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 asyncio |
26 |
1 |
import json |
27 |
1 |
import logging |
28 |
1 |
import operator |
29 |
1 |
import functools |
30 |
|
|
31 |
1 |
import requests |
32 |
1 |
from requests.exceptions import ConnectionError, RequestException |
33 |
|
|
34 |
1 |
from osm_policy_module.common.common_db_client import CommonDbClient |
35 |
1 |
from osm_policy_module.common.lcm_client import LcmClient |
36 |
1 |
from osm_policy_module.common.mon_client import MonClient |
37 |
1 |
from osm_policy_module.core import database |
38 |
1 |
from osm_policy_module.core.config import Config |
39 |
1 |
from osm_policy_module.core.database import ( |
40 |
|
VnfAlarm, |
41 |
|
VnfAlarmRepository, |
42 |
|
AlarmActionRepository, |
43 |
|
) |
44 |
1 |
from osm_policy_module.core.exceptions import VdurNotFound |
45 |
|
|
46 |
1 |
log = logging.getLogger(__name__) |
47 |
|
|
48 |
|
|
49 |
1 |
class AlarmingService: |
50 |
1 |
def __init__(self, config: Config, loop=None): |
51 |
1 |
self.conf = config |
52 |
1 |
if not loop: |
53 |
1 |
loop = asyncio.get_event_loop() |
54 |
1 |
self.loop = loop |
55 |
1 |
self.db_client = CommonDbClient(config) |
56 |
1 |
self.mon_client = MonClient(config, loop=self.loop) |
57 |
1 |
self.lcm_client = LcmClient(config, loop=self.loop) |
58 |
|
|
59 |
1 |
async def configure_vnf_alarms(self, nsr_id: str, vnf_member_index=None): |
60 |
1 |
log.info("Configuring vnf alarms for network service %s", nsr_id) |
61 |
1 |
alarms_created = [] |
62 |
1 |
database.db.connect() |
63 |
1 |
try: |
64 |
1 |
with database.db.atomic(): |
65 |
1 |
if vnf_member_index is None: |
66 |
1 |
vnfrs = self.db_client.get_vnfrs(nsr_id) |
67 |
|
else: |
68 |
0 |
vnfrs = [] |
69 |
0 |
vnfr = self.db_client.get_vnfr(nsr_id, vnf_member_index) |
70 |
0 |
vnfrs.append(vnfr) |
71 |
|
# vnfrs = self.db_client.get_vnfrs(nsr_id) |
72 |
1 |
for vnfr in vnfrs: |
73 |
1 |
log.debug("Processing vnfr: %s", vnfr) |
74 |
1 |
vnfd = self.db_client.get_vnfd(vnfr["vnfd-id"]) |
75 |
1 |
for vdur in vnfr["vdur"]: |
76 |
1 |
vdu = next( |
77 |
|
filter( |
78 |
|
lambda vdu: vdu["id"] == vdur["vdu-id-ref"], vnfd["vdu"] |
79 |
|
) |
80 |
|
) |
81 |
1 |
if "alarm" in vdu: |
82 |
1 |
alarm_descriptors = vdu["alarm"] |
83 |
1 |
for alarm_descriptor in alarm_descriptors: |
84 |
1 |
try: |
85 |
1 |
VnfAlarmRepository.get( |
86 |
|
VnfAlarm.alarm_id |
87 |
|
== alarm_descriptor["alarm-id"], |
88 |
|
VnfAlarm.vnf_member_index |
89 |
|
== vnfr["member-vnf-index-ref"], |
90 |
|
VnfAlarm.vdu_name == vdur["name"], |
91 |
|
VnfAlarm.nsr_id == nsr_id, |
92 |
|
) |
93 |
0 |
log.debug( |
94 |
|
"vdu %s already has an alarm configured with same id %s", |
95 |
|
vdur["name"], |
96 |
|
alarm_descriptor["alarm-id"], |
97 |
|
) |
98 |
0 |
continue |
99 |
1 |
except VnfAlarm.DoesNotExist: |
100 |
1 |
pass |
101 |
1 |
vnf_monitoring_param = next( |
102 |
|
filter( |
103 |
|
lambda param: param["id"] |
104 |
|
== alarm_descriptor["vnf-monitoring-param-ref"], |
105 |
|
vdu.get("monitoring-parameter", []), |
106 |
|
), |
107 |
|
{}, |
108 |
|
) |
109 |
1 |
metric_name = self._get_metric_name( |
110 |
|
vnf_monitoring_param |
111 |
|
) |
112 |
1 |
alarm_action = dict() |
113 |
1 |
for action_type in ["ok", "insufficient-data", "alarm"]: |
114 |
1 |
if ( |
115 |
|
"actions" in alarm_descriptor |
116 |
|
and action_type in alarm_descriptor["actions"] |
117 |
|
): |
118 |
1 |
for url in alarm_descriptor["actions"][ |
119 |
|
action_type |
120 |
|
]: |
121 |
1 |
if "webhook" in alarm_action: |
122 |
1 |
alarm_action["webhook"].append( |
123 |
|
url["url"] |
124 |
|
) |
125 |
|
else: |
126 |
1 |
alarm_action["webhook"] = [url["url"]] |
127 |
1 |
alarm_uuid = await self.mon_client.create_alarm( |
128 |
|
metric_name=metric_name, |
129 |
|
ns_id=nsr_id, |
130 |
|
vdu_name=vdur["name"], |
131 |
|
vnf_member_index=vnfr["member-vnf-index-ref"], |
132 |
|
threshold=alarm_descriptor["value"], |
133 |
|
operation=alarm_descriptor["operation"], |
134 |
|
action=str(alarm_action), |
135 |
|
) |
136 |
1 |
alarm = VnfAlarmRepository.create( |
137 |
|
alarm_id=alarm_descriptor["alarm-id"], |
138 |
|
alarm_uuid=alarm_uuid, |
139 |
|
nsr_id=nsr_id, |
140 |
|
vnf_member_index=vnfr["member-vnf-index-ref"], |
141 |
|
vdu_name=vdur["name"], |
142 |
|
last_action="insufficient-data", |
143 |
|
id_suffix=0, |
144 |
|
ok_ack=False, |
145 |
|
alarm_ack=False, |
146 |
|
) |
147 |
1 |
for action_type in ["ok", "insufficient-data", "alarm"]: |
148 |
1 |
if ( |
149 |
|
"actions" in alarm_descriptor |
150 |
|
and action_type in alarm_descriptor["actions"] |
151 |
|
): |
152 |
1 |
for url in alarm_descriptor["actions"][ |
153 |
|
action_type |
154 |
|
]: |
155 |
1 |
AlarmActionRepository.create( |
156 |
|
type=action_type, |
157 |
|
url=url["url"], |
158 |
|
alarm=alarm, |
159 |
|
) |
160 |
1 |
alarms_created.append(alarm) |
161 |
|
|
162 |
0 |
except Exception as e: |
163 |
0 |
log.exception("Error configuring VNF alarms:") |
164 |
0 |
if len(alarms_created) > 0: |
165 |
0 |
log.debug("Cleaning alarm resources in MON") |
166 |
0 |
for alarm in alarms_created: |
167 |
0 |
try: |
168 |
0 |
await self.mon_client.delete_alarm( |
169 |
|
alarm.nsr_id, |
170 |
|
alarm.vnf_member_index, |
171 |
|
alarm.vdu_name, |
172 |
|
alarm.alarm_uuid, |
173 |
|
) |
174 |
0 |
except ValueError: |
175 |
0 |
log.exception( |
176 |
|
"Error deleting alarm in MON %s", alarm.alarm_uuid |
177 |
|
) |
178 |
0 |
raise e |
179 |
|
finally: |
180 |
1 |
database.db.close() |
181 |
|
|
182 |
1 |
async def delete_orphaned_alarms(self, nsr_id): |
183 |
|
# TODO: Review as it seems this code is never called |
184 |
0 |
log.info("Deleting orphaned vnf alarms for network service %s", nsr_id) |
185 |
0 |
database.db.connect() |
186 |
0 |
try: |
187 |
0 |
with database.db.atomic(): |
188 |
0 |
for alarm in VnfAlarmRepository.list(VnfAlarm.nsr_id == nsr_id): |
189 |
0 |
try: |
190 |
0 |
self.db_client.get_vdur( |
191 |
|
nsr_id, alarm.vnf_member_index, alarm.vdu_name |
192 |
|
) |
193 |
0 |
except VdurNotFound: |
194 |
0 |
log.debug("Deleting orphaned alarm %s", alarm.alarm_uuid) |
195 |
0 |
try: |
196 |
0 |
await self.mon_client.delete_alarm( |
197 |
|
alarm.nsr_id, |
198 |
|
alarm.vnf_member_index, |
199 |
|
alarm.vdu_name, |
200 |
|
alarm.alarm_uuid, |
201 |
|
) |
202 |
0 |
except ValueError: |
203 |
0 |
log.exception( |
204 |
|
"Error deleting alarm in MON %s", alarm.alarm_uuid |
205 |
|
) |
206 |
0 |
alarm.delete_instance() |
207 |
0 |
except Exception as e: |
208 |
0 |
log.exception("Error deleting orphaned alarms:") |
209 |
0 |
raise e |
210 |
|
finally: |
211 |
0 |
database.db.close() |
212 |
|
|
213 |
1 |
async def delete_vnf_alarms(self, nsr_id, vnf_member_index=None): |
214 |
0 |
log.info("Deleting vnf alarms for network service %s", nsr_id) |
215 |
0 |
database.db.connect() |
216 |
0 |
try: |
217 |
0 |
with database.db.atomic(): |
218 |
0 |
if vnf_member_index is None: |
219 |
0 |
alarm_conditions = VnfAlarm.nsr_id == nsr_id |
220 |
|
else: |
221 |
0 |
query_list = [ |
222 |
|
VnfAlarm.nsr_id == nsr_id, |
223 |
|
VnfAlarm.vnf_member_index == vnf_member_index, |
224 |
|
] |
225 |
0 |
alarm_conditions = functools.reduce(operator.and_, query_list) |
226 |
0 |
for alarm in VnfAlarmRepository.list(alarm_conditions): |
227 |
0 |
log.debug("Deleting vnf alarm %s", alarm.alarm_uuid) |
228 |
0 |
try: |
229 |
0 |
await self.mon_client.delete_alarm( |
230 |
|
alarm.nsr_id, |
231 |
|
alarm.vnf_member_index, |
232 |
|
alarm.vdu_name, |
233 |
|
alarm.alarm_uuid, |
234 |
|
) |
235 |
0 |
except ValueError: |
236 |
0 |
log.exception( |
237 |
|
"Error deleting alarm in MON %s", alarm.alarm_uuid |
238 |
|
) |
239 |
0 |
alarm.delete_instance() |
240 |
|
|
241 |
0 |
except Exception as e: |
242 |
0 |
log.exception("Error deleting vnf alarms:") |
243 |
0 |
raise e |
244 |
|
finally: |
245 |
0 |
database.db.close() |
246 |
|
|
247 |
1 |
async def handle_alarm(self, alarm_uuid: str, status: str, payload: dict): |
248 |
1 |
alert_timeout = int(self.conf.get("alert", "timeout")) |
249 |
1 |
database.db.connect() |
250 |
1 |
try: |
251 |
1 |
with database.db.atomic(): |
252 |
1 |
alarm = VnfAlarmRepository.get(VnfAlarm.alarm_uuid == alarm_uuid) |
253 |
1 |
log.debug( |
254 |
|
"Handling vnf alarm %s with status %s", alarm.alarm_id, status |
255 |
|
) |
256 |
1 |
for action in alarm.actions: |
257 |
|
""" |
258 |
|
Compares the current status with the last_action status. |
259 |
|
If both the status are 'alarm', it avoid sending repetitive alarm notification. |
260 |
|
If both the status are 'ok', it avoid sending repetitive ok notification. |
261 |
|
""" |
262 |
1 |
if action.type == status: |
263 |
1 |
if bool(self.conf.get("alert", "enhanced_alarms")): |
264 |
1 |
if ( |
265 |
|
status != "ok" |
266 |
|
or (status == "ok" and alarm.ok_ack is False) |
267 |
|
) and ( |
268 |
|
status != "alarm" |
269 |
|
or (status == "alarm" and alarm.alarm_ack is False) |
270 |
|
): |
271 |
1 |
log.info( |
272 |
|
"Executing request to url %s for vnf alarm %s with status %s", |
273 |
|
action.url, |
274 |
|
alarm.alarm_id, |
275 |
|
status, |
276 |
|
) |
277 |
1 |
try: |
278 |
1 |
if status == "alarm" and alarm.last_action == "ok": |
279 |
1 |
alarm.id_suffix += 1 |
280 |
1 |
alarm.ok_ack = False |
281 |
1 |
if status == "ok" and alarm.last_action == "alarm": |
282 |
1 |
alarm.alarm_ack = False |
283 |
1 |
alarm.last_action = status |
284 |
1 |
alarm.save() |
285 |
0 |
except Exception as e: |
286 |
0 |
log.exception(e) |
287 |
|
|
288 |
1 |
payload["notify_details"][ |
289 |
|
"alarm_number" |
290 |
|
] = alarm.id_suffix |
291 |
1 |
headers = {"content-type": "application/json"} |
292 |
1 |
try: |
293 |
1 |
resp = requests.post( |
294 |
|
url=action.url, |
295 |
|
data=json.dumps(payload), |
296 |
|
headers=headers, |
297 |
|
verify=False, |
298 |
|
timeout=alert_timeout, |
299 |
|
) |
300 |
1 |
log.info("Response %s", resp) |
301 |
1 |
if resp.status_code == 200: |
302 |
0 |
if status == "ok": |
303 |
0 |
alarm.ok_ack = True |
304 |
0 |
alarm.save() |
305 |
0 |
if status == "alarm": |
306 |
0 |
alarm.alarm_ack = True |
307 |
0 |
alarm.save() |
308 |
0 |
if status == "insufficient-data": |
309 |
0 |
alarm.alarm_ack = False |
310 |
0 |
alarm.ok_ack = False |
311 |
0 |
alarm.save() |
312 |
0 |
except ConnectionError: |
313 |
0 |
log.exception( |
314 |
|
"Error connecting to url %s", action.url |
315 |
|
) |
316 |
0 |
except RequestException as e: |
317 |
0 |
log.info( |
318 |
|
"Error: RequestException while connecting to url %s", |
319 |
|
action.url, |
320 |
|
) |
321 |
0 |
log.debug("RequestException %s", e) |
322 |
|
|
323 |
|
else: |
324 |
0 |
log.info( |
325 |
|
"Executing request to url %s for vnf alarm %s with status %s", |
326 |
|
action.url, |
327 |
|
alarm.alarm_id, |
328 |
|
status, |
329 |
|
) |
330 |
0 |
try: |
331 |
0 |
requests.post( |
332 |
|
url=action.url, |
333 |
|
json=json.dumps(payload), |
334 |
|
timeout=alert_timeout, |
335 |
|
) |
336 |
0 |
except ConnectionError: |
337 |
0 |
log.exception("Error connecting to url %s", action.url) |
338 |
0 |
except RequestException as e: |
339 |
0 |
log.info( |
340 |
|
"Error: RequestException while connecting to url %s", |
341 |
|
action.url, |
342 |
|
) |
343 |
0 |
log.debug("RequestException %s", e) |
344 |
|
|
345 |
0 |
except VnfAlarm.DoesNotExist: |
346 |
0 |
log.debug( |
347 |
|
"There is no alarming action configured for alarm %s.", alarm_uuid |
348 |
|
) |
349 |
|
finally: |
350 |
1 |
database.db.close() |
351 |
|
|
352 |
1 |
def _get_metric_name(self, vnf_monitoring_param: dict): |
353 |
1 |
if "performance-metric" in vnf_monitoring_param: |
354 |
1 |
return vnf_monitoring_param["performance-metric"] |
355 |
0 |
raise ValueError( |
356 |
|
"No metric name found for vnf_monitoring_param %s" |
357 |
|
% vnf_monitoring_param["id"] |
358 |
|
) |