| 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 |
1 |
import json |
| 26 |
1 |
import logging |
| 27 |
1 |
import operator |
| 28 |
1 |
import functools |
| 29 |
|
|
| 30 |
1 |
import requests |
| 31 |
1 |
from requests.exceptions import ConnectionError, RequestException |
| 32 |
|
|
| 33 |
1 |
from osm_policy_module.common.common_db_client import CommonDbClient |
| 34 |
1 |
from osm_policy_module.common.lcm_client import LcmClient |
| 35 |
1 |
from osm_policy_module.common.mon_client import MonClient |
| 36 |
1 |
from osm_policy_module.core import database |
| 37 |
1 |
from osm_policy_module.core.config import Config |
| 38 |
1 |
from osm_policy_module.core.database import ( |
| 39 |
|
VnfAlarm, |
| 40 |
|
VnfAlarmRepository, |
| 41 |
|
AlarmActionRepository, |
| 42 |
|
) |
| 43 |
1 |
from osm_policy_module.core.exceptions import VdurNotFound |
| 44 |
|
|
| 45 |
1 |
log = logging.getLogger(__name__) |
| 46 |
|
|
| 47 |
|
|
| 48 |
1 |
class AlarmingService: |
| 49 |
1 |
def __init__(self, config: Config): |
| 50 |
1 |
self.conf = config |
| 51 |
1 |
self.db_client = CommonDbClient(config) |
| 52 |
1 |
self.mon_client = MonClient(config) |
| 53 |
1 |
self.lcm_client = LcmClient(config) |
| 54 |
|
|
| 55 |
1 |
async def configure_vnf_alarms(self, nsr_id: str, vnf_member_index=None): |
| 56 |
1 |
log.info("Configuring vnf alarms for network service %s", nsr_id) |
| 57 |
1 |
alarms_created = [] |
| 58 |
1 |
database.db.connect() |
| 59 |
1 |
try: |
| 60 |
1 |
with database.db.atomic(): |
| 61 |
1 |
if vnf_member_index is None: |
| 62 |
1 |
vnfrs = self.db_client.get_vnfrs(nsr_id) |
| 63 |
|
else: |
| 64 |
0 |
vnfrs = [] |
| 65 |
0 |
vnfr = self.db_client.get_vnfr(nsr_id, vnf_member_index) |
| 66 |
0 |
vnfrs.append(vnfr) |
| 67 |
|
# vnfrs = self.db_client.get_vnfrs(nsr_id) |
| 68 |
1 |
for vnfr in vnfrs: |
| 69 |
1 |
log.debug("Processing vnfr: %s", vnfr) |
| 70 |
1 |
vnfd = self.db_client.get_vnfd(vnfr["vnfd-id"]) |
| 71 |
1 |
for vdur in vnfr["vdur"]: |
| 72 |
1 |
vdu = next( |
| 73 |
|
filter( |
| 74 |
|
lambda vdu: vdu["id"] == vdur["vdu-id-ref"], vnfd["vdu"] |
| 75 |
|
) |
| 76 |
|
) |
| 77 |
1 |
if "alarm" in vdu: |
| 78 |
1 |
alarm_descriptors = vdu["alarm"] |
| 79 |
1 |
for alarm_descriptor in alarm_descriptors: |
| 80 |
1 |
try: |
| 81 |
1 |
VnfAlarmRepository.get( |
| 82 |
|
VnfAlarm.alarm_id |
| 83 |
|
== alarm_descriptor["alarm-id"], |
| 84 |
|
VnfAlarm.vnf_member_index |
| 85 |
|
== vnfr["member-vnf-index-ref"], |
| 86 |
|
VnfAlarm.vdu_name == vdur["name"], |
| 87 |
|
VnfAlarm.nsr_id == nsr_id, |
| 88 |
|
) |
| 89 |
0 |
log.debug( |
| 90 |
|
"vdu %s already has an alarm configured with same id %s", |
| 91 |
|
vdur["name"], |
| 92 |
|
alarm_descriptor["alarm-id"], |
| 93 |
|
) |
| 94 |
0 |
continue |
| 95 |
1 |
except VnfAlarm.DoesNotExist: |
| 96 |
1 |
pass |
| 97 |
1 |
vnf_monitoring_param = next( |
| 98 |
|
filter( |
| 99 |
|
lambda param: param["id"] |
| 100 |
|
== alarm_descriptor["vnf-monitoring-param-ref"], |
| 101 |
|
vdu.get("monitoring-parameter", []), |
| 102 |
|
), |
| 103 |
|
{}, |
| 104 |
|
) |
| 105 |
1 |
metric_name = self._get_metric_name( |
| 106 |
|
vnf_monitoring_param |
| 107 |
|
) |
| 108 |
1 |
alarm_action = dict() |
| 109 |
1 |
for action_type in ["ok", "insufficient-data", "alarm"]: |
| 110 |
1 |
if ( |
| 111 |
|
"actions" in alarm_descriptor |
| 112 |
|
and action_type in alarm_descriptor["actions"] |
| 113 |
|
): |
| 114 |
1 |
for url in alarm_descriptor["actions"][ |
| 115 |
|
action_type |
| 116 |
|
]: |
| 117 |
1 |
if "webhook" in alarm_action: |
| 118 |
1 |
alarm_action["webhook"].append( |
| 119 |
|
url["url"] |
| 120 |
|
) |
| 121 |
|
else: |
| 122 |
1 |
alarm_action["webhook"] = [url["url"]] |
| 123 |
1 |
alarm_uuid = await self.mon_client.create_alarm( |
| 124 |
|
metric_name=metric_name, |
| 125 |
|
ns_id=nsr_id, |
| 126 |
|
vdu_name=vdur["name"], |
| 127 |
|
vnf_member_index=vnfr["member-vnf-index-ref"], |
| 128 |
|
threshold=alarm_descriptor["value"], |
| 129 |
|
operation=alarm_descriptor["operation"], |
| 130 |
|
action=str(alarm_action), |
| 131 |
|
) |
| 132 |
1 |
alarm = VnfAlarmRepository.create( |
| 133 |
|
alarm_id=alarm_descriptor["alarm-id"], |
| 134 |
|
alarm_uuid=alarm_uuid, |
| 135 |
|
nsr_id=nsr_id, |
| 136 |
|
vnf_member_index=vnfr["member-vnf-index-ref"], |
| 137 |
|
vdu_name=vdur["name"], |
| 138 |
|
last_action="insufficient-data", |
| 139 |
|
id_suffix=0, |
| 140 |
|
ok_ack=False, |
| 141 |
|
alarm_ack=False, |
| 142 |
|
) |
| 143 |
1 |
for action_type in ["ok", "insufficient-data", "alarm"]: |
| 144 |
1 |
if ( |
| 145 |
|
"actions" in alarm_descriptor |
| 146 |
|
and action_type in alarm_descriptor["actions"] |
| 147 |
|
): |
| 148 |
1 |
for url in alarm_descriptor["actions"][ |
| 149 |
|
action_type |
| 150 |
|
]: |
| 151 |
1 |
AlarmActionRepository.create( |
| 152 |
|
type=action_type, |
| 153 |
|
url=url["url"], |
| 154 |
|
alarm=alarm, |
| 155 |
|
) |
| 156 |
1 |
alarms_created.append(alarm) |
| 157 |
|
|
| 158 |
0 |
except Exception as e: |
| 159 |
0 |
log.exception("Error configuring VNF alarms:") |
| 160 |
0 |
if len(alarms_created) > 0: |
| 161 |
0 |
log.debug("Cleaning alarm resources in MON") |
| 162 |
0 |
for alarm in alarms_created: |
| 163 |
0 |
try: |
| 164 |
0 |
await self.mon_client.delete_alarm( |
| 165 |
|
alarm.nsr_id, |
| 166 |
|
alarm.vnf_member_index, |
| 167 |
|
alarm.vdu_name, |
| 168 |
|
alarm.alarm_uuid, |
| 169 |
|
) |
| 170 |
0 |
except ValueError: |
| 171 |
0 |
log.exception( |
| 172 |
|
"Error deleting alarm in MON %s", alarm.alarm_uuid |
| 173 |
|
) |
| 174 |
0 |
raise e |
| 175 |
|
finally: |
| 176 |
1 |
database.db.close() |
| 177 |
|
|
| 178 |
1 |
async def delete_orphaned_alarms(self, nsr_id): |
| 179 |
|
# TODO: Review as it seems this code is never called |
| 180 |
0 |
log.info("Deleting orphaned vnf alarms for network service %s", nsr_id) |
| 181 |
0 |
database.db.connect() |
| 182 |
0 |
try: |
| 183 |
0 |
with database.db.atomic(): |
| 184 |
0 |
for alarm in VnfAlarmRepository.list(VnfAlarm.nsr_id == nsr_id): |
| 185 |
0 |
try: |
| 186 |
0 |
self.db_client.get_vdur( |
| 187 |
|
nsr_id, alarm.vnf_member_index, alarm.vdu_name |
| 188 |
|
) |
| 189 |
0 |
except VdurNotFound: |
| 190 |
0 |
log.debug("Deleting orphaned alarm %s", alarm.alarm_uuid) |
| 191 |
0 |
try: |
| 192 |
0 |
await self.mon_client.delete_alarm( |
| 193 |
|
alarm.nsr_id, |
| 194 |
|
alarm.vnf_member_index, |
| 195 |
|
alarm.vdu_name, |
| 196 |
|
alarm.alarm_uuid, |
| 197 |
|
) |
| 198 |
0 |
except ValueError: |
| 199 |
0 |
log.exception( |
| 200 |
|
"Error deleting alarm in MON %s", alarm.alarm_uuid |
| 201 |
|
) |
| 202 |
0 |
alarm.delete_instance() |
| 203 |
0 |
except Exception as e: |
| 204 |
0 |
log.exception("Error deleting orphaned alarms:") |
| 205 |
0 |
raise e |
| 206 |
|
finally: |
| 207 |
0 |
database.db.close() |
| 208 |
|
|
| 209 |
1 |
async def delete_vnf_alarms(self, nsr_id, vnf_member_index=None): |
| 210 |
0 |
log.info("Deleting vnf alarms for network service %s", nsr_id) |
| 211 |
0 |
database.db.connect() |
| 212 |
0 |
try: |
| 213 |
0 |
with database.db.atomic(): |
| 214 |
0 |
if vnf_member_index is None: |
| 215 |
0 |
alarm_conditions = VnfAlarm.nsr_id == nsr_id |
| 216 |
|
else: |
| 217 |
0 |
query_list = [ |
| 218 |
|
VnfAlarm.nsr_id == nsr_id, |
| 219 |
|
VnfAlarm.vnf_member_index == vnf_member_index, |
| 220 |
|
] |
| 221 |
0 |
alarm_conditions = functools.reduce(operator.and_, query_list) |
| 222 |
0 |
for alarm in VnfAlarmRepository.list(alarm_conditions): |
| 223 |
0 |
log.debug("Deleting vnf alarm %s", alarm.alarm_uuid) |
| 224 |
0 |
try: |
| 225 |
0 |
await self.mon_client.delete_alarm( |
| 226 |
|
alarm.nsr_id, |
| 227 |
|
alarm.vnf_member_index, |
| 228 |
|
alarm.vdu_name, |
| 229 |
|
alarm.alarm_uuid, |
| 230 |
|
) |
| 231 |
0 |
except ValueError: |
| 232 |
0 |
log.exception( |
| 233 |
|
"Error deleting alarm in MON %s", alarm.alarm_uuid |
| 234 |
|
) |
| 235 |
0 |
alarm.delete_instance() |
| 236 |
|
|
| 237 |
0 |
except Exception as e: |
| 238 |
0 |
log.exception("Error deleting vnf alarms:") |
| 239 |
0 |
raise e |
| 240 |
|
finally: |
| 241 |
0 |
database.db.close() |
| 242 |
|
|
| 243 |
1 |
async def handle_alarm(self, alarm_uuid: str, status: str, payload: dict): |
| 244 |
1 |
alert_timeout = int(self.conf.get("alert", "timeout")) |
| 245 |
1 |
database.db.connect() |
| 246 |
1 |
try: |
| 247 |
1 |
with database.db.atomic(): |
| 248 |
1 |
alarm = VnfAlarmRepository.get(VnfAlarm.alarm_uuid == alarm_uuid) |
| 249 |
1 |
log.debug( |
| 250 |
|
"Handling vnf alarm %s with status %s", alarm.alarm_id, status |
| 251 |
|
) |
| 252 |
1 |
for action in alarm.actions: |
| 253 |
|
""" |
| 254 |
|
Compares the current status with the last_action status. |
| 255 |
|
If both the status are 'alarm', it avoid sending repetitive alarm notification. |
| 256 |
|
If both the status are 'ok', it avoid sending repetitive ok notification. |
| 257 |
|
""" |
| 258 |
1 |
if action.type == status: |
| 259 |
1 |
if bool(self.conf.get("alert", "enhanced_alarms")): |
| 260 |
1 |
if ( |
| 261 |
|
status != "ok" |
| 262 |
|
or (status == "ok" and alarm.ok_ack is False) |
| 263 |
|
) and ( |
| 264 |
|
status != "alarm" |
| 265 |
|
or (status == "alarm" and alarm.alarm_ack is False) |
| 266 |
|
): |
| 267 |
1 |
log.info( |
| 268 |
|
"Executing request to url %s for vnf alarm %s with status %s", |
| 269 |
|
action.url, |
| 270 |
|
alarm.alarm_id, |
| 271 |
|
status, |
| 272 |
|
) |
| 273 |
1 |
try: |
| 274 |
1 |
if status == "alarm" and alarm.last_action == "ok": |
| 275 |
1 |
alarm.id_suffix += 1 |
| 276 |
1 |
alarm.ok_ack = False |
| 277 |
1 |
if status == "ok" and alarm.last_action == "alarm": |
| 278 |
1 |
alarm.alarm_ack = False |
| 279 |
1 |
alarm.last_action = status |
| 280 |
1 |
alarm.save() |
| 281 |
0 |
except Exception as e: |
| 282 |
0 |
log.exception(e) |
| 283 |
|
|
| 284 |
1 |
payload["notify_details"][ |
| 285 |
|
"alarm_number" |
| 286 |
|
] = alarm.id_suffix |
| 287 |
1 |
headers = {"content-type": "application/json"} |
| 288 |
1 |
try: |
| 289 |
1 |
resp = requests.post( |
| 290 |
|
url=action.url, |
| 291 |
|
data=json.dumps(payload), |
| 292 |
|
headers=headers, |
| 293 |
|
verify=False, |
| 294 |
|
timeout=alert_timeout, |
| 295 |
|
) |
| 296 |
1 |
log.info("Response %s", resp) |
| 297 |
1 |
if resp.status_code == 200: |
| 298 |
0 |
if status == "ok": |
| 299 |
0 |
alarm.ok_ack = True |
| 300 |
0 |
alarm.save() |
| 301 |
0 |
if status == "alarm": |
| 302 |
0 |
alarm.alarm_ack = True |
| 303 |
0 |
alarm.save() |
| 304 |
0 |
if status == "insufficient-data": |
| 305 |
0 |
alarm.alarm_ack = False |
| 306 |
0 |
alarm.ok_ack = False |
| 307 |
0 |
alarm.save() |
| 308 |
0 |
except ConnectionError: |
| 309 |
0 |
log.exception( |
| 310 |
|
"Error connecting to url %s", action.url |
| 311 |
|
) |
| 312 |
0 |
except RequestException as e: |
| 313 |
0 |
log.info( |
| 314 |
|
"Error: RequestException while connecting to url %s", |
| 315 |
|
action.url, |
| 316 |
|
) |
| 317 |
0 |
log.debug("RequestException %s", e) |
| 318 |
|
|
| 319 |
|
else: |
| 320 |
0 |
log.info( |
| 321 |
|
"Executing request to url %s for vnf alarm %s with status %s", |
| 322 |
|
action.url, |
| 323 |
|
alarm.alarm_id, |
| 324 |
|
status, |
| 325 |
|
) |
| 326 |
0 |
try: |
| 327 |
0 |
requests.post( |
| 328 |
|
url=action.url, |
| 329 |
|
json=json.dumps(payload), |
| 330 |
|
timeout=alert_timeout, |
| 331 |
|
) |
| 332 |
0 |
except ConnectionError: |
| 333 |
0 |
log.exception("Error connecting to url %s", action.url) |
| 334 |
0 |
except RequestException as e: |
| 335 |
0 |
log.info( |
| 336 |
|
"Error: RequestException while connecting to url %s", |
| 337 |
|
action.url, |
| 338 |
|
) |
| 339 |
0 |
log.debug("RequestException %s", e) |
| 340 |
|
|
| 341 |
0 |
except VnfAlarm.DoesNotExist: |
| 342 |
0 |
log.debug( |
| 343 |
|
"There is no alarming action configured for alarm %s.", alarm_uuid |
| 344 |
|
) |
| 345 |
|
finally: |
| 346 |
1 |
database.db.close() |
| 347 |
|
|
| 348 |
1 |
def _get_metric_name(self, vnf_monitoring_param: dict): |
| 349 |
1 |
if "performance-metric" in vnf_monitoring_param: |
| 350 |
1 |
return vnf_monitoring_param["performance-metric"] |
| 351 |
0 |
raise ValueError( |
| 352 |
|
"No metric name found for vnf_monitoring_param %s" |
| 353 |
|
% vnf_monitoring_param["id"] |
| 354 |
|
) |
![[Jenkins]](/jenkins/static/c06f6c43/images/jenkins-header-logo-v2.svg){kind=logo}
