| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 1 | # -*- coding: utf-8 -*- |
| Atul Agarwal | db8c105 | 2020-04-28 15:42:28 +0530 | [diff] [blame] | 2 | # pylint: disable=no-member |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 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 json |
| 27 | import logging |
| 28 | |
| 29 | import requests |
| Atul Agarwal | 46e7efd | 2021-10-18 17:22:58 +0000 | [diff] [blame] | 30 | from requests.exceptions import ConnectionError, RequestException |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 31 | |
| 32 | from osm_policy_module.common.common_db_client import CommonDbClient |
| 33 | from osm_policy_module.common.lcm_client import LcmClient |
| 34 | from osm_policy_module.common.mon_client import MonClient |
| 35 | from osm_policy_module.core import database |
| 36 | from osm_policy_module.core.config import Config |
| garciadeblas | 4584f8e | 2021-05-14 16:50:06 +0200 | [diff] [blame] | 37 | from osm_policy_module.core.database import ( |
| 38 | VnfAlarm, |
| 39 | VnfAlarmRepository, |
| 40 | AlarmActionRepository, |
| 41 | ) |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 42 | from osm_policy_module.core.exceptions import VdurNotFound |
| 43 | |
| 44 | log = logging.getLogger(__name__) |
| 45 | |
| 46 | |
| 47 | class AlarmingService: |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 48 | def __init__(self, config: Config, loop=None): |
| 49 | self.conf = config |
| 50 | if not loop: |
| 51 | loop = asyncio.get_event_loop() |
| 52 | self.loop = loop |
| 53 | self.db_client = CommonDbClient(config) |
| 54 | self.mon_client = MonClient(config, loop=self.loop) |
| 55 | self.lcm_client = LcmClient(config, loop=self.loop) |
| 56 | |
| 57 | async def configure_vnf_alarms(self, nsr_id: str): |
| 58 | log.info("Configuring vnf alarms for network service %s", nsr_id) |
| 59 | alarms_created = [] |
| Benjamin Diaz | acac755 | 2019-05-23 14:19:06 -0300 | [diff] [blame] | 60 | database.db.connect() |
| 61 | try: |
| 62 | with database.db.atomic(): |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 63 | vnfrs = self.db_client.get_vnfrs(nsr_id) |
| 64 | for vnfr in vnfrs: |
| 65 | log.debug("Processing vnfr: %s", vnfr) |
| garciadeblas | 4584f8e | 2021-05-14 16:50:06 +0200 | [diff] [blame] | 66 | vnfd = self.db_client.get_vnfd(vnfr["vnfd-id"]) |
| 67 | for vdur in vnfr["vdur"]: |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 68 | vdu = next( |
| 69 | filter( |
| garciadeblas | 4584f8e | 2021-05-14 16:50:06 +0200 | [diff] [blame] | 70 | lambda vdu: vdu["id"] == vdur["vdu-id-ref"], vnfd["vdu"] |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 71 | ) |
| 72 | ) |
| garciadeblas | 4584f8e | 2021-05-14 16:50:06 +0200 | [diff] [blame] | 73 | if "alarm" in vdu: |
| 74 | alarm_descriptors = vdu["alarm"] |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 75 | for alarm_descriptor in alarm_descriptors: |
| 76 | try: |
| 77 | VnfAlarmRepository.get( |
| garciadeblas | 4584f8e | 2021-05-14 16:50:06 +0200 | [diff] [blame] | 78 | VnfAlarm.alarm_id |
| 79 | == alarm_descriptor["alarm-id"], |
| 80 | VnfAlarm.vnf_member_index |
| 81 | == vnfr["member-vnf-index-ref"], |
| 82 | VnfAlarm.vdu_name == vdur["name"], |
| 83 | VnfAlarm.nsr_id == nsr_id, |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 84 | ) |
| garciadeblas | 4584f8e | 2021-05-14 16:50:06 +0200 | [diff] [blame] | 85 | log.debug( |
| 86 | "vdu %s already has an alarm configured with same id %s", |
| 87 | vdur["name"], |
| 88 | alarm_descriptor["alarm-id"], |
| 89 | ) |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 90 | continue |
| 91 | except VnfAlarm.DoesNotExist: |
| 92 | pass |
| 93 | vnf_monitoring_param = next( |
| 94 | filter( |
| garciadeblas | 4584f8e | 2021-05-14 16:50:06 +0200 | [diff] [blame] | 95 | lambda param: param["id"] |
| 96 | == alarm_descriptor["vnf-monitoring-param-ref"], |
| 97 | vdu.get("monitoring-parameter", []), |
| garciaale | 7ce998f | 2020-12-03 18:05:24 -0300 | [diff] [blame] | 98 | ), |
| garciadeblas | 4584f8e | 2021-05-14 16:50:06 +0200 | [diff] [blame] | 99 | {}, |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 100 | ) |
| garciadeblas | 4584f8e | 2021-05-14 16:50:06 +0200 | [diff] [blame] | 101 | metric_name = self._get_metric_name( |
| 102 | vnf_monitoring_param |
| 103 | ) |
| Atul Agarwal | e9228cf | 2021-03-19 10:11:38 +0000 | [diff] [blame] | 104 | alarm_action = dict() |
| 105 | for action_type in ["ok", "insufficient-data", "alarm"]: |
| 106 | if "actions" in alarm_descriptor and action_type in alarm_descriptor["actions"]: |
| 107 | for url in alarm_descriptor["actions"][action_type]: |
| 108 | if "webhook" in alarm_action: |
| 109 | alarm_action["webhook"].append(url["url"]) |
| 110 | else: |
| 111 | alarm_action["webhook"] = [url["url"]] |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 112 | alarm_uuid = await self.mon_client.create_alarm( |
| Gianpietro Lavado | 4161019 | 2019-12-06 15:46:23 +0000 | [diff] [blame] | 113 | metric_name=metric_name, |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 114 | ns_id=nsr_id, |
| garciadeblas | 4584f8e | 2021-05-14 16:50:06 +0200 | [diff] [blame] | 115 | vdu_name=vdur["name"], |
| 116 | vnf_member_index=vnfr["member-vnf-index-ref"], |
| 117 | threshold=alarm_descriptor["value"], |
| 118 | operation=alarm_descriptor["operation"], |
| Atul Agarwal | e9228cf | 2021-03-19 10:11:38 +0000 | [diff] [blame] | 119 | action=str(alarm_action), |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 120 | ) |
| 121 | alarm = VnfAlarmRepository.create( |
| garciadeblas | 4584f8e | 2021-05-14 16:50:06 +0200 | [diff] [blame] | 122 | alarm_id=alarm_descriptor["alarm-id"], |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 123 | alarm_uuid=alarm_uuid, |
| 124 | nsr_id=nsr_id, |
| garciadeblas | 4584f8e | 2021-05-14 16:50:06 +0200 | [diff] [blame] | 125 | vnf_member_index=vnfr["member-vnf-index-ref"], |
| 126 | vdu_name=vdur["name"], |
| sritharan | 3fbf2fb | 2022-04-07 05:20:38 +0000 | [diff] [blame^] | 127 | last_action='insufficient-data', |
| 128 | id_suffix=0, |
| 129 | ok_ack=False, |
| 130 | alarm_ack=False |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 131 | ) |
| garciadeblas | 4584f8e | 2021-05-14 16:50:06 +0200 | [diff] [blame] | 132 | for action_type in ["ok", "insufficient-data", "alarm"]: |
| 133 | if ( |
| 134 | "actions" in alarm_descriptor |
| 135 | and action_type in alarm_descriptor["actions"] |
| 136 | ): |
| 137 | for url in alarm_descriptor["actions"][ |
| 138 | action_type |
| 139 | ]: |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 140 | AlarmActionRepository.create( |
| 141 | type=action_type, |
| garciadeblas | 4584f8e | 2021-05-14 16:50:06 +0200 | [diff] [blame] | 142 | url=url["url"], |
| 143 | alarm=alarm, |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 144 | ) |
| 145 | alarms_created.append(alarm) |
| 146 | |
| Benjamin Diaz | acac755 | 2019-05-23 14:19:06 -0300 | [diff] [blame] | 147 | except Exception as e: |
| 148 | log.exception("Error configuring VNF alarms:") |
| 149 | if len(alarms_created) > 0: |
| 150 | log.debug("Cleaning alarm resources in MON") |
| 151 | for alarm in alarms_created: |
| 152 | try: |
| garciadeblas | 4584f8e | 2021-05-14 16:50:06 +0200 | [diff] [blame] | 153 | await self.mon_client.delete_alarm( |
| 154 | alarm.nsr_id, |
| 155 | alarm.vnf_member_index, |
| 156 | alarm.vdu_name, |
| 157 | alarm.alarm_uuid, |
| 158 | ) |
| Benjamin Diaz | acac755 | 2019-05-23 14:19:06 -0300 | [diff] [blame] | 159 | except ValueError: |
| garciadeblas | 4584f8e | 2021-05-14 16:50:06 +0200 | [diff] [blame] | 160 | log.exception( |
| 161 | "Error deleting alarm in MON %s", alarm.alarm_uuid |
| 162 | ) |
| Benjamin Diaz | acac755 | 2019-05-23 14:19:06 -0300 | [diff] [blame] | 163 | raise e |
| 164 | finally: |
| 165 | database.db.close() |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 166 | |
| 167 | async def delete_orphaned_alarms(self, nsr_id): |
| garciaale | 7ce998f | 2020-12-03 18:05:24 -0300 | [diff] [blame] | 168 | # TODO: Review as it seems this code is never called |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 169 | log.info("Deleting orphaned vnf alarms for network service %s", nsr_id) |
| 170 | database.db.connect() |
| Benjamin Diaz | acac755 | 2019-05-23 14:19:06 -0300 | [diff] [blame] | 171 | try: |
| 172 | with database.db.atomic(): |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 173 | for alarm in VnfAlarmRepository.list(VnfAlarm.nsr_id == nsr_id): |
| 174 | try: |
| garciadeblas | 4584f8e | 2021-05-14 16:50:06 +0200 | [diff] [blame] | 175 | self.db_client.get_vdur( |
| 176 | nsr_id, alarm.vnf_member_index, alarm.vdu_name |
| 177 | ) |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 178 | except VdurNotFound: |
| 179 | log.debug("Deleting orphaned alarm %s", alarm.alarm_uuid) |
| 180 | try: |
| 181 | await self.mon_client.delete_alarm( |
| 182 | alarm.nsr_id, |
| 183 | alarm.vnf_member_index, |
| 184 | alarm.vdu_name, |
| garciadeblas | 4584f8e | 2021-05-14 16:50:06 +0200 | [diff] [blame] | 185 | alarm.alarm_uuid, |
| 186 | ) |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 187 | except ValueError: |
| garciadeblas | 4584f8e | 2021-05-14 16:50:06 +0200 | [diff] [blame] | 188 | log.exception( |
| 189 | "Error deleting alarm in MON %s", alarm.alarm_uuid |
| 190 | ) |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 191 | alarm.delete_instance() |
| Benjamin Diaz | acac755 | 2019-05-23 14:19:06 -0300 | [diff] [blame] | 192 | except Exception as e: |
| 193 | log.exception("Error deleting orphaned alarms:") |
| 194 | raise e |
| 195 | finally: |
| 196 | database.db.close() |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 197 | |
| 198 | async def delete_vnf_alarms(self, nsr_id): |
| 199 | log.info("Deleting vnf alarms for network service %s", nsr_id) |
| 200 | database.db.connect() |
| Benjamin Diaz | acac755 | 2019-05-23 14:19:06 -0300 | [diff] [blame] | 201 | try: |
| 202 | with database.db.atomic(): |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 203 | for alarm in VnfAlarmRepository.list(VnfAlarm.nsr_id == nsr_id): |
| 204 | log.debug("Deleting vnf alarm %s", alarm.alarm_uuid) |
| 205 | try: |
| 206 | await self.mon_client.delete_alarm( |
| 207 | alarm.nsr_id, |
| 208 | alarm.vnf_member_index, |
| 209 | alarm.vdu_name, |
| garciadeblas | 4584f8e | 2021-05-14 16:50:06 +0200 | [diff] [blame] | 210 | alarm.alarm_uuid, |
| 211 | ) |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 212 | except ValueError: |
| garciadeblas | 4584f8e | 2021-05-14 16:50:06 +0200 | [diff] [blame] | 213 | log.exception( |
| 214 | "Error deleting alarm in MON %s", alarm.alarm_uuid |
| 215 | ) |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 216 | alarm.delete_instance() |
| 217 | |
| Benjamin Diaz | acac755 | 2019-05-23 14:19:06 -0300 | [diff] [blame] | 218 | except Exception as e: |
| 219 | log.exception("Error deleting vnf alarms:") |
| 220 | raise e |
| 221 | finally: |
| 222 | database.db.close() |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 223 | |
| 224 | async def handle_alarm(self, alarm_uuid: str, status: str, payload: dict): |
| Atul Agarwal | 10e97a4 | 2021-10-24 17:05:55 +0000 | [diff] [blame] | 225 | alert_timeout = int(self.conf.get('alert', 'timeout')) |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 226 | database.db.connect() |
| 227 | try: |
| 228 | with database.db.atomic(): |
| 229 | alarm = VnfAlarmRepository.get(VnfAlarm.alarm_uuid == alarm_uuid) |
| garciadeblas | 4584f8e | 2021-05-14 16:50:06 +0200 | [diff] [blame] | 230 | log.debug( |
| 231 | "Handling vnf alarm %s with status %s", alarm.alarm_id, status |
| 232 | ) |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 233 | for action in alarm.actions: |
| sritharan | 3fbf2fb | 2022-04-07 05:20:38 +0000 | [diff] [blame^] | 234 | """ |
| 235 | Compares the current status with the last_action status. |
| 236 | If both the status are 'alarm', it avoid sending repetitive alarm notification. |
| 237 | If both the status are 'ok', it avoid sending repetitive ok notification. |
| 238 | """ |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 239 | if action.type == status: |
| sritharan | 3fbf2fb | 2022-04-07 05:20:38 +0000 | [diff] [blame^] | 240 | if bool(self.conf.get('alert', 'enhanced_alarms')): |
| 241 | if ((status != "ok" or (status == "ok" and alarm.ok_ack is False)) and |
| 242 | (status != "alarm" or (status == "alarm" and alarm.alarm_ack is False))): |
| 243 | log.info( |
| 244 | "Executing request to url %s for vnf alarm %s with status %s", |
| 245 | action.url, |
| 246 | alarm.alarm_id, |
| 247 | status |
| 248 | ) |
| 249 | try: |
| 250 | if status == "alarm" and alarm.last_action == "ok": |
| 251 | alarm.id_suffix += 1 |
| 252 | alarm.ok_ack = False |
| 253 | if status == "ok" and alarm.last_action == "alarm": |
| 254 | alarm.alarm_ack = False |
| 255 | alarm.last_action = status |
| 256 | alarm.save() |
| 257 | except Exception as e: |
| 258 | log.exception(e) |
| 259 | |
| 260 | payload["notify_details"]["alarm_number"] = alarm.id_suffix |
| 261 | headers = {"content-type": "application/json"} |
| 262 | try: |
| 263 | resp = requests.post(url=action.url, data=json.dumps(payload), |
| 264 | headers=headers, verify=False, timeout=alert_timeout) |
| 265 | log.info("Response %s", resp) |
| 266 | if resp.status_code == 200: |
| 267 | if status == "ok": |
| 268 | alarm.ok_ack = True |
| 269 | alarm.save() |
| 270 | if status == "alarm": |
| 271 | alarm.alarm_ack = True |
| 272 | alarm.save() |
| 273 | if status == "insufficient-data": |
| 274 | alarm.alarm_ack = False |
| 275 | alarm.ok_ack = False |
| 276 | alarm.save() |
| 277 | except RequestException as e: |
| 278 | log.info("Error: RequestException while connecting to url %s", action.url) |
| 279 | log.debug("RequestException %s", e) |
| 280 | |
| 281 | except ConnectionError: |
| 282 | log.exception("Error connecting to url %s", action.url) |
| 283 | else: |
| 284 | log.info( |
| 285 | "Executing request to url %s for vnf alarm %s with status %s", |
| 286 | action.url, |
| 287 | alarm.alarm_id, |
| 288 | status |
| 289 | ) |
| 290 | try: |
| 291 | requests.post(url=action.url, json=json.dumps(payload), timeout=alert_timeout) |
| 292 | except RequestException as e: |
| 293 | log.info("Error: RequestException while connecting to url %s", action.url) |
| 294 | log.debug("RequestException %s", e) |
| 295 | except ConnectionError: |
| 296 | log.exception("Error connecting to url %s", action.url) |
| Atul Agarwal | db8c105 | 2020-04-28 15:42:28 +0530 | [diff] [blame] | 297 | |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 298 | except VnfAlarm.DoesNotExist: |
| garciadeblas | 4584f8e | 2021-05-14 16:50:06 +0200 | [diff] [blame] | 299 | log.debug( |
| 300 | "There is no alarming action configured for alarm %s.", alarm_uuid |
| 301 | ) |
| Benjamin Diaz | f7451f8 | 2019-04-01 14:56:26 -0300 | [diff] [blame] | 302 | finally: |
| 303 | database.db.close() |
| Gianpietro Lavado | 4161019 | 2019-12-06 15:46:23 +0000 | [diff] [blame] | 304 | |
| garciaale | 7ce998f | 2020-12-03 18:05:24 -0300 | [diff] [blame] | 305 | def _get_metric_name(self, vnf_monitoring_param: dict): |
| garciadeblas | 4584f8e | 2021-05-14 16:50:06 +0200 | [diff] [blame] | 306 | if "performance-metric" in vnf_monitoring_param: |
| 307 | return vnf_monitoring_param["performance-metric"] |
| 308 | raise ValueError( |
| 309 | "No metric name found for vnf_monitoring_param %s" |
| 310 | % vnf_monitoring_param["id"] |
| 311 | ) |