 - Initial version
"""
RPC_API_VERSION = '1.0'
target = messaging.Target(version=RPC_API_VERSION)
def __init__(self, threads=None):
super(Service, self).__init__(threads=threads)
# Build the Pool (and related) Object from Config
self.pool = objects.Pool.from_config(CONF)
# Get a pool manager cache connection.
self.cache = cache.get_pool_manager_cache(
CONF['service:pool_manager'].cache_driver)
# Store some settings for quick access later
self.threshold = CONF['service:pool_manager'].threshold_percentage
self.timeout = CONF['service:pool_manager'].poll_timeout
self.retry_interval = CONF['service:pool_manager'].poll_retry_interval
self.max_retries = CONF['service:pool_manager'].poll_max_retries
self.delay = CONF['service:pool_manager'].poll_delay
# Create the necessary Backend instances for each target
self._setup_target_backends()
def _setup_target_backends(self):
self.target_backends = {}
for target in self.pool.targets:
# Fetch an instance of the Backend class, passing in the options
# and masters
self.target_backends[target.id] = backend.get_backend(
target.type, target)
LOG.info(_LI('%d targets setup'), len(self.pool.targets))
if not self.target_backends:
raise exceptions.NoPoolTargetsConfigured()
@property
def service_name(self):
return 'pool_manager'
@property
def _rpc_topic(self):
# Modify the default topic so it's pool manager instance specific.
topic = super(Service, self)._rpc_topic
topic = '%s.%s' % (topic, CONF['service:pool_manager'].pool_id)
LOG.info(_LI('Using topic %(topic)s for this pool manager instance.')
% {'topic': topic})
return topic
def start(self):
for target in self.pool.targets:
self.target_backends[target.id].start()
super(Service, self).start()
if CONF['service:pool_manager'].enable_recovery_timer:
LOG.info(_LI('Starting periodic recovery timer'))
self.tg.add_timer(
CONF['service:pool_manager'].periodic_recovery_interval,
self.periodic_recovery,
CONF['service:pool_manager'].periodic_recovery_interval)
if CONF['service:pool_manager'].enable_sync_timer:
LOG.info(_LI('Starting periodic synchronization timer'))
self.tg.add_timer(
CONF['service:pool_manager'].periodic_sync_interval,
self.periodic_sync,
CONF['service:pool_manager'].periodic_sync_interval)
def stop(self):
for target in self.pool.targets:
self.target_backends[target.id].stop()
super(Service, self).stop()
@property
def central_api(self):
return central_api.CentralAPI.get_instance()
@property
def mdns_api(self):
return mdns_api.MdnsAPI.get_instance()
# Periodioc Tasks
def periodic_recovery(self):
"""
:return: None
"""
# TODO(kiall): Replace this inter-process-lock with a distributed
# lock, likely using the tooz library - see bug 1445127.
with lockutils.lock('periodic_recovery', external=True, delay=30):
context = DesignateContext.get_admin_context(all_tenants=True)
LOG.debug("Starting Periodic Recovery")
try:
# Handle Deletion Failures
domains = self._get_failed_domains(context, DELETE_ACTION)
for domain in domains:
self.delete_domain(context, domain)
# Handle Creation Failures
domains = self._get_failed_domains(context, CREATE_ACTION)
for domain in domains:
self.create_domain(context, domain)
# Handle Update Failures
domains = self._get_failed_domains(context, UPDATE_ACTION)
for domain in domains:
self.update_domain(context, domain)
except Exception:
LOG.exception(_LE('An unhandled exception in periodic '
'recovery occurred'))
def periodic_sync(self):
"""
:return: None
"""
# TODO(kiall): Replace this inter-process-lock with a distributed
# lock, likely using the tooz library - see bug 1445127.
with lockutils.lock('periodic_sync', external=True, delay=30):
context = DesignateContext.get_admin_context(all_tenants=True)
LOG.debug("Starting Periodic Synchronization")
criterion = {
'pool_id': CONF['service:pool_manager'].pool_id,
'status': '!%s' % ERROR_STATUS
}
periodic_sync_seconds = \
CONF['service:pool_manager'].periodic_sync_seconds
if periodic_sync_seconds is not None:
# Generate the current serial, will provide a UTC Unix TS.
current = utils.increment_serial()
criterion['serial'] = ">%s" % (current - periodic_sync_seconds)
domains = self.central_api.find_domains(context, criterion)
try:
for domain in domains:
# TODO(kiall): If the domain was created within the last
# periodic_sync_seconds, attempt to recreate
# to fill in targets which may have failed.
self.update_domain(context, domain)
except Exception:
LOG.exception(_LE('An unhandled exception in periodic '
'synchronization occurred.'))
# Standard Create/Update/Delete Methods
def create_domain(self, context, domain):
"""
:param context: Security context information.
:param domain: Domain to be created
:return: None
"""
LOG.info(_LI("Creating new domain %s"), domain.name)
results = []
# Create the domain on each of the Pool Targets
for target in self.pool.targets:
results.append(
self._create_domain_on_target(context, target, domain))
if self._exceed_or_meet_threshold(results.count(True)):
LOG.debug('Consensus reached for creating domain %(domain)s '
'on pool targets' % {'domain': domain.name})
else:
LOG.warn(_LW('Consensus not reached for creating domain %(domain)s'
' on pool targets') % {'domain': domain.name})
self.central_api.update_status(
context, domain.id, ERROR_STATUS, domain.serial)
return
# Send a NOTIFY to each also-notifies
for also_notify in self.pool.also_notifies:
self._update_domain_on_also_notify(context, also_notify, domain)
# Send a NOTIFY to each nameserver
for nameserver in self.pool.nameservers:
create_status = self._build_status_object(
nameserver, domain, CREATE_ACTION)
self.cache.store(context, create_status)
self._update_domain_on_nameserver(context, nameserver, domain)
def _create_domain_on_target(self, context, target, domain):
"""
:param context: Security context information.
:param target: Target to create Domain on
:param domain: Domain to be created
:return: True/False
"""
LOG.debug("Creating domain %s on target %s", domain.name, target.id)
backend = self.target_backends[target.id]
try:
backend.create_domain(context, domain)
return True
except Exception:
LOG.exception(_LE("Failed to create domain %(domain)s on target "
"%(target)s"),
{'domain': domain.name, 'target': target.id})
return False
def update_domain(self, context, domain):
"""
:param context: Security context information.
:param domain: Domain to be updated
:return: None
"""
LOG.info(_LI("Updating domain %s"), domain.name)
results = []
# Update the domain on each of the Pool Targets
for target in self.pool.targets:
results.append(
self._update_domain_on_target(context, target, domain))
if self._exceed_or_meet_threshold(results.count(True)):
LOG.debug('Consensus reached for updating domain %(domain)s '
'on pool targets' % {'domain': domain.name})
else:
LOG.warn(_LW('Consensus not reached for updating domain %(domain)s'
' on pool targets') % {'domain': domain.name})
self.central_api.update_status(
context, domain.id, ERROR_STATUS, domain.serial)
return
# Send a NOTIFY to each also-notifies
for also_notify in self.pool.also_notifies:
self._update_domain_on_also_notify(context, also_notify, domain)
# Send a NOTIFY to each nameserver
for nameserver in self.pool.nameservers:
# See if there is already another update in progress
try:
update_status = self.cache.retrieve(
context, nameserver.id, domain.id, UPDATE_ACTION)
except exceptions.PoolManagerStatusNotFound:
update_status = self._build_status_object(
nameserver, domain, UPDATE_ACTION)
self.cache.store(context, update_status)
self._update_domain_on_nameserver(context, nameserver, domain)
def _update_domain_on_target(self, context, target, domain):
"""
:param context: Security context information.
:param target: Target to update Domain on
:param domain: Domain to be updated
:return: True/False
"""
LOG.debug("Updating domain %s on target %s", domain.name, target.id)
backend = self.target_backends[target.id]
try:
backend.update_domain(context, domain)
return True
except Exception:
LOG.exception(_LE("Failed to update domain %(domain)s on target "
"%(target)s"),
{'domain': domain.name, 'target': target.id})
return False
def _update_domain_on_also_notify(self, context, also_notify, domain):
LOG.info(_LI('Updating domain %(domain)s on also_notify %(server)s.') %
{'domain': domain.name,
'server': self._get_destination(also_notify)})
self.mdns_api.notify_zone_changed(
context, domain, also_notify, self.timeout, self.retry_interval,
self.max_retries, 0)
def _update_domain_on_nameserver(self, context, nameserver, domain):
LOG.info(_LI('Updating domain %(domain)s on nameserver %(server)s.') %
{'domain': domain.name,
'server': self._get_destination(nameserver)})
self.mdns_api.notify_zone_changed(
context, domain, nameserver, self.timeout, self.retry_interval,
self.max_retries, 0)
self.mdns_api.poll_for_serial_number(
context, domain, nameserver, self.timeout, self.retry_interval,
self.max_retries, self.delay)
def delete_domain(self, context, domain):
"""
:param context: Security context information.
:param domain: Domain to be deleted
:return: None
"""
LOG.info(_LI("Deleting domain %s"), domain.name)
results = []
# Delete the domain on each of the Pool Targets
for target in self.pool.targets:
results.append(
self._delete_domain_on_target(context, target, domain))
# TODO(kiall): We should monitor that the Domain is actually deleted
# correctly on each of the nameservers, rather than
# assuming a sucessful delete-on-target is OK as we have
# in the past.
if self._exceed_or_meet_threshold(
results.count(True), MAXIMUM_THRESHOLD):
LOG.debug('Consensus reached for deleting domain %(domain)s '
'on pool targets' % {'domain': domain.name})
self.central_api.update_status(
context, domain.id, SUCCESS_STATUS, domain.serial)
else:
LOG.warn(_LW('Consensus not reached for deleting domain %(domain)s'
' on pool targets') % {'domain': domain.name})
self.central_api.update_status(
context, domain.id, ERROR_STATUS, domain.serial)
def _delete_domain_on_target(self, context, target, domain):
"""
:param context: Security context information.
:param target: Target to delete Domain from
:param domain: Domain to be deleted
:return: True/False
"""
LOG.debug("Deleting domain %s on target %s", domain.name, target.id)
backend = self.target_backends[target.id]
try:
backend.delete_domain(context, domain)
return True
except Exception:
LOG.exception(_LE("Failed to delete domain %(domain)s on target "
"%(target)s"),
{'domain': domain.name, 'target': target.id})
return False
def update_status(self, context, domain, nameserver, status,
actual_serial):
"""
update_status is called by mdns for creates and updates.
deletes are handled by the backend entirely and status is determined
at the time of delete itself.
:param context: Security context information.
:param domain: The designate domain object.
:param nameserver: The nameserver for which a status update is being
sent.
:param status: The status, 'SUCCESS' or 'ERROR'.
:param actual_serial: The actual serial number received from the name
server for the domain.
:return: None
"""
LOG.debug("Calling update_status for %s : %s : %s : %s" %
(domain.name, domain.action, status, actual_serial))
action = UPDATE_ACTION if domain.action == 'NONE' else domain.action
with lockutils.lock('update-status-%s' % domain.id):
try:
current_status = self.cache.retrieve(
context, nameserver.id, domain.id, action)
except exceptions.PoolManagerStatusNotFound:
current_status = self._build_status_object(
nameserver, domain, action)
self.cache.store(context, current_status)
cache_serial = current_status.serial_number
LOG.debug('For domain %s : %s on nameserver %s the cache serial '
'is %s and the actual serial is %s.' %
(domain.name, action,
self._get_destination(nameserver),
cache_serial, actual_serial))
if actual_serial and cache_serial <= actual_serial:
current_status.status = status
current_status.serial_number = actual_serial
self.cache.store(context, current_status)
consensus_serial = self._get_consensus_serial(context, domain)
# If there is a valid consensus serial we can still send a success
# for that serial.
# If there is a higher error serial we can also send an error for
# the error serial.
if consensus_serial != 0 and cache_serial <= consensus_serial \
and domain.status != 'ACTIVE':
LOG.info(_LI('For domain %(domain)s '
'the consensus serial is %(consensus_serial)s.') %
{'domain': domain.name,
'consensus_serial': consensus_serial})
self.central_api.update_status(
context, domain.id, SUCCESS_STATUS, consensus_serial)
if status == ERROR_STATUS:
error_serial = self._get_error_serial(
context, domain, consensus_serial)
if error_serial > consensus_serial or error_serial == 0:
LOG.warn(_LW('For domain %(domain)s '
'the error serial is %(error_serial)s.') %
{'domain': domain.name,
'error_serial': error_serial})
self.central_api.update_status(
context, domain.id, ERROR_STATUS, error_serial)
if consensus_serial == domain.serial and self._is_consensus(
context, domain, action, SUCCESS_STATUS,
MAXIMUM_THRESHOLD):
self._clear_cache(context, domain, action)
# Utility Methods
def _get_failed_domains(self, context, action):
criterion = {
'pool_id': CONF['service:pool_manager'].pool_id,
'action': action,
'status': 'ERROR'
}
return self.central_api.find_domains(context, criterion)
@staticmethod
def _get_destination(nameserver):
return '%s:%s' % (nameserver.host, nameserver.port)
@staticmethod
def _percentage(count, total_count):
return (Decimal(count) / Decimal(total_count)) * Decimal(100)
def _exceed_or_meet_threshold(self, count, threshold=None):
threshold = threshold or self.threshold
return self._percentage(
count, len(self.pool.targets)) >= Decimal(threshold)
@staticmethod
def _get_sorted_serials(pool_manager_statuses, descending=False):
serials = []
for pool_manager_status in pool_manager_statuses:
serials.append(pool_manager_status.serial_number)
serials.sort(reverse=descending)
return serials
def _get_serials_ascending(self, pool_manager_statuses):
return self._get_sorted_serials(pool_manager_statuses)
def _get_serials_descending(self, pool_manager_statuses):
return self._get_sorted_serials(pool_manager_statuses, descending=True)
def _is_consensus(self, context, domain, action, status, threshold=None):
status_count = 0
pool_manager_statuses = self._retrieve_statuses(
context, domain, action)
for pool_manager_status in pool_manager_statuses:
if pool_manager_status.status == status:
status_count += 1
if threshold is None:
threshold = self.threshold
return self._exceed_or_meet_threshold(status_count, threshold)
def _get_consensus_serial(self, context, domain):
consensus_serial = 0
action = UPDATE_ACTION if domain.action == 'NONE' else domain.action
pm_statuses = self._retrieve_statuses(context, domain, action)
for serial in self._get_serials_descending(pm_statuses):
serial_count = 0
for pm_status in pm_statuses:
if pm_status.serial_number >= serial:
serial_count += 1
if self._exceed_or_meet_threshold(serial_count, self.threshold):
consensus_serial = serial
break
return consensus_serial
def _get_error_serial(self, context, domain, consensus_serial):
error_serial = 0
action = UPDATE_ACTION if domain.action == 'NONE' else domain.action
if self._is_consensus(context, domain, action, ERROR_STATUS):
pm_statuses = self._retrieve_statuses(context, domain, action)
for serial in self._get_serials_ascending(pm_statuses):
if serial > consensus_serial:
error_serial = serial
break
return error_serial
# When we hear back from the nameserver, the serial_number is set to the
# value the nameserver
@staticmethod
def _build_status_object(nameserver, domain, action):
values = {
'nameserver_id': nameserver.id,
'domain_id': domain.id,
'status': None,
'serial_number': 0,
'action': action
}
return objects.PoolManagerStatus(**values)
# Methods for manipulating the cache.
def _clear_cache(self, context, domain, action=None):
LOG.debug('Clearing cache for domain %s with action %s.' %
(domain.name, action))
pool_manager_statuses = []
if action:
actions = [action]
else:
actions = [CREATE_ACTION, UPDATE_ACTION, DELETE_ACTION]
for nameserver in self.pool.nameservers:
for action in actions:
pool_manager_status = self._build_status_object(
nameserver, domain, action)
pool_manager_statuses.append(pool_manager_status)
for pool_manager_status in pool_manager_statuses:
# Ignore any not found errors while clearing the cache
try:
self.cache.clear(context, pool_manager_status)
except exceptions.PoolManagerStatusNotFound:
pass
def _retrieve_from_mdns(self, context, nameserver, domain, action):
try:
(status, actual_serial, retries) = \
self.mdns_api.get_serial_number(
context, domain, nameserver, self.timeout,
self.retry_interval, self.max_retries, self.delay)
except messaging.MessagingException as msg_ex:
LOG.debug('Could not retrieve status and serial for domain %s on '
'nameserver %s with action %s (%s: %s)' %
(domain.name, self._get_destination(nameserver), action,
type(msg_ex), str(msg_ex)))
return None
pool_manager_status = self._build_status_object(
nameserver, domain, action)
if status == NO_DOMAIN_STATUS:
if action == CREATE_ACTION:
pool_manager_status.status = 'ERROR'
elif action == DELETE_ACTION:
pool_manager_status.status = 'SUCCESS'
# TODO(Ron): Handle this case properly.
elif action == UPDATE_ACTION:
pool_manager_status.status = 'ERROR'
else:
pool_manager_status.status = status
pool_manager_status.serial_number = actual_serial \
if actual_serial is not None else 0
LOG.debug('Retrieved status %s and serial %s for domain %s '
'on nameserver %s with action %s from mdns.' %
(pool_manager_status.status,
pool_manager_status.serial_number,
domain.name, self._get_destination(nameserver), action))
self.cache.store(context, pool_manager_status)
return pool_manager_status
def _retrieve_statuses(self, context, domain, action):
pool_manager_statuses = []
for nameserver in self.pool.nameservers:
try:
pool_manager_status = self.cache.retrieve(
context, nameserver.id, domain.id, action)
LOG.debug('Cache hit! Retrieved status %s and serial %s '
'for domain %s on nameserver %s with action %s from '
'the cache.' %
(pool_manager_status.status,
pool_manager_status.serial_number,
domain.name,
self._get_destination(nameserver), action))
except exceptions.PoolManagerStatusNotFound:
LOG.debug('Cache miss! Did not retrieve status and serial '
'for domain %s on nameserver %s with action %s from '
'the cache. Getting it from the server.' %
(domain.name,
self._get_destination(nameserver),
action))
pool_manager_status = self._retrieve_from_mdns(
context, nameserver, domain, action)
if pool_manager_status is not None:
pool_manager_statuses.append(pool_manager_status)
return pool_manager_statuses