osm_policy_module/autoscaling/service.py

   1 # -*- coding: utf-8 -*-
   2 # pylint: disable=no-member
   3
   4 # Copyright 2018 Whitestack, LLC
   5 # *************************************************************
   6
   7 # This file is part of OSM Monitoring module
   8 # All Rights Reserved to Whitestack, LLC
   9
  10 # Licensed under the Apache License, Version 2.0 (the "License"); you may
  11 # not use this file except in compliance with the License. You may obtain
  12 # a copy of the License at
  13
  14 #         http://www.apache.org/licenses/LICENSE-2.0
  15
  16 # Unless required by applicable law or agreed to in writing, software
  17 # distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
  18 # WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
  19 # License for the specific language governing permissions and limitations
  20 # under the License.
  21
  22 # For those usages not covered by the Apache License, Version 2.0 please
  23 # contact: bdiaz@whitestack.com or glavado@whitestack.com
  24 ##
  25 import asyncio
  26 import datetime
  27 import json
  28 import logging
  29 from typing import List
  30
  31 from osm_policy_module.common.common_db_client import CommonDbClient
  32 from osm_policy_module.common.lcm_client import LcmClient
  33 from osm_policy_module.common.mon_client import MonClient
  34 from osm_policy_module.core import database
  35 from osm_policy_module.core.config import Config
  36 from osm_policy_module.core.database import ScalingGroup, ScalingAlarm, ScalingPolicy, ScalingCriteria, \
  37     ScalingAlarmRepository, ScalingGroupRepository, ScalingPolicyRepository, ScalingCriteriaRepository
  38 from osm_policy_module.core.exceptions import VdurNotFound
  39 from osm_policy_module.utils.vnfd import VnfdUtils
  40
  41 log = logging.getLogger(__name__)
  42
  43
  44 class AutoscalingService:
  45
  46     def __init__(self, config: Config, loop=None):
  47         self.conf = config
  48         if not loop:
  49             loop = asyncio.get_event_loop()
  50         self.loop = loop
  51         self.db_client = CommonDbClient(config)
  52         self.mon_client = MonClient(config, loop=self.loop)
  53         self.lcm_client = LcmClient(config, loop=self.loop)
  54
  55     async def configure_scaling_groups(self, nsr_id: str):
  56         """
  57         Configures scaling groups for a network service. Creates records in DB. Creates alarms in MON.
  58         :param nsr_id: Network service record id
  59         :return:
  60         """
  61         log.info("Configuring scaling groups for network service with nsr_id: %s",
  62                  nsr_id)
  63         alarms_created = []
  64         database.db.connect()
  65         try:
  66             with database.db.atomic() as tx:
  67                 try:
  68                     vnfrs = self.db_client.get_vnfrs(nsr_id)
  69                     for vnfr in vnfrs:
  70                         log.debug("Processing vnfr: %s", vnfr)
  71                         vnfd = self.db_client.get_vnfd(vnfr['vnfd-id'])
  72                         if 'scaling-group-descriptor' not in vnfd:
  73                             log.debug("No scaling group present in vnfd")
  74                             continue
  75                         scaling_groups = vnfd['scaling-group-descriptor']
  76                         vnf_monitoring_params = vnfd['monitoring-param']
  77                         for scaling_group in scaling_groups:
  78                             scaling_group_record = self._get_or_create_scaling_group(nsr_id,
  79                                                                                      vnfr['member-vnf-index-ref'],
  80                                                                                      scaling_group)
  81                             for scaling_policy in scaling_group['scaling-policy']:
  82                                 if scaling_policy['scaling-type'] != 'automatic':
  83                                     continue
  84                                 scaling_policy_record = self._get_or_create_scaling_policy(nsr_id,
  85                                                                                            scaling_policy,
  86                                                                                            scaling_group_record)
  87
  88                                 for scaling_criteria in scaling_policy['scaling-criteria']:
  89                                     scaling_criteria_record = self._get_or_create_scaling_criteria(
  90                                         nsr_id,
  91                                         scaling_criteria,
  92                                         scaling_policy_record
  93                                     )
  94                                     vnf_monitoring_param = next(
  95                                         filter(
  96                                             lambda param: param['id'] == scaling_criteria[
  97                                                 'vnf-monitoring-param-ref'
  98                                             ],
  99                                             vnf_monitoring_params)
 100                                     )
 101                                     vdurs = self._get_monitored_vdurs(vnf_monitoring_param, vnfr['vdur'], vnfd)
 102                                     for vdur in vdurs:
 103                                         log.debug("Creating alarm for vdur %s ", vdur)
 104                                         try:
 105                                             ScalingAlarmRepository.get(ScalingAlarm.vdu_name == vdur['name'],
 106                                                                        ScalingCriteria.name == scaling_criteria['name'],
 107                                                                        ScalingPolicy.name == scaling_policy['name'],
 108                                                                        ScalingGroup.nsr_id == nsr_id,
 109                                                                        join_classes=[ScalingCriteria,
 110                                                                                      ScalingPolicy,
 111                                                                                      ScalingGroup])
 112                                             log.debug("vdu %s already has an alarm configured", vdur['name'])
 113                                             continue
 114                                         except ScalingAlarm.DoesNotExist:
 115                                             pass
 116                                         metric_name = self._get_metric_name(vnf_monitoring_param, vdur, vnfd)
 117
 118                                         db_nsr = self.db_client.get_nsr(nsr_id)
 119                                         nb_scale_op = 0
 120                                         if db_nsr["_admin"].get("scaling-group"):
 121                                             db_nsr_admin = db_nsr["_admin"]["scaling-group"]
 122                                             for admin_scale_index, admin_scale_info in enumerate(db_nsr_admin):
 123                                                 if admin_scale_info["name"] == scaling_group["name"]:
 124                                                     nb_scale_op = admin_scale_info.get("nb-scale-op", 0)
 125                                                     break
 126                                         min_instance_count = int(scaling_group["min-instance-count"])
 127                                         if nb_scale_op > min_instance_count:
 128                                             alarm_uuid = await self.mon_client.create_alarm(
 129                                                 metric_name=metric_name,
 130                                                 ns_id=nsr_id,
 131                                                 vdu_name=vdur['name'],
 132                                                 vnf_member_index=vnfr['member-vnf-index-ref'],
 133                                                 threshold=scaling_criteria['scale-in-threshold'],
 134                                                 operation=scaling_criteria['scale-in-relational-operation'],
 135                                                 statistic=vnf_monitoring_param['aggregation-type']
 136                                             )
 137                                             alarm = ScalingAlarmRepository.create(
 138                                                 alarm_uuid=alarm_uuid,
 139                                                 action='scale_in',
 140                                                 vnf_member_index=vnfr['member-vnf-index-ref'],
 141                                                 vdu_name=vdur['name'],
 142                                                 scaling_criteria=scaling_criteria_record
 143                                             )
 144                                             alarms_created.append(alarm)
 145                                         alarm_uuid = await self.mon_client.create_alarm(
 146                                             metric_name=metric_name,
 147                                             ns_id=nsr_id,
 148                                             vdu_name=vdur['name'],
 149                                             vnf_member_index=vnfr['member-vnf-index-ref'],
 150                                             threshold=scaling_criteria['scale-out-threshold'],
 151                                             operation=scaling_criteria['scale-out-relational-operation'],
 152                                             statistic=vnf_monitoring_param['aggregation-type']
 153                                         )
 154                                         alarm = ScalingAlarmRepository.create(
 155                                             alarm_uuid=alarm_uuid,
 156                                             action='scale_out',
 157                                             vnf_member_index=vnfr['member-vnf-index-ref'],
 158                                             vdu_name=vdur['name'],
 159                                             scaling_criteria=scaling_criteria_record
 160                                         )
 161                                         alarms_created.append(alarm)
 162
 163                 except Exception as e:
 164                     log.exception("Error configuring scaling groups:")
 165                     tx.rollback()
 166                     if len(alarms_created) > 0:
 167                         log.info("Cleaning alarm resources in MON")
 168                         for alarm in alarms_created:
 169                             await self.mon_client.delete_alarm(
 170                                 alarm.scaling_criteria.scaling_policy.scaling_group.nsr_id,
 171                                 alarm.vnf_member_index,
 172                                 alarm.vdu_name,
 173                                 alarm.alarm_uuid)
 174                     raise e
 175         finally:
 176             database.db.close()
 177
 178     async def delete_scaling_groups(self, nsr_id: str):
 179         log.debug("Deleting scaling groups for network service %s", nsr_id)
 180         database.db.connect()
 181         try:
 182             with database.db.atomic() as tx:
 183                 try:
 184                     for scaling_group in ScalingGroupRepository.list(ScalingGroup.nsr_id == nsr_id):
 185                         for scaling_policy in scaling_group.scaling_policies:
 186                             for scaling_criteria in scaling_policy.scaling_criterias:
 187                                 for alarm in scaling_criteria.scaling_alarms:
 188                                     try:
 189                                         await self.mon_client.delete_alarm(
 190                                             alarm.scaling_criteria.scaling_policy.scaling_group.nsr_id,
 191                                             alarm.vnf_member_index,
 192                                             alarm.vdu_name,
 193                                             alarm.alarm_uuid)
 194                                     except ValueError:
 195                                         log.exception("Error deleting alarm in MON %s", alarm.alarm_uuid)
 196                                     alarm.delete_instance()
 197                                 scaling_criteria.delete_instance()
 198                             scaling_policy.delete_instance()
 199                         scaling_group.delete_instance()
 200
 201                 except Exception as e:
 202                     log.exception("Error deleting scaling groups and alarms:")
 203                     tx.rollback()
 204                     raise e
 205         finally:
 206             database.db.close()
 207
 208     async def delete_orphaned_alarms(self, nsr_id):
 209         log.info("Deleting orphaned scaling alarms for network service %s", nsr_id)
 210         database.db.connect()
 211         try:
 212             with database.db.atomic() as tx:
 213                 try:
 214                     for scaling_group in ScalingGroupRepository.list(ScalingGroup.nsr_id == nsr_id):
 215                         for scaling_policy in scaling_group.scaling_policies:
 216                             for scaling_criteria in scaling_policy.scaling_criterias:
 217                                 for alarm in scaling_criteria.scaling_alarms:
 218                                     try:
 219                                         self.db_client.get_vdur(nsr_id, alarm.vnf_member_index, alarm.vdu_name)
 220                                     except VdurNotFound:
 221                                         log.debug("Deleting orphaned scaling alarm %s", alarm.alarm_uuid)
 222                                         try:
 223                                             await self.mon_client.delete_alarm(
 224                                                 alarm.scaling_criteria.scaling_policy.scaling_group.nsr_id,
 225                                                 alarm.vnf_member_index,
 226                                                 alarm.vdu_name,
 227                                                 alarm.alarm_uuid)
 228                                         except ValueError:
 229                                             log.exception("Error deleting alarm in MON %s", alarm.alarm_uuid)
 230                                         alarm.delete_instance()
 231
 232                 except Exception as e:
 233                     log.exception("Error deleting orphaned alarms:")
 234                     tx.rollback()
 235                     raise e
 236         finally:
 237             database.db.close()
 238
 239     async def handle_alarm(self, alarm_uuid: str, status: str):
 240         await self.update_alarm_status(alarm_uuid, status)
 241         await self.evaluate_policy(alarm_uuid)
 242
 243     async def update_alarm_status(self, alarm_uuid: str, status: str):
 244         database.db.connect()
 245         try:
 246             with database.db.atomic():
 247                 alarm = ScalingAlarmRepository.get(ScalingAlarm.alarm_uuid == alarm_uuid)
 248                 alarm.last_status = status
 249                 alarm.save()
 250         except ScalingAlarm.DoesNotExist:
 251             log.debug("There is no autoscaling action configured for alarm %s.", alarm_uuid)
 252         finally:
 253             database.db.close()
 254
 255     async def evaluate_policy(self, alarm_uuid):
 256         database.db.connect()
 257         try:
 258             with database.db.atomic():
 259                 alarm = ScalingAlarmRepository.get(ScalingAlarm.alarm_uuid == alarm_uuid)
 260                 vnf_member_index = alarm.vnf_member_index
 261                 action = alarm.action
 262                 scaling_policy = alarm.scaling_criteria.scaling_policy
 263                 if not scaling_policy.enabled:
 264                     return
 265                 if action == 'scale_in':
 266                     operation = scaling_policy.scale_in_operation
 267                 elif action == 'scale_out':
 268                     operation = scaling_policy.scale_out_operation
 269                 else:
 270                     raise Exception('Unknown alarm action {}'.format(alarm.action))
 271                 alarms = ScalingAlarmRepository.list(ScalingAlarm.scaling_criteria == alarm.scaling_criteria,
 272                                                      ScalingAlarm.action == alarm.action,
 273                                                      ScalingAlarm.vnf_member_index == vnf_member_index,
 274                                                      ScalingAlarm.vdu_name == alarm.vdu_name)
 275                 statuses = []
 276                 for alarm in alarms:
 277                     statuses.append(alarm.last_status)
 278                 if (operation == 'AND' and set(statuses) == {'alarm'}) or (operation == 'OR' and 'alarm' in statuses):
 279                     delta = datetime.datetime.now() - scaling_policy.last_scale
 280                     if delta.total_seconds() > scaling_policy.cooldown_time:
 281                         log.info("Sending %s action message for ns: %s",
 282                                  alarm.action,
 283                                  scaling_policy.scaling_group.nsr_id)
 284                         await self.lcm_client.scale(scaling_policy.scaling_group.nsr_id,
 285                                                     scaling_policy.scaling_group.name,
 286                                                     vnf_member_index,
 287                                                     action)
 288                         scaling_policy.last_scale = datetime.datetime.now()
 289                         scaling_policy.save()
 290
 291         except ScalingAlarm.DoesNotExist:
 292             log.debug("There is no autoscaling action configured for alarm %s.", alarm_uuid)
 293         finally:
 294             database.db.close()
 295
 296     def _get_or_create_scaling_group(self, nsr_id: str, vnf_member_index: str, scaling_group: dict):
 297         try:
 298             scaling_group_record = ScalingGroupRepository.get(
 299                 ScalingGroup.nsr_id == nsr_id,
 300                 ScalingGroup.vnf_member_index == vnf_member_index,
 301                 ScalingGroup.name == scaling_group['name']
 302             )
 303             log.debug("Found existing scaling group record in DB...")
 304         except ScalingGroup.DoesNotExist:
 305             log.debug("Creating scaling group record in DB...")
 306             scaling_group_record = ScalingGroupRepository.create(
 307                 nsr_id=nsr_id,
 308                 vnf_member_index=vnf_member_index,
 309                 name=scaling_group['name'],
 310                 content=json.dumps(scaling_group)
 311             )
 312             log.debug(
 313                 "Created scaling group record in DB : nsr_id=%s, vnf_member_index=%s, name=%s",
 314                 scaling_group_record.nsr_id,
 315                 scaling_group_record.vnf_member_index,
 316                 scaling_group_record.name)
 317         return scaling_group_record
 318
 319     def _get_or_create_scaling_policy(self, nsr_id: str, scaling_policy: dict, scaling_group_record: ScalingGroup):
 320         try:
 321             scaling_policy_record = ScalingPolicyRepository.get(
 322                 ScalingPolicy.name == scaling_policy['name'],
 323                 ScalingGroup.id == scaling_group_record.id,
 324                 join_classes=[ScalingGroup]
 325             )
 326             log.debug("Found existing scaling policy record in DB...")
 327         except ScalingPolicy.DoesNotExist:
 328             log.debug("Creating scaling policy record in DB...")
 329             scaling_policy_record = ScalingPolicyRepository.create(
 330                 nsr_id=nsr_id,
 331                 name=scaling_policy['name'],
 332                 cooldown_time=scaling_policy['cooldown-time'],
 333                 scaling_group=scaling_group_record,
 334             )
 335             if 'scale-in-operation-type' in scaling_policy:
 336                 scaling_policy_record.scale_in_operation = scaling_policy[
 337                     'scale-in-operation-type']
 338             if 'scale-out-operation-type' in scaling_policy:
 339                 scaling_policy_record.scale_out_operation = scaling_policy[
 340                     'scale-out-operation-type']
 341             if 'enabled' in scaling_policy:
 342                 scaling_policy_record.enabled = scaling_policy['enabled']
 343             scaling_policy_record.save()
 344             log.debug("Created scaling policy record in DB : name=%s, scaling_group.name=%s",
 345                       scaling_policy_record.name,
 346                       scaling_policy_record.scaling_group.name)
 347         return scaling_policy_record
 348
 349     def _get_or_create_scaling_criteria(self, nsr_id: str, scaling_criteria: dict,
 350                                         scaling_policy_record: ScalingPolicy):
 351         try:
 352             scaling_criteria_record = ScalingCriteriaRepository.get(
 353                 ScalingPolicy.id == scaling_policy_record.id,
 354                 ScalingCriteria.name == scaling_criteria['name'],
 355                 join_classes=[ScalingPolicy]
 356             )
 357             log.debug("Found existing scaling criteria record in DB...")
 358         except ScalingCriteria.DoesNotExist:
 359             log.debug("Creating scaling criteria record in DB...")
 360             scaling_criteria_record = ScalingCriteriaRepository.create(
 361                 nsr_id=nsr_id,
 362                 name=scaling_criteria['name'],
 363                 scaling_policy=scaling_policy_record
 364             )
 365             log.debug(
 366                 "Created scaling criteria record in DB : name=%s, scaling_policy.name=%s",
 367                 scaling_criteria_record.name,
 368                 scaling_criteria_record.scaling_policy.name)
 369         return scaling_criteria_record
 370
 371     def _get_monitored_vdurs(self, vnf_monitoring_param: dict, vdurs: List[dict], vnfd: dict):
 372         monitored_vdurs = []
 373         if 'vdu-monitoring-param' in vnf_monitoring_param:
 374             monitored_vdurs = list(
 375                 filter(
 376                     lambda vdur: vdur['vdu-id-ref'] == vnf_monitoring_param
 377                     ['vdu-monitoring-param']
 378                     ['vdu-ref'],
 379                     vdurs
 380                 )
 381             )
 382         elif 'vdu-metric' in vnf_monitoring_param:
 383             monitored_vdurs = list(
 384                 filter(
 385                     lambda vdur: vdur['vdu-id-ref'] == vnf_monitoring_param
 386                     ['vdu-metric']
 387                     ['vdu-ref'],
 388                     vdurs
 389                 )
 390             )
 391         elif 'vnf-metric' in vnf_monitoring_param:
 392             vdu = VnfdUtils.get_mgmt_vdu(vnfd)
 393             monitored_vdurs = list(
 394                 filter(
 395                     lambda vdur: vdur['vdu-id-ref'] == vdu['id'],
 396                     vdurs
 397                 )
 398             )
 399         else:
 400             log.warning(
 401                 "Scaling criteria is referring to a vnf-monitoring-param that does not "
 402                 "contain a reference to a vdu or vnf metric.")
 403         return monitored_vdurs
 404
 405     def _get_metric_name(self, vnf_monitoring_param: dict, vdur: dict, vnfd: dict):
 406         vdu = next(
 407             filter(lambda vdu: vdu['id'] == vdur['vdu-id-ref'], vnfd['vdu'])
 408         )
 409         if 'vdu-monitoring-param' in vnf_monitoring_param:
 410             vdu_monitoring_param = next(filter(
 411                 lambda param: param['id'] == vnf_monitoring_param['vdu-monitoring-param'][
 412                     'vdu-monitoring-param-ref'], vdu['monitoring-param']))
 413             nfvi_metric = vdu_monitoring_param['nfvi-metric']
 414             return nfvi_metric
 415         if 'vdu-metric' in vnf_monitoring_param:
 416             vnf_metric_name = vnf_monitoring_param['vdu-metric']['vdu-metric-name-ref']
 417             return vnf_metric_name
 418         if 'vnf-metric' in vnf_monitoring_param:
 419             vnf_metric_name = vnf_monitoring_param['vnf-metric']['vnf-metric-name-ref']
 420             return vnf_metric_name
 421         raise ValueError('No metric name found for vnf_monitoring_param %s' % vnf_monitoring_param['id'])