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 datetime
  26 import json
  27 import logging
  28 import operator
  29 import functools
  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 (
  37     ScalingGroup,
  38     ScalingAlarm,
  39     ScalingPolicy,
  40     ScalingCriteria,
  41     ScalingAlarmRepository,
  42     ScalingGroupRepository,
  43     ScalingPolicyRepository,
  44     ScalingCriteriaRepository,
  45 )
  46 from osm_policy_module.core.exceptions import VdurNotFound
  47
  48 log = logging.getLogger(__name__)
  49
  50
  51 class AutoscalingService:
  52     def __init__(self, config: Config):
  53         self.conf = config
  54         self.db_client = CommonDbClient(config)
  55         self.mon_client = MonClient(config)
  56         self.lcm_client = LcmClient(config)
  57
  58     async def configure_scaling_groups(self, nsr_id: str, vnf_member_index=None):
  59         """
  60         Configures scaling groups for a network service. Creates records in DB. Creates alarms in MON.
  61         :param nsr_id: Network service record id
  62         :return:
  63         """
  64         log.info(
  65             "Configuring scaling groups for network service with nsr_id: %s", nsr_id
  66         )
  67         alarms_created = []
  68         database.db.connect()
  69         try:
  70             with database.db.atomic() as tx:
  71                 try:
  72                     if vnf_member_index is None:
  73                         vnfrs = self.db_client.get_vnfrs(nsr_id)
  74                     else:
  75                         vnfrs = []
  76                         vnfr = self.db_client.get_vnfr(nsr_id, vnf_member_index)
  77                         vnfrs.append(vnfr)
  78                     # vnfrs = self.db_client.get_vnfrs(nsr_id)
  79                     for vnfr in vnfrs:
  80                         log.debug("Processing vnfr: %s", vnfr)
  81                         vnfd = self.db_client.get_vnfd(vnfr["vnfd-id"])
  82                         # TODO: Change for multiple DF support
  83                         df = vnfd.get("df", [{}])[0]
  84                         if "scaling-aspect" not in df:
  85                             log.debug("No scaling aspect present in vnfd")
  86                             continue
  87                         # TODO: Change for multiple instantiation levels support
  88                         scaling_aspects = df["scaling-aspect"]
  89                         all_vnfd_monitoring_params = (
  90                             self._get_all_vnfd_monitoring_params(vnfd)
  91                         )
  92                         for scaling_aspect in scaling_aspects:
  93                             scaling_group_record = self._get_or_create_scaling_group(
  94                                 nsr_id, vnfr["member-vnf-index-ref"], scaling_aspect
  95                             )
  96                             vdurs = self._get_monitored_vdurs(
  97                                 scaling_aspect, vnfr["vdur"]
  98                             )
  99                             for scaling_policy in scaling_aspect.get(
 100                                 "scaling-policy", ()
 101                             ):
 102                                 if scaling_policy["scaling-type"] != "automatic":
 103                                     continue
 104                                 scaling_policy_record = (
 105                                     self._get_or_create_scaling_policy(
 106                                         nsr_id, scaling_policy, scaling_group_record
 107                                     )
 108                                 )
 109
 110                                 for scaling_criteria in scaling_policy[
 111                                     "scaling-criteria"
 112                                 ]:
 113                                     scaling_criteria_record = (
 114                                         self._get_or_create_scaling_criteria(
 115                                             nsr_id,
 116                                             scaling_criteria,
 117                                             scaling_policy_record,
 118                                         )
 119                                     )
 120                                     monitoring_param_ref = scaling_criteria.get(
 121                                         "vnf-monitoring-param-ref"
 122                                     )
 123                                     vnf_monitoring_param = all_vnfd_monitoring_params[
 124                                         monitoring_param_ref
 125                                     ]
 126
 127                                     for vdur in vdurs:
 128                                         vdu_id = vdur["vdu-id-ref"]
 129                                         log.debug("Creating alarm for vdur %s ", vdur)
 130                                         try:
 131                                             ScalingAlarmRepository.get(
 132                                                 ScalingAlarm.vdu_name == vdur["name"],
 133                                                 ScalingCriteria.name
 134                                                 == scaling_criteria["name"],
 135                                                 ScalingPolicy.name
 136                                                 == scaling_policy["name"],
 137                                                 ScalingGroup.nsr_id == nsr_id,
 138                                                 join_classes=[
 139                                                     ScalingCriteria,
 140                                                     ScalingPolicy,
 141                                                     ScalingGroup,
 142                                                 ],
 143                                             )
 144                                             log.debug(
 145                                                 "vdu %s already has an alarm configured",
 146                                                 vdur["name"],
 147                                             )
 148                                             continue
 149                                         except ScalingAlarm.DoesNotExist:
 150                                             pass
 151                                         metric_name = self._get_metric_name(
 152                                             vnf_monitoring_param
 153                                         )
 154
 155                                         db_nsr = self.db_client.get_nsr(nsr_id)
 156                                         nb_scale_op = 0
 157                                         if db_nsr["_admin"].get("scaling-group"):
 158                                             db_nsr_admin = db_nsr["_admin"][
 159                                                 "scaling-group"
 160                                             ]
 161                                             for (
 162                                                 admin_scale_index,
 163                                                 admin_scale_info,
 164                                             ) in enumerate(db_nsr_admin):
 165                                                 if (
 166                                                     admin_scale_info["name"]
 167                                                     == scaling_aspect["name"]
 168                                                 ):
 169                                                     nb_scale_op = admin_scale_info.get(
 170                                                         "nb-scale-op", 0
 171                                                     )
 172                                                     break
 173                                         min_instance_count = 1
 174                                         for vdu_profile in df.get("vdu-profile", ()):
 175                                             if vdu_profile.get("id") == vdu_id:
 176                                                 min_instance_count = int(
 177                                                     vdu_profile.get(
 178                                                         "min-number-of-instances ", 1
 179                                                     )
 180                                                 )
 181                                                 break
 182                                         if nb_scale_op >= min_instance_count:
 183                                             alarm_uuid = (
 184                                                 await self.mon_client.create_alarm(
 185                                                     metric_name=metric_name,
 186                                                     ns_id=nsr_id,
 187                                                     vdu_name=vdur["name"],
 188                                                     vnf_member_index=vnfr[
 189                                                         "member-vnf-index-ref"
 190                                                     ],
 191                                                     threshold=scaling_criteria[
 192                                                         "scale-in-threshold"
 193                                                     ],
 194                                                     operation=scaling_criteria[
 195                                                         "scale-in-relational-operation"
 196                                                     ],
 197                                                     action="scale_in",
 198                                                     vnfr=vnfr,
 199                                                     vnfd=vnfd,
 200                                                 )
 201                                             )
 202                                             alarm = ScalingAlarmRepository.create(
 203                                                 alarm_uuid=alarm_uuid,
 204                                                 action="scale_in",
 205                                                 vnf_member_index=vnfr[
 206                                                     "member-vnf-index-ref"
 207                                                 ],
 208                                                 vdu_name=vdur["name"],
 209                                                 scaling_criteria=scaling_criteria_record,
 210                                             )
 211                                             alarms_created.append(alarm)
 212                                         alarm_uuid = await self.mon_client.create_alarm(
 213                                             metric_name=metric_name,
 214                                             ns_id=nsr_id,
 215                                             vdu_name=vdur["name"],
 216                                             vnf_member_index=vnfr[
 217                                                 "member-vnf-index-ref"
 218                                             ],
 219                                             threshold=scaling_criteria[
 220                                                 "scale-out-threshold"
 221                                             ],
 222                                             operation=scaling_criteria[
 223                                                 "scale-out-relational-operation"
 224                                             ],
 225                                             action="scale_out",
 226                                             vnfr=vnfr,
 227                                             vnfd=vnfd,
 228                                         )
 229                                         alarm = ScalingAlarmRepository.create(
 230                                             alarm_uuid=alarm_uuid,
 231                                             action="scale_out",
 232                                             vnf_member_index=vnfr[
 233                                                 "member-vnf-index-ref"
 234                                             ],
 235                                             vdu_name=vdur["name"],
 236                                             scaling_criteria=scaling_criteria_record,
 237                                         )
 238                                         alarms_created.append(alarm)
 239
 240                 except Exception as e:
 241                     log.exception("Error configuring scaling groups:")
 242                     tx.rollback()
 243                     if len(alarms_created) > 0:
 244                         log.info("Cleaning alarm resources in MON")
 245                         for alarm in alarms_created:
 246                             await self.mon_client.delete_alarm(
 247                                 alarm.scaling_criteria.scaling_policy.scaling_group.nsr_id,
 248                                 alarm.vnf_member_index,
 249                                 alarm.vdu_name,
 250                                 alarm.alarm_uuid,
 251                             )
 252                     raise e
 253         finally:
 254             database.db.close()
 255
 256     async def delete_scaling_groups(self, nsr_id: str, vnf_member_index=None):
 257         log.debug("Deleting scaling groups for network service %s", nsr_id)
 258         database.db.connect()
 259         try:
 260             with database.db.atomic() as tx:
 261                 try:
 262                     if vnf_member_index is None:
 263                         scale_conditions = ScalingGroup.nsr_id == nsr_id
 264                     else:
 265                         query_list = [
 266                             ScalingGroup.nsr_id == nsr_id,
 267                             ScalingGroup.vnf_member_index == vnf_member_index,
 268                         ]
 269                         scale_conditions = functools.reduce(operator.and_, query_list)
 270                     for scaling_group in ScalingGroupRepository.list(scale_conditions):
 271                         for scaling_policy in scaling_group.scaling_policies:
 272                             for scaling_criteria in scaling_policy.scaling_criterias:
 273                                 for alarm in scaling_criteria.scaling_alarms:
 274                                     try:
 275                                         await self.mon_client.delete_alarm(
 276                                             alarm.scaling_criteria.scaling_policy.scaling_group.nsr_id,
 277                                             alarm.vnf_member_index,
 278                                             alarm.vdu_name,
 279                                             alarm.alarm_uuid,
 280                                         )
 281                                     except ValueError:
 282                                         log.exception(
 283                                             "Error deleting alarm in MON %s",
 284                                             alarm.alarm_uuid,
 285                                         )
 286                                     alarm.delete_instance()
 287                                 scaling_criteria.delete_instance()
 288                             scaling_policy.delete_instance()
 289                         scaling_group.delete_instance()
 290
 291                 except Exception as e:
 292                     log.exception("Error deleting scaling groups and alarms:")
 293                     tx.rollback()
 294                     raise e
 295         finally:
 296             database.db.close()
 297
 298     async def delete_orphaned_alarms(self, nsr_id):
 299         log.info("Deleting orphaned scaling alarms for network service %s", nsr_id)
 300         database.db.connect()
 301         try:
 302             with database.db.atomic() as tx:
 303                 try:
 304                     for scaling_group in ScalingGroupRepository.list(
 305                         ScalingGroup.nsr_id == nsr_id
 306                     ):
 307                         for scaling_policy in scaling_group.scaling_policies:
 308                             for scaling_criteria in scaling_policy.scaling_criterias:
 309                                 for alarm in scaling_criteria.scaling_alarms:
 310                                     try:
 311                                         self.db_client.get_vdur(
 312                                             nsr_id,
 313                                             alarm.vnf_member_index,
 314                                             alarm.vdu_name,
 315                                         )
 316                                     except VdurNotFound:
 317                                         log.debug(
 318                                             "Deleting orphaned scaling alarm %s",
 319                                             alarm.alarm_uuid,
 320                                         )
 321                                         try:
 322                                             await self.mon_client.delete_alarm(
 323                                                 alarm.scaling_criteria.scaling_policy.scaling_group.nsr_id,
 324                                                 alarm.vnf_member_index,
 325                                                 alarm.vdu_name,
 326                                                 alarm.alarm_uuid,
 327                                             )
 328                                         except ValueError:
 329                                             log.exception(
 330                                                 "Error deleting alarm in MON %s",
 331                                                 alarm.alarm_uuid,
 332                                             )
 333                                         alarm.delete_instance()
 334
 335                 except Exception as e:
 336                     log.exception("Error deleting orphaned alarms:")
 337                     tx.rollback()
 338                     raise e
 339         finally:
 340             database.db.close()
 341
 342     async def handle_alarm(self, alarm_uuid: str, status: str):
 343         await self.update_alarm_status(alarm_uuid, status)
 344         await self.evaluate_policy(alarm_uuid)
 345
 346     async def update_alarm_status(self, alarm_uuid: str, status: str):
 347         database.db.connect()
 348         try:
 349             with database.db.atomic():
 350                 alarm = ScalingAlarmRepository.get(
 351                     ScalingAlarm.alarm_uuid == alarm_uuid
 352                 )
 353                 alarm.last_status = status
 354                 alarm.save()
 355         except ScalingAlarm.DoesNotExist:
 356             log.debug(
 357                 "There is no autoscaling action configured for alarm %s.", alarm_uuid
 358             )
 359         finally:
 360             database.db.close()
 361
 362     async def evaluate_policy(self, alarm_uuid):
 363         database.db.connect()
 364         try:
 365             if self.conf.get("autoscale", "enabled") == "True":
 366                 with database.db.atomic():
 367                     alarm = ScalingAlarmRepository.get(
 368                         ScalingAlarm.alarm_uuid == alarm_uuid
 369                     )
 370                     vnf_member_index = alarm.vnf_member_index
 371                     action = alarm.action
 372                     scaling_policy = alarm.scaling_criteria.scaling_policy
 373                     if not scaling_policy.enabled:
 374                         return
 375                     if action == "scale_in":
 376                         operation = scaling_policy.scale_in_operation
 377                     elif action == "scale_out":
 378                         operation = scaling_policy.scale_out_operation
 379                     else:
 380                         raise Exception("Unknown alarm action {}".format(alarm.action))
 381                     alarms = ScalingAlarmRepository.list(
 382                         ScalingAlarm.scaling_criteria == alarm.scaling_criteria,
 383                         ScalingAlarm.action == alarm.action,
 384                         ScalingAlarm.vnf_member_index == vnf_member_index,
 385                         ScalingAlarm.vdu_name == alarm.vdu_name,
 386                     )
 387                     statuses = []
 388                     for alarm in alarms:
 389                         statuses.append(alarm.last_status)
 390                     if (operation == "AND" and set(statuses) == {"alarm"}) or (
 391                         operation == "OR" and "alarm" in statuses
 392                     ):
 393                         delta = datetime.datetime.now() - scaling_policy.last_scale
 394                         if delta.total_seconds() > scaling_policy.cooldown_time:
 395                             log.info(
 396                                 "Sending %s action message for ns: %s",
 397                                 alarm.action,
 398                                 scaling_policy.scaling_group.nsr_id,
 399                             )
 400                             await self.lcm_client.scale(
 401                                 scaling_policy.scaling_group.nsr_id,
 402                                 scaling_policy.scaling_group.name,
 403                                 vnf_member_index,
 404                                 action,
 405                             )
 406                             scaling_policy.last_scale = datetime.datetime.now()
 407                             scaling_policy.save()
 408
 409         except ScalingAlarm.DoesNotExist:
 410             log.debug(
 411                 "There is no autoscaling action configured for alarm %s.", alarm_uuid
 412             )
 413         finally:
 414             database.db.close()
 415
 416     def _get_all_vnfd_monitoring_params(self, vnfd):
 417         all_monitoring_params = {}
 418         for ivld in vnfd.get("int-virtual-link-desc", ()):
 419             for mp in ivld.get("monitoring-parameters", ()):
 420                 all_monitoring_params[mp.get("id")] = mp
 421
 422         for vdu in vnfd.get("vdu", ()):
 423             for mp in vdu.get("monitoring-parameter", ()):
 424                 all_monitoring_params[mp.get("id")] = mp
 425
 426         for df in vnfd.get("df", ()):
 427             for mp in df.get("monitoring-parameter", ()):
 428                 all_monitoring_params[mp.get("id")] = mp
 429
 430         return all_monitoring_params
 431
 432     def _get_or_create_scaling_group(
 433         self, nsr_id: str, vnf_member_index: str, scaling_aspect: dict
 434     ):
 435         try:
 436             scaling_group_record = ScalingGroupRepository.get(
 437                 ScalingGroup.nsr_id == nsr_id,
 438                 ScalingGroup.vnf_member_index == vnf_member_index,
 439                 ScalingGroup.name == scaling_aspect["name"],
 440             )
 441             log.debug("Found existing scaling group record in DB...")
 442         except ScalingGroup.DoesNotExist:
 443             log.debug("Creating scaling group record in DB...")
 444             scaling_group_record = ScalingGroupRepository.create(
 445                 nsr_id=nsr_id,
 446                 vnf_member_index=vnf_member_index,
 447                 name=scaling_aspect["name"],
 448                 content=json.dumps(scaling_aspect),
 449             )
 450             log.debug(
 451                 "Created scaling group record in DB : nsr_id=%s, vnf_member_index=%s, name=%s",
 452                 scaling_group_record.nsr_id,
 453                 scaling_group_record.vnf_member_index,
 454                 scaling_group_record.name,
 455             )
 456         return scaling_group_record
 457
 458     def _get_or_create_scaling_policy(
 459         self, nsr_id: str, scaling_policy: dict, scaling_group_record: ScalingGroup
 460     ):
 461         try:
 462             scaling_policy_record = ScalingPolicyRepository.get(
 463                 ScalingPolicy.name == scaling_policy["name"],
 464                 ScalingGroup.id == scaling_group_record.id,
 465                 join_classes=[ScalingGroup],
 466             )
 467             log.debug("Found existing scaling policy record in DB...")
 468         except ScalingPolicy.DoesNotExist:
 469             log.debug("Creating scaling policy record in DB...")
 470             scaling_policy_record = ScalingPolicyRepository.create(
 471                 nsr_id=nsr_id,
 472                 name=scaling_policy["name"],
 473                 cooldown_time=scaling_policy["cooldown-time"],
 474                 scaling_group=scaling_group_record,
 475             )
 476             if "scale-in-operation-type" in scaling_policy:
 477                 scaling_policy_record.scale_in_operation = scaling_policy[
 478                     "scale-in-operation-type"
 479                 ]
 480             if "scale-out-operation-type" in scaling_policy:
 481                 scaling_policy_record.scale_out_operation = scaling_policy[
 482                     "scale-out-operation-type"
 483                 ]
 484             if "enabled" in scaling_policy:
 485                 scaling_policy_record.enabled = scaling_policy["enabled"]
 486             scaling_policy_record.save()
 487             log.debug(
 488                 "Created scaling policy record in DB : name=%s, scaling_group.name=%s",
 489                 scaling_policy_record.name,
 490                 scaling_policy_record.scaling_group.name,
 491             )
 492         return scaling_policy_record
 493
 494     def _get_or_create_scaling_criteria(
 495         self, nsr_id: str, scaling_criteria: dict, scaling_policy_record: ScalingPolicy
 496     ):
 497         try:
 498             scaling_criteria_record = ScalingCriteriaRepository.get(
 499                 ScalingPolicy.id == scaling_policy_record.id,
 500                 ScalingCriteria.name == scaling_criteria["name"],
 501                 join_classes=[ScalingPolicy],
 502             )
 503             log.debug("Found existing scaling criteria record in DB...")
 504         except ScalingCriteria.DoesNotExist:
 505             log.debug("Creating scaling criteria record in DB...")
 506             scaling_criteria_record = ScalingCriteriaRepository.create(
 507                 nsr_id=nsr_id,
 508                 name=scaling_criteria["name"],
 509                 scaling_policy=scaling_policy_record,
 510             )
 511             log.debug(
 512                 "Created scaling criteria record in DB : name=%s, scaling_policy.name=%s",
 513                 scaling_criteria_record.name,
 514                 scaling_criteria_record.scaling_policy.name,
 515             )
 516         return scaling_criteria_record
 517
 518     def _get_monitored_vdurs(self, scaling_aspect: dict, vdurs):
 519         all_monitored_vdus = set()
 520         for delta in scaling_aspect.get("aspect-delta-details", {}).get("deltas", ()):
 521             for vdu_delta in delta.get("vdu-delta", ()):
 522                 all_monitored_vdus.add(vdu_delta.get("id"))
 523
 524         monitored_vdurs = list(
 525             filter(lambda vdur: vdur["vdu-id-ref"] in all_monitored_vdus, vdurs)
 526         )
 527
 528         if not monitored_vdurs:
 529             log.warning(
 530                 "Scaling criteria is referring to a vnf-monitoring-param that does not "
 531                 "contain a reference to a vdu or vnf metric."
 532             )
 533         return monitored_vdurs
 534
 535     def _get_metric_name(self, vnf_monitoring_param: dict):
 536         if "performance-metric" in vnf_monitoring_param:
 537             return vnf_monitoring_param["performance-metric"]
 538         raise ValueError(
 539             "No metric name found for vnf_monitoring_param %s"
 540             % vnf_monitoring_param["id"]
 541         )