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 |
|
vnfr=vnfr, |
132 |
|
vnfd=vnfd, |
133 |
|
) |
134 |
1 |
alarm = VnfAlarmRepository.create( |
135 |
|
alarm_id=alarm_descriptor["alarm-id"], |
136 |
|
alarm_uuid=alarm_uuid, |
137 |
|
nsr_id=nsr_id, |
138 |
|
vnf_member_index=vnfr["member-vnf-index-ref"], |
139 |
|
vdu_name=vdur["name"], |
140 |
|
last_action="insufficient-data", |
141 |
|
id_suffix=0, |
142 |
|
ok_ack=False, |
143 |
|
alarm_ack=False, |
144 |
|
) |
145 |
1 |
for action_type in ["ok", "insufficient-data", "alarm"]: |
146 |
1 |
if ( |
147 |
|
"actions" in alarm_descriptor |
148 |
|
and action_type in alarm_descriptor["actions"] |
149 |
|
): |
150 |
1 |
for url in alarm_descriptor["actions"][ |
151 |
|
action_type |
152 |
|
]: |
153 |
1 |
AlarmActionRepository.create( |
154 |
|
type=action_type, |
155 |
|
url=url["url"], |
156 |
|
alarm=alarm, |
157 |
|
) |
158 |
1 |
alarms_created.append(alarm) |
159 |
|
|
160 |
0 |
except Exception as e: |
161 |
0 |
log.exception("Error configuring VNF alarms:") |
162 |
0 |
if len(alarms_created) > 0: |
163 |
0 |
log.debug("Cleaning alarm resources in MON") |
164 |
0 |
for alarm in alarms_created: |
165 |
0 |
try: |
166 |
0 |
await self.mon_client.delete_alarm( |
167 |
|
alarm.nsr_id, |
168 |
|
alarm.vnf_member_index, |
169 |
|
alarm.vdu_name, |
170 |
|
alarm.alarm_uuid, |
171 |
|
) |
172 |
0 |
except ValueError: |
173 |
0 |
log.exception( |
174 |
|
"Error deleting alarm in MON %s", alarm.alarm_uuid |
175 |
|
) |
176 |
0 |
raise e |
177 |
|
finally: |
178 |
1 |
database.db.close() |
179 |
|
|
180 |
1 |
async def delete_orphaned_alarms(self, nsr_id): |
181 |
|
# TODO: Review as it seems this code is never called |
182 |
0 |
log.info("Deleting orphaned vnf alarms for network service %s", nsr_id) |
183 |
0 |
database.db.connect() |
184 |
0 |
try: |
185 |
0 |
with database.db.atomic(): |
186 |
0 |
for alarm in VnfAlarmRepository.list(VnfAlarm.nsr_id == nsr_id): |
187 |
0 |
try: |
188 |
0 |
self.db_client.get_vdur( |
189 |
|
nsr_id, alarm.vnf_member_index, alarm.vdu_name |
190 |
|
) |
191 |
0 |
except VdurNotFound: |
192 |
0 |
log.debug("Deleting orphaned alarm %s", alarm.alarm_uuid) |
193 |
0 |
try: |
194 |
0 |
await self.mon_client.delete_alarm( |
195 |
|
alarm.nsr_id, |
196 |
|
alarm.vnf_member_index, |
197 |
|
alarm.vdu_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 |
0 |
except Exception as e: |
206 |
0 |
log.exception("Error deleting orphaned alarms:") |
207 |
0 |
raise e |
208 |
|
finally: |
209 |
0 |
database.db.close() |
210 |
|
|
211 |
1 |
async def delete_vnf_alarms(self, nsr_id, vnf_member_index=None): |
212 |
0 |
log.info("Deleting vnf alarms for network service %s", nsr_id) |
213 |
0 |
database.db.connect() |
214 |
0 |
try: |
215 |
0 |
with database.db.atomic(): |
216 |
0 |
if vnf_member_index is None: |
217 |
0 |
alarm_conditions = VnfAlarm.nsr_id == nsr_id |
218 |
|
else: |
219 |
0 |
query_list = [ |
220 |
|
VnfAlarm.nsr_id == nsr_id, |
221 |
|
VnfAlarm.vnf_member_index == vnf_member_index, |
222 |
|
] |
223 |
0 |
alarm_conditions = functools.reduce(operator.and_, query_list) |
224 |
0 |
for alarm in VnfAlarmRepository.list(alarm_conditions): |
225 |
0 |
log.debug("Deleting vnf alarm %s", alarm.alarm_uuid) |
226 |
0 |
try: |
227 |
0 |
await self.mon_client.delete_alarm( |
228 |
|
alarm.nsr_id, |
229 |
|
alarm.vnf_member_index, |
230 |
|
alarm.vdu_name, |
231 |
|
alarm.alarm_uuid, |
232 |
|
) |
233 |
0 |
except ValueError: |
234 |
0 |
log.exception( |
235 |
|
"Error deleting alarm in MON %s", alarm.alarm_uuid |
236 |
|
) |
237 |
0 |
alarm.delete_instance() |
238 |
|
|
239 |
0 |
except Exception as e: |
240 |
0 |
log.exception("Error deleting vnf alarms:") |
241 |
0 |
raise e |
242 |
|
finally: |
243 |
0 |
database.db.close() |
244 |
|
|
245 |
1 |
async def handle_alarm(self, alarm_uuid: str, status: str, payload: dict): |
246 |
1 |
alert_timeout = int(self.conf.get("alert", "timeout")) |
247 |
1 |
database.db.connect() |
248 |
1 |
try: |
249 |
1 |
with database.db.atomic(): |
250 |
1 |
alarm = VnfAlarmRepository.get(VnfAlarm.alarm_uuid == alarm_uuid) |
251 |
1 |
log.debug( |
252 |
|
"Handling vnf alarm %s with status %s", alarm.alarm_id, status |
253 |
|
) |
254 |
1 |
for action in alarm.actions: |
255 |
|
""" |
256 |
|
Compares the current status with the last_action status. |
257 |
|
If both the status are 'alarm', it avoid sending repetitive alarm notification. |
258 |
|
If both the status are 'ok', it avoid sending repetitive ok notification. |
259 |
|
""" |
260 |
1 |
if action.type == status: |
261 |
1 |
if bool(self.conf.get("alert", "enhanced_alarms")): |
262 |
1 |
if ( |
263 |
|
status != "ok" |
264 |
|
or (status == "ok" and alarm.ok_ack is False) |
265 |
|
) and ( |
266 |
|
status != "alarm" |
267 |
|
or (status == "alarm" and alarm.alarm_ack is False) |
268 |
|
): |
269 |
1 |
log.info( |
270 |
|
"Executing request to url %s for vnf alarm %s with status %s", |
271 |
|
action.url, |
272 |
|
alarm.alarm_id, |
273 |
|
status, |
274 |
|
) |
275 |
1 |
try: |
276 |
1 |
if ( |
277 |
|
status == "alarm" |
278 |
|
and alarm.last_action == "ok" |
279 |
|
or ( |
280 |
|
status == "alarm" |
281 |
|
and alarm.last_action == "insufficient-data" |
282 |
|
) |
283 |
|
): |
284 |
1 |
alarm.id_suffix += 1 |
285 |
1 |
alarm.ok_ack = False |
286 |
1 |
if status == "ok" and alarm.last_action == "alarm": |
287 |
1 |
alarm.alarm_ack = False |
288 |
1 |
alarm.last_action = status |
289 |
1 |
alarm.save() |
290 |
0 |
except Exception as e: |
291 |
0 |
log.exception(e) |
292 |
|
|
293 |
1 |
payload["notify_details"][ |
294 |
|
"alarm_number" |
295 |
|
] = alarm.id_suffix |
296 |
1 |
headers = {"content-type": "application/json"} |
297 |
1 |
try: |
298 |
1 |
resp = requests.post( |
299 |
|
url=action.url, |
300 |
|
data=json.dumps(payload), |
301 |
|
headers=headers, |
302 |
|
timeout=alert_timeout, |
303 |
|
) |
304 |
1 |
log.info("Response %s", resp) |
305 |
1 |
if resp.status_code == 200: |
306 |
0 |
if status == "ok": |
307 |
0 |
alarm.ok_ack = True |
308 |
0 |
alarm.save() |
309 |
0 |
if status == "alarm": |
310 |
0 |
alarm.alarm_ack = True |
311 |
0 |
alarm.save() |
312 |
0 |
if status == "insufficient-data": |
313 |
0 |
alarm.alarm_ack = False |
314 |
0 |
alarm.ok_ack = False |
315 |
0 |
alarm.save() |
316 |
0 |
except ConnectionError: |
317 |
0 |
log.exception( |
318 |
|
"Error connecting to url %s", action.url |
319 |
|
) |
320 |
0 |
except RequestException as e: |
321 |
0 |
log.info( |
322 |
|
"Error: RequestException while connecting to url %s", |
323 |
|
action.url, |
324 |
|
) |
325 |
0 |
log.debug("RequestException %s", e) |
326 |
|
|
327 |
|
else: |
328 |
0 |
log.info( |
329 |
|
"Executing request to url %s for vnf alarm %s with status %s", |
330 |
|
action.url, |
331 |
|
alarm.alarm_id, |
332 |
|
status, |
333 |
|
) |
334 |
0 |
try: |
335 |
0 |
requests.post( |
336 |
|
url=action.url, |
337 |
|
json=json.dumps(payload), |
338 |
|
timeout=alert_timeout, |
339 |
|
) |
340 |
0 |
except ConnectionError: |
341 |
0 |
log.exception("Error connecting to url %s", action.url) |
342 |
0 |
except RequestException as e: |
343 |
0 |
log.info( |
344 |
|
"Error: RequestException while connecting to url %s", |
345 |
|
action.url, |
346 |
|
) |
347 |
0 |
log.debug("RequestException %s", e) |
348 |
|
|
349 |
0 |
except VnfAlarm.DoesNotExist: |
350 |
0 |
log.debug( |
351 |
|
"There is no alarming action configured for alarm %s.", alarm_uuid |
352 |
|
) |
353 |
|
finally: |
354 |
1 |
database.db.close() |
355 |
|
|
356 |
1 |
def _get_metric_name(self, vnf_monitoring_param: dict): |
357 |
1 |
if "performance-metric" in vnf_monitoring_param: |
358 |
1 |
return vnf_monitoring_param["performance-metric"] |
359 |
0 |
raise ValueError( |
360 |
|
"No metric name found for vnf_monitoring_param %s" |
361 |
|
% vnf_monitoring_param["id"] |
362 |
|
) |