Cesar77RR's picture
download
raw
17.3 kB
import datetime
import random
import time
from neutron_lib import constants
from neutron_lib import context as ncontext
from oslo_config import cfg
from oslo_log import log as logging
import oslo_messaging
from oslo_utils import timeutils
from sqlalchemy import orm
from sqlalchemy.orm import exc
from neutron._i18n import _
from neutron.agent.common import utils as agent_utils
from neutron.common import constants as n_const
from neutron.common import utils
from neutron.db import agents_db
from neutron.db.availability_zone import network as network_az
from neutron.db.models import agent as agent_model
from neutron.db.network_dhcp_agent_binding import models as ndab_model
from neutron.extensions import agent as ext_agent
from neutron.extensions import dhcpagentscheduler
from neutron import worker as neutron_worker
LOG = logging.getLogger(__name__)
AGENTS_SCHEDULER_OPTS = [cfg.StrOpt('network_scheduler_driver', default='neutron.scheduler.dhcp_agent_scheduler.WeightScheduler', help=_('Driver to use for scheduling network to DHCP agent')), cfg.BoolOpt('network_auto_schedule', default=True, help=_('Allow auto scheduling networks to DHCP agent.')), cfg.BoolOpt('allow_automatic_dhcp_failover', default=True, help=_('Automatically remove networks from offline DHCP agents.')), cfg.IntOpt('dhcp_agents_per_network', default=1, help=_('Number of DHCP agents scheduled to host a tenant network. If this number is greater than 1, the scheduler automatically assigns multiple DHCP agents for a given tenant network, providing high availability for DHCP service.')), cfg.BoolOpt('enable_services_on_agents_with_admin_state_down', default=False, help=_('Enable services on an agent with admin_state_up False. If this option is False, when admin_state_up of an agent is turned False, services on it will be disabled. Agents with admin_state_up False are not selected for automatic scheduling regardless of this option. But manual scheduling to such agents is available if this option is True.'))]
cfg.CONF.register_opts(AGENTS_SCHEDULER_OPTS)
class AgentSchedulerDbMixin(agents_db.AgentDbMixin, ):
'Common class for agent scheduler mixins.'
agent_notifiers = {constants.AGENT_TYPE_DHCP: None, constants.AGENT_TYPE_L3: None, constants.AGENT_TYPE_LOADBALANCER: None}
@staticmethod
def is_eligible_agent(active, agent):
if (active is None):
return True
else:
return (not agent_utils.is_agent_down(agent['heartbeat_timestamp']))
def update_agent(self, context, id, agent):
original_agent = self.get_agent(context, id)
result = super(AgentSchedulerDbMixin, self).update_agent(context, id, agent)
agent_data = agent['agent']
agent_notifier = self.agent_notifiers.get(original_agent['agent_type'])
if (agent_notifier and ('admin_state_up' in agent_data) and (original_agent['admin_state_up'] != agent_data['admin_state_up'])):
agent_notifier.agent_updated(context, agent_data['admin_state_up'], original_agent['host'])
return result
def add_agent_status_check_worker(self, function):
interval = max((cfg.CONF.agent_down_time // 2), 1)
initial_delay = random.randint(interval, (interval * 2))
check_worker = neutron_worker.PeriodicWorker(function, interval, initial_delay)
self.add_worker(check_worker)
def agent_dead_limit_seconds(self):
return (cfg.CONF.agent_down_time * 2)
def wait_down_agents(self, agent_type, agent_dead_limit):
'Gives chance for agents to send a heartbeat.'
tdelta = (timeutils.utcnow() - getattr(self, '_clock_jump_canary', timeutils.utcnow()))
if (tdelta.total_seconds() > cfg.CONF.agent_down_time):
LOG.warning('Time since last %s agent reschedule check has exceeded the interval between checks. Waiting before check to allow agents to send a heartbeat in case there was a clock adjustment.', agent_type)
time.sleep(agent_dead_limit)
self._clock_jump_canary = timeutils.utcnow()
def get_cutoff_time(self, agent_dead_limit):
cutoff = (timeutils.utcnow() - datetime.timedelta(seconds=agent_dead_limit))
return cutoff
def reschedule_resources_from_down_agents(self, agent_type, get_down_bindings, agent_id_attr, resource_id_attr, resource_name, reschedule_resource, rescheduling_failed):
'Reschedule resources from down neutron agents\n if admin state is up.\n '
agent_dead_limit = self.agent_dead_limit_seconds()
self.wait_down_agents(agent_type, agent_dead_limit)
context = ncontext.get_admin_context()
try:
down_bindings = get_down_bindings(context, agent_dead_limit)
agents_back_online = set()
for binding in down_bindings:
binding_agent_id = getattr(binding, agent_id_attr)
binding_resource_id = getattr(binding, resource_id_attr)
if (binding_agent_id in agents_back_online):
continue
else:
context = ncontext.get_admin_context()
agent = self._get_agent(context, binding_agent_id)
if agent.is_active:
agents_back_online.add(binding_agent_id)
continue
LOG.warning('Rescheduling %(resource_name)s %(resource)s from agent %(agent)s because the agent did not report to the server in the last %(dead_time)s seconds.', {'resource_name': resource_name, 'resource': binding_resource_id, 'agent': binding_agent_id, 'dead_time': agent_dead_limit})
try:
reschedule_resource(context, binding_resource_id)
except (rescheduling_failed, oslo_messaging.RemoteError):
LOG.exception('Failed to reschedule %(resource_name)s %(resource)s', {'resource_name': resource_name, 'resource': binding_resource_id})
except Exception:
LOG.exception('Exception encountered during %(resource_name)s rescheduling.', {'resource_name': resource_name})
class DhcpAgentSchedulerDbMixin(dhcpagentscheduler.DhcpAgentSchedulerPluginBase, AgentSchedulerDbMixin, ):
'Mixin class to add DHCP agent scheduler extension to db_base_plugin_v2.\n '
network_scheduler = None
def add_periodic_dhcp_agent_status_check(self):
if (not cfg.CONF.allow_automatic_dhcp_failover):
LOG.info('Skipping periodic DHCP agent status check because automatic network rescheduling is disabled.')
return
self.add_agent_status_check_worker(self.remove_networks_from_down_agents)
def is_eligible_agent(self, context, active, agent):
return (AgentSchedulerDbMixin.is_eligible_agent(active, agent) or self.agent_starting_up(context, agent))
def agent_starting_up(self, context, agent):
"Check if agent was just started.\n\n Method returns True if agent is in its 'starting up' period.\n Return value depends on amount of networks assigned to the agent.\n It doesn't look at latest heartbeat timestamp as it is assumed\n that this method is called for agents that are considered dead.\n "
agent_dead_limit = datetime.timedelta(seconds=self.agent_dead_limit_seconds())
network_count = context.session.query(ndab_model.NetworkDhcpAgentBinding).filter_by(dhcp_agent_id=agent['id']).count()
additional_time = datetime.timedelta(seconds=(2 * network_count))
LOG.debug('Checking if agent starts up and giving it additional %s', additional_time)
agent_expected_up = ((agent['started_at'] + agent_dead_limit) + additional_time)
return (agent_expected_up > timeutils.utcnow())
def _schedule_network(self, context, network_id, dhcp_notifier):
LOG.info('Scheduling unhosted network %s', network_id)
try:
network = self.get_network(context, network_id)
agents = self.schedule_network(context, network)
if (not agents):
LOG.info('Failed to schedule network %s, no eligible agents or it might be already scheduled by another server', network_id)
return
if (not dhcp_notifier):
return
for agent in agents:
LOG.info('Adding network %(net)s to agent %(agent)s on host %(host)s', {'net': network_id, 'agent': agent.id, 'host': agent.host})
dhcp_notifier.network_added_to_agent(context, network_id, agent.host)
except Exception:
LOG.exception('Failed to schedule network %s', network_id)
def _filter_bindings(self, context, bindings):
'Skip bindings for which the agent is dead, but starting up.'
checked_agents = {}
for binding in bindings:
try:
agent_id = binding.dhcp_agent['id']
if (agent_id not in checked_agents):
if self.agent_starting_up(context, binding.dhcp_agent):
checked_agents[agent_id] = True
LOG.debug('Agent %s is starting up, skipping', agent_id)
else:
checked_agents[agent_id] = False
if (not checked_agents[agent_id]):
(yield binding)
except exc.ObjectDeletedError:
LOG.debug('binding was removed concurrently, skipping it')
def remove_networks_from_down_agents(self):
'Remove networks from down DHCP agents if admin state is up.\n\n Reschedule them if configured so.\n '
agent_dead_limit = self.agent_dead_limit_seconds()
self.wait_down_agents('DHCP', agent_dead_limit)
cutoff = self.get_cutoff_time(agent_dead_limit)
context = ncontext.get_admin_context()
try:
down_bindings = context.session.query(ndab_model.NetworkDhcpAgentBinding).join(agent_model.Agent).filter((agent_model.Agent.heartbeat_timestamp < cutoff), agent_model.Agent.admin_state_up)
dhcp_notifier = self.agent_notifiers.get(constants.AGENT_TYPE_DHCP)
dead_bindings = [b for b in self._filter_bindings(context, down_bindings)]
agents = self.get_agents_db(context, {'agent_type': [constants.AGENT_TYPE_DHCP]})
if (not agents):
return
active_agents = [agent for agent in agents if self.is_eligible_agent(context, True, agent)]
if (not active_agents):
LOG.warning('No DHCP agents available, skipping rescheduling')
return
for binding in dead_bindings:
LOG.warning('Removing network %(network)s from agent %(agent)s because the agent did not report to the server in the last %(dead_time)s seconds.', {'network': binding.network_id, 'agent': binding.dhcp_agent_id, 'dead_time': agent_dead_limit})
saved_binding = {'net': binding.network_id, 'agent': binding.dhcp_agent_id}
try:
self.remove_network_from_dhcp_agent(context, binding.dhcp_agent_id, binding.network_id, notify=False)
except dhcpagentscheduler.NetworkNotHostedByDhcpAgent:
LOG.debug('Network %(net)s already removed from DHCP agent %(agent)s', saved_binding)
except Exception:
LOG.exception('Unexpected exception occurred while removing network %(net)s from agent %(agent)s', saved_binding)
if cfg.CONF.network_auto_schedule:
self._schedule_network(context, saved_binding['net'], dhcp_notifier)
except Exception:
LOG.exception('Exception encountered during network rescheduling')
def get_dhcp_agents_hosting_networks(self, context, network_ids, active=None, admin_state_up=None, hosts=None):
if (not network_ids):
return []
query = context.session.query(ndab_model.NetworkDhcpAgentBinding)
query = query.options(orm.contains_eager(ndab_model.NetworkDhcpAgentBinding.dhcp_agent))
query = query.join(ndab_model.NetworkDhcpAgentBinding.dhcp_agent)
if network_ids:
query = query.filter(ndab_model.NetworkDhcpAgentBinding.network_id.in_(network_ids))
if hosts:
query = query.filter(agent_model.Agent.host.in_(hosts))
if (admin_state_up is not None):
query = query.filter((agent_model.Agent.admin_state_up == admin_state_up))
return [binding.dhcp_agent for binding in query if self.is_eligible_agent(context, active, binding.dhcp_agent)]
def add_network_to_dhcp_agent(self, context, id, network_id):
self._get_network(context, network_id)
with context.session.begin(subtransactions=True):
agent_db = self._get_agent(context, id)
if ((agent_db['agent_type'] != constants.AGENT_TYPE_DHCP) or (not services_available(agent_db['admin_state_up']))):
raise dhcpagentscheduler.InvalidDHCPAgent(id=id)
dhcp_agents = self.get_dhcp_agents_hosting_networks(context, [network_id])
for dhcp_agent in dhcp_agents:
if (id == dhcp_agent.id):
raise dhcpagentscheduler.NetworkHostedByDHCPAgent(network_id=network_id, agent_id=id)
binding = ndab_model.NetworkDhcpAgentBinding()
binding.dhcp_agent_id = id
binding.network_id = network_id
context.session.add(binding)
dhcp_notifier = self.agent_notifiers.get(constants.AGENT_TYPE_DHCP)
if dhcp_notifier:
dhcp_notifier.network_added_to_agent(context, network_id, agent_db.host)
def remove_network_from_dhcp_agent(self, context, id, network_id, notify=True):
agent = self._get_agent(context, id)
try:
query = context.session.query(ndab_model.NetworkDhcpAgentBinding)
binding = query.filter((ndab_model.NetworkDhcpAgentBinding.network_id == network_id), (ndab_model.NetworkDhcpAgentBinding.dhcp_agent_id == id)).one()
except exc.NoResultFound:
raise dhcpagentscheduler.NetworkNotHostedByDhcpAgent(network_id=network_id, agent_id=id)
device_id = utils.get_dhcp_agent_device_id(network_id, agent['host'])
filters = dict(device_id=[device_id])
ports = self.get_ports(context, filters=filters)
for port in ports:
port['device_id'] = n_const.DEVICE_ID_RESERVED_DHCP_PORT
self.update_port(context, port['id'], dict(port=port))
with context.session.begin():
context.session.delete(binding)
if (not notify):
return
dhcp_notifier = self.agent_notifiers.get(constants.AGENT_TYPE_DHCP)
if dhcp_notifier:
dhcp_notifier.network_removed_from_agent(context, network_id, agent.host)
def list_networks_on_dhcp_agent(self, context, id):
query = context.session.query(ndab_model.NetworkDhcpAgentBinding.network_id)
query = query.filter((ndab_model.NetworkDhcpAgentBinding.dhcp_agent_id == id))
net_ids = [item[0] for item in query]
if net_ids:
return {'networks': self.get_networks(context, filters={'id': net_ids})}
else:
self._get_agent(context, id)
return {'networks': []}
def list_active_networks_on_active_dhcp_agent(self, context, host):
try:
agent = self._get_agent_by_type_and_host(context, constants.AGENT_TYPE_DHCP, host)
except ext_agent.AgentNotFoundByTypeHost:
LOG.debug('DHCP Agent not found on host %s', host)
return []
if (not services_available(agent.admin_state_up)):
return []
query = context.session.query(ndab_model.NetworkDhcpAgentBinding.network_id)
query = query.filter((ndab_model.NetworkDhcpAgentBinding.dhcp_agent_id == agent.id))
net_ids = [item[0] for item in query]
if net_ids:
return self.get_networks(context, filters={'id': net_ids, 'admin_state_up': [True]})
else:
return []
def list_dhcp_agents_hosting_network(self, context, network_id):
dhcp_agents = self.get_dhcp_agents_hosting_networks(context, [network_id])
agent_ids = [dhcp_agent.id for dhcp_agent in dhcp_agents]
if agent_ids:
return {'agents': self.get_agents(context, filters={'id': agent_ids})}
else:
return {'agents': []}
def schedule_network(self, context, created_network):
if self.network_scheduler:
return self.network_scheduler.schedule(self, context, created_network)
def auto_schedule_networks(self, context, host):
if self.network_scheduler:
pass
class AZDhcpAgentSchedulerDbMixin(DhcpAgentSchedulerDbMixin, network_az.NetworkAvailabilityZoneMixin, ):
'Mixin class to add availability_zone supported DHCP agent scheduler.'
def get_network_availability_zones(self, network):
zones = {agent.availability_zone for agent in network.dhcp_agents}
return list(zones)
def services_available(admin_state_up):
if cfg.CONF.enable_services_on_agents_with_admin_state_down:
return True
return admin_state_up
def get_admin_state_up_filter():
if cfg.CONF.enable_services_on_agents_with_admin_state_down:
return None
return True

Xet Storage Details

Size:
17.3 kB
·
Xet hash:
c758dba7cb497931831412b50a6097c2d6b5561cc978cfc3460d5f16ab6e4ac8

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.