| 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 logging |
| 26 |
1 |
import datetime |
| 27 |
|
|
| 28 |
1 |
from osm_policy_module.common.common_db_client import CommonDbClient |
| 29 |
1 |
from osm_policy_module.common.lcm_client import LcmClient |
| 30 |
1 |
from osm_policy_module.common.mon_client import MonClient |
| 31 |
1 |
from osm_policy_module.core import database |
| 32 |
1 |
from osm_policy_module.core.config import Config |
| 33 |
1 |
from osm_policy_module.core.database import ( |
| 34 |
|
HealingAction, |
| 35 |
|
HealingActionRepository, |
| 36 |
|
) |
| 37 |
1 |
from osm_policy_module.core.exceptions import VdurNotFound |
| 38 |
|
|
| 39 |
1 |
log = logging.getLogger(__name__) |
| 40 |
|
|
| 41 |
|
|
| 42 |
1 |
class HealingService: |
| 43 |
1 |
def __init__(self, config: Config): |
| 44 |
|
""" |
| 45 |
|
Initializing the HealingService |
| 46 |
|
""" |
| 47 |
1 |
log.info("HealingService Initialized") |
| 48 |
1 |
self.conf = config |
| 49 |
1 |
self.db_client = CommonDbClient(config) |
| 50 |
1 |
self.mon_client = MonClient(config) |
| 51 |
1 |
self.lcm_client = LcmClient(config) |
| 52 |
1 |
log.info("Constructor created for HealingService") |
| 53 |
|
|
| 54 |
1 |
async def configure_healing_alarms(self, nsr_id: str): |
| 55 |
|
""" |
| 56 |
|
Configuring the Healing alarms |
| 57 |
|
:param nsr_id: Network service record id |
| 58 |
|
""" |
| 59 |
1 |
log.info("Configuring Healing alarm for NS %s", nsr_id) |
| 60 |
1 |
alarms_created = [] |
| 61 |
1 |
database.db.connect() |
| 62 |
1 |
try: |
| 63 |
1 |
with database.db.atomic(): |
| 64 |
1 |
vnfrs = self.db_client.get_vnfrs(nsr_id) |
| 65 |
1 |
for vnfr in vnfrs: |
| 66 |
1 |
vnfd = self.db_client.get_vnfd(vnfr["vnfd-id"]) |
| 67 |
1 |
df = vnfd.get("df", [{}])[0] |
| 68 |
1 |
if "healing-aspect" not in df: |
| 69 |
0 |
log.info("No healing configuration present in vnfd") |
| 70 |
0 |
continue |
| 71 |
1 |
healing_aspects = df["healing-aspect"] |
| 72 |
1 |
for healing_aspect in healing_aspects: |
| 73 |
1 |
for healing_policy in healing_aspect.get("healing-policy", ()): |
| 74 |
1 |
vdu_id = healing_policy["vdu-id"] |
| 75 |
1 |
for vdur in vnfr["vdur"]: |
| 76 |
1 |
if vdu_id == vdur["vdu-id-ref"]: |
| 77 |
1 |
try: |
| 78 |
1 |
HealingActionRepository.get( |
| 79 |
|
HealingAction.alarm_id |
| 80 |
|
== healing_policy["event-name"], |
| 81 |
|
HealingAction.vdur_name == vdur["name"], |
| 82 |
|
HealingAction.nsr_id == nsr_id, |
| 83 |
|
HealingAction.cooldown_time |
| 84 |
|
== healing_policy["cooldown-time"], |
| 85 |
|
HealingAction.recovery_action |
| 86 |
|
== healing_policy["action-on-recovery"], |
| 87 |
|
HealingAction.vnfinstance_id == vnfr["id"], |
| 88 |
|
HealingAction.vdu_id |
| 89 |
|
== healing_policy["vdu-id"], |
| 90 |
|
HealingAction.count_index |
| 91 |
|
== vdur["count-index"], |
| 92 |
|
) |
| 93 |
0 |
log.debug( |
| 94 |
|
"vdu %s already has an alarm configured with same id %s", |
| 95 |
|
healing_policy["vdu-id"], |
| 96 |
|
healing_policy["event-name"], |
| 97 |
|
) |
| 98 |
0 |
continue |
| 99 |
1 |
except HealingAction.DoesNotExist: |
| 100 |
1 |
pass |
| 101 |
|
|
| 102 |
1 |
metric_name = "vm_status" |
| 103 |
1 |
alarm_uuid = await self.mon_client.create_alarm( |
| 104 |
|
metric_name=metric_name, |
| 105 |
|
ns_id=nsr_id, |
| 106 |
|
vdu_name=vdur["name"], |
| 107 |
|
vnf_member_index=vnfr["member-vnf-index-ref"], |
| 108 |
|
threshold=1, |
| 109 |
|
operation="LT", |
| 110 |
|
statistic="AVERAGE", |
| 111 |
|
) |
| 112 |
1 |
alarm = HealingActionRepository.create( |
| 113 |
|
alarm_id=healing_policy["event-name"], |
| 114 |
|
alarm_uuid=alarm_uuid, |
| 115 |
|
nsr_id=nsr_id, |
| 116 |
|
vnf_member_index=vnfr["member-vnf-index-ref"], |
| 117 |
|
vdur_name=vdur["name"], |
| 118 |
|
recovery_action=healing_policy[ |
| 119 |
|
"action-on-recovery" |
| 120 |
|
], |
| 121 |
|
cooldown_time=healing_policy["cooldown-time"], |
| 122 |
|
day1=healing_policy["day1"], |
| 123 |
|
vdu_id=healing_policy["vdu-id"], |
| 124 |
|
vnfinstance_id=vnfr["id"], |
| 125 |
|
count_index=vdur["count-index"], |
| 126 |
|
) |
| 127 |
1 |
alarms_created.append(alarm) |
| 128 |
|
|
| 129 |
0 |
except Exception as e: |
| 130 |
0 |
log.exception("Error configuring VNF alarms:") |
| 131 |
0 |
if len(alarms_created) > 0: |
| 132 |
0 |
for alarm in alarms_created: |
| 133 |
0 |
try: |
| 134 |
0 |
await self.mon_client.delete_alarm( |
| 135 |
|
alarm.nsr_id, |
| 136 |
|
alarm.vnf_member_index, |
| 137 |
|
alarm.vdu_name, |
| 138 |
|
alarm.alarm_uuid, |
| 139 |
|
) |
| 140 |
0 |
except ValueError: |
| 141 |
0 |
log.exception( |
| 142 |
|
"Error deleting alarm in MON %s", alarm.alarm_uuid |
| 143 |
|
) |
| 144 |
0 |
raise e |
| 145 |
|
finally: |
| 146 |
1 |
database.db.close() |
| 147 |
|
|
| 148 |
1 |
async def delete_orphaned_healing_alarms(self, nsr_id): |
| 149 |
0 |
log.info("Deleting orphaned healing alarms for network service %s", nsr_id) |
| 150 |
0 |
database.db.connect() |
| 151 |
0 |
try: |
| 152 |
0 |
with database.db.atomic(): |
| 153 |
0 |
for alarm in HealingActionRepository.list( |
| 154 |
|
HealingAction.nsr_id == nsr_id |
| 155 |
|
): |
| 156 |
0 |
try: |
| 157 |
0 |
self.db_client.get_vdur( |
| 158 |
|
nsr_id, alarm.vnf_member_index, alarm.vdur_name |
| 159 |
|
) |
| 160 |
0 |
except VdurNotFound: |
| 161 |
0 |
log.info("Deleting orphaned alarm %s", alarm.alarm_uuid) |
| 162 |
0 |
try: |
| 163 |
0 |
await self.mon_client.delete_alarm( |
| 164 |
|
alarm.nsr_id, |
| 165 |
|
alarm.vnf_member_index, |
| 166 |
|
alarm.vdur_name, |
| 167 |
|
alarm.alarm_uuid, |
| 168 |
|
) |
| 169 |
0 |
except ValueError: |
| 170 |
0 |
log.exception( |
| 171 |
|
"Error deleting alarm in MON %s", alarm.alarm_uuid |
| 172 |
|
) |
| 173 |
0 |
alarm.delete_instance() |
| 174 |
|
|
| 175 |
0 |
except Exception as e: |
| 176 |
0 |
log.exception("Error deleting orphaned alarms:") |
| 177 |
0 |
raise e |
| 178 |
|
finally: |
| 179 |
0 |
database.db.close() |
| 180 |
|
|
| 181 |
1 |
async def delete_healing_alarms(self, nsr_id): |
| 182 |
|
""" |
| 183 |
|
Deleting the healing alarms |
| 184 |
|
:param nsr_id: Network service record id |
| 185 |
|
""" |
| 186 |
0 |
log.info("Deleting healing vnf alarms for network service %s", nsr_id) |
| 187 |
0 |
database.db.connect() |
| 188 |
0 |
try: |
| 189 |
0 |
with database.db.atomic(): |
| 190 |
0 |
for alarm in HealingActionRepository.list( |
| 191 |
|
HealingAction.nsr_id == nsr_id |
| 192 |
|
): |
| 193 |
0 |
try: |
| 194 |
0 |
await self.mon_client.delete_alarm( |
| 195 |
|
alarm.nsr_id, |
| 196 |
|
alarm.vnf_member_index, |
| 197 |
|
alarm.vdur_name, |
| 198 |
|
alarm.alarm_uuid, |
| 199 |
|
) |
| 200 |
0 |
except ValueError: |
| 201 |
0 |
log.exception( |
| 202 |
|
"Error deleting alarm in MON %s", alarm.alarm_uuid |
| 203 |
|
) |
| 204 |
0 |
alarm.delete_instance() |
| 205 |
|
|
| 206 |
0 |
except Exception as e: |
| 207 |
0 |
log.exception("Error deleting vnf alarms:") |
| 208 |
0 |
raise e |
| 209 |
|
finally: |
| 210 |
0 |
database.db.close() |
| 211 |
|
|
| 212 |
1 |
async def update_alarm_status(self, alarm_uuid: str, status: str): |
| 213 |
|
""" |
| 214 |
|
For updating the alarm status |
| 215 |
|
:param alarm_uuid: vdu uuid |
| 216 |
|
:param status: Status of an alarm |
| 217 |
|
""" |
| 218 |
0 |
database.db.connect() |
| 219 |
0 |
try: |
| 220 |
0 |
with database.db.atomic(): |
| 221 |
0 |
alarm = HealingActionRepository.get( |
| 222 |
|
HealingAction.alarm_uuid == alarm_uuid |
| 223 |
|
) |
| 224 |
0 |
alarm.last_status = status |
| 225 |
0 |
alarm.save() |
| 226 |
0 |
except HealingAction.DoesNotExist: |
| 227 |
0 |
log.debug("There is no healing action configured for alarm %s.", alarm_uuid) |
| 228 |
|
finally: |
| 229 |
0 |
database.db.close() |
| 230 |
|
|
| 231 |
1 |
async def handle_alarm(self, alarm_uuid: str, status: str): |
| 232 |
|
""" |
| 233 |
|
For Handling the healing alarms |
| 234 |
|
:param alarm_uuid: vdu uuid |
| 235 |
|
:param status: Status of an alarm |
| 236 |
|
""" |
| 237 |
0 |
await self.update_alarm_status(alarm_uuid, status) |
| 238 |
0 |
database.db.connect() |
| 239 |
0 |
try: |
| 240 |
0 |
if status == "alarm": |
| 241 |
0 |
with database.db.atomic(): |
| 242 |
0 |
alarm = HealingActionRepository.get( |
| 243 |
|
HealingAction.alarm_uuid == alarm_uuid |
| 244 |
|
) |
| 245 |
0 |
vnf_member_index = alarm.vnf_member_index |
| 246 |
0 |
vdur_name = alarm.vdur_name |
| 247 |
0 |
vdu_id = alarm.vdu_id |
| 248 |
0 |
nsr_id = alarm.nsr_id |
| 249 |
0 |
heal_type = alarm.recovery_action |
| 250 |
0 |
cooldown_time = alarm.cooldown_time |
| 251 |
0 |
count_index = alarm.count_index |
| 252 |
0 |
last_heal = alarm.last_heal |
| 253 |
0 |
day1 = alarm.day1 |
| 254 |
0 |
vnfinstance_id = alarm.vnfinstance_id |
| 255 |
0 |
alarms = HealingActionRepository.list( |
| 256 |
|
HealingAction.vnf_member_index == vnf_member_index, |
| 257 |
|
HealingAction.vdur_name == vdur_name, |
| 258 |
|
) |
| 259 |
0 |
statuses = [] |
| 260 |
|
|
| 261 |
0 |
for alarm in alarms: |
| 262 |
0 |
statuses.append(alarm.last_status) |
| 263 |
0 |
if (set(statuses) == {"alarm"}) or ("alarm" in statuses): |
| 264 |
0 |
delta = datetime.datetime.now() - last_heal |
| 265 |
0 |
if delta.total_seconds() > cooldown_time: |
| 266 |
0 |
await self.lcm_client.heal( |
| 267 |
|
nsr_id, |
| 268 |
|
vnfinstance_id, |
| 269 |
|
vdur_name, |
| 270 |
|
vdu_id, |
| 271 |
|
vnf_member_index, |
| 272 |
|
heal_type, |
| 273 |
|
day1, |
| 274 |
|
count_index, |
| 275 |
|
) |
| 276 |
0 |
last_heal = datetime.datetime.now() |
| 277 |
0 |
log.info("datetime.datetime.now %s", datetime.datetime.now) |
| 278 |
0 |
alarm.last_heal = last_heal |
| 279 |
0 |
alarm.save() |
| 280 |
|
|
| 281 |
0 |
except HealingAction.DoesNotExist: |
| 282 |
0 |
log.info("There is no healing action configured for alarm %s.", alarm_uuid) |
| 283 |
|
finally: |
| 284 |
0 |
database.db.close() |
![[Jenkins]](/jenkins/static/c06f6c43/images/jenkins-header-logo-v2.svg){kind=logo}
