osm_mon/plugins/OpenStack/Aodh/alarming.py

   1 # Copyright 2017 Intel Research and Development Ireland Limited
   2 # *************************************************************
   3
   4 # This file is part of OSM Monitoring module
   5 # All Rights Reserved to Intel Corporation
   6
   7 # Licensed under the Apache License, Version 2.0 (the "License"); you may
   8 # not use this file except in compliance with the License. You may obtain
   9 # a copy of the License at
  10
  11 #         http://www.apache.org/licenses/LICENSE-2.0
  12
  13 # Unless required by applicable law or agreed to in writing, software
  14 # distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
  15 # WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
  16 # License for the specific language governing permissions and limitations
  17 # under the License.
  18
  19 # For those usages not covered by the Apache License, Version 2.0 please
  20 # contact: helena.mcgough@intel.com or adrian.hoban@intel.com
  21 ##
  22 """Carry out alarming requests via Aodh API."""
  23
  24 import json
  25 import logging
  26
  27 import six
  28 import yaml
  29
  30 from osm_mon.core.auth import AuthManager
  31 from osm_mon.core.database import DatabaseManager
  32 from osm_mon.core.message_bus.producer import KafkaProducer
  33 from osm_mon.core.settings import Config
  34 from osm_mon.plugins.OpenStack.common import Common
  35 from osm_mon.plugins.OpenStack.response import OpenStack_Response
  36
  37 log = logging.getLogger(__name__)
  38
  39 ALARM_NAMES = {
  40     "average_memory_usage_above_threshold": "average_memory_utilization",
  41     "disk_read_ops": "disk_read_ops",
  42     "disk_write_ops": "disk_write_ops",
  43     "disk_read_bytes": "disk_read_bytes",
  44     "disk_write_bytes": "disk_write_bytes",
  45     "net_packets_dropped": "packets_dropped",
  46     "packets_in_above_threshold": "packets_received",
  47     "packets_out_above_threshold": "packets_sent",
  48     "cpu_utilization_above_threshold": "cpu_utilization"}
  49
  50 METRIC_MAPPINGS = {
  51     "average_memory_utilization": "memory.percent",
  52     "disk_read_ops": "disk.read.requests",
  53     "disk_write_ops": "disk.write.requests",
  54     "disk_read_bytes": "disk.read.bytes",
  55     "disk_write_bytes": "disk.write.bytes",
  56     "packets_dropped": "interface.if_dropped",
  57     "packets_received": "interface.if_packets",
  58     "packets_sent": "interface.if_packets",
  59     "cpu_utilization": "cpu_util",
  60 }
  61
  62 SEVERITIES = {
  63     "warning": "low",
  64     "minor": "low",
  65     "major": "moderate",
  66     "critical": "critical",
  67     "indeterminate": "critical"}
  68
  69 STATISTICS = {
  70     "average": "mean",
  71     "minimum": "min",
  72     "maximum": "max",
  73     "count": "count",
  74     "sum": "sum"}
  75
  76
  77 class Alarming(object):
  78     """Carries out alarming requests and responses via Aodh API."""
  79
  80     def __init__(self):
  81         """Create the OpenStack alarming instance."""
  82         # Initialize configuration and notifications
  83         config = Config.instance()
  84         config.read_environ()
  85
  86         self._database_manager = DatabaseManager()
  87         self._auth_manager = AuthManager()
  88
  89         # Use the Response class to generate valid json response messages
  90         self._response = OpenStack_Response()
  91
  92         # Initializer a producer to send responses back to SO
  93         self._producer = KafkaProducer("alarm_response")
  94
  95     def configure_alarm(self, alarm_endpoint, metric_endpoint, auth_token, values, vim_config):
  96         """Create requested alarm in Aodh."""
  97         url = "{}/v2/alarms/".format(alarm_endpoint)
  98
  99         # Check if the desired alarm is supported
 100         alarm_name = values['alarm_name'].lower()
 101         metric_name = values['metric_name'].lower()
 102         resource_id = values['resource_uuid']
 103
 104         if metric_name not in METRIC_MAPPINGS.keys():
 105             log.warning("This metric is not supported.")
 106             return None, False
 107
 108         # Check for the required metric
 109         metric_id = self.check_for_metric(auth_token, metric_endpoint, metric_name, resource_id)
 110
 111         try:
 112             if metric_id is not None:
 113                 # Create the alarm if metric is available
 114                 if 'granularity' in vim_config and 'granularity' not in values:
 115                     values['granularity'] = vim_config['granularity']
 116                 payload = self.check_payload(values, metric_name, resource_id,
 117                                              alarm_name)
 118                 new_alarm = Common.perform_request(
 119                     url, auth_token, req_type="post", payload=payload)
 120                 return json.loads(new_alarm.text)['alarm_id'], True
 121             else:
 122                 log.warning("The required Gnocchi metric does not exist.")
 123                 return None, False
 124
 125         except Exception as exc:
 126             log.warning("Failed to create the alarm: %s", exc)
 127         return None, False
 128
 129     def alarming(self, message):
 130         """Consume info from the message bus to manage alarms."""
 131         try:
 132             values = json.loads(message.value)
 133         except ValueError:
 134             values = yaml.safe_load(message.value)
 135
 136         log.info("OpenStack alarm action required.")
 137         vim_uuid = values['vim_uuid']
 138
 139         auth_token = Common.get_auth_token(vim_uuid)
 140
 141         alarm_endpoint = Common.get_endpoint("alarming", vim_uuid)
 142         metric_endpoint = Common.get_endpoint("metric", vim_uuid)
 143
 144         vim_account = self._auth_manager.get_credentials(vim_uuid)
 145         vim_config = json.loads(vim_account.config)
 146
 147         if message.key == "create_alarm_request":
 148             # Configure/Update an alarm
 149             alarm_details = values['alarm_create_request']
 150
 151             alarm_id, alarm_status = self.configure_alarm(
 152                 alarm_endpoint, metric_endpoint, auth_token, alarm_details, vim_config)
 153
 154             # Generate a valid response message, send via producer
 155             try:
 156                 if alarm_status is True:
 157                     log.info("Alarm successfully created")
 158                     self._database_manager.save_alarm(alarm_id, vim_uuid)
 159
 160                 resp_message = self._response.generate_response(
 161                     'create_alarm_response', status=alarm_status,
 162                     alarm_id=alarm_id,
 163                     cor_id=alarm_details['correlation_id'])
 164                 log.info("Response Message: %s", resp_message)
 165                 self._producer.create_alarm_response(
 166                     'create_alarm_response', resp_message)
 167             except Exception:
 168                 log.exception("Response creation failed:")
 169
 170         elif message.key == "list_alarm_request":
 171             # Check for a specified: alarm_name, resource_uuid, severity
 172             # and generate the appropriate list
 173             list_details = values['alarm_list_request']
 174
 175             alarm_list = self.list_alarms(
 176                 alarm_endpoint, auth_token, list_details)
 177
 178             try:
 179                 # Generate and send a list response back
 180                 resp_message = self._response.generate_response(
 181                     'list_alarm_response', alarm_list=alarm_list,
 182                     cor_id=list_details['correlation_id'])
 183                 log.info("Response Message: %s", resp_message)
 184                 self._producer.list_alarm_response(
 185                     'list_alarm_response', resp_message)
 186             except Exception:
 187                 log.exception("Failed to send a valid response back.")
 188
 189         elif message.key == "delete_alarm_request":
 190             request_details = values['alarm_delete_request']
 191             alarm_id = request_details['alarm_uuid']
 192
 193             resp_status = self.delete_alarm(
 194                 alarm_endpoint, auth_token, alarm_id)
 195
 196             # Generate and send a response message
 197             try:
 198                 resp_message = self._response.generate_response(
 199                     'delete_alarm_response', alarm_id=alarm_id,
 200                     status=resp_status,
 201                     cor_id=request_details['correlation_id'])
 202                 log.info("Response message: %s", resp_message)
 203                 self._producer.delete_alarm_response(
 204                     'delete_alarm_response', resp_message)
 205             except Exception:
 206                 log.exception("Failed to create delete response: ")
 207
 208         elif message.key == "acknowledge_alarm":
 209             # Acknowledge that an alarm has been dealt with by the SO
 210             alarm_id = values['ack_details']['alarm_uuid']
 211
 212             response = self.update_alarm_state(
 213                 alarm_endpoint, auth_token, alarm_id)
 214
 215             # Log if an alarm was reset
 216             if response is True:
 217                 log.info("Acknowledged the alarm and cleared it.")
 218             else:
 219                 log.warning("Failed to acknowledge/clear the alarm.")
 220
 221         elif message.key == "update_alarm_request":
 222             # Update alarm configurations
 223             alarm_details = values['alarm_update_request']
 224
 225             alarm_id, status = self.update_alarm(
 226                 alarm_endpoint, auth_token, alarm_details, vim_config)
 227
 228             # Generate a response for an update request
 229             try:
 230                 resp_message = self._response.generate_response(
 231                     'update_alarm_response', alarm_id=alarm_id,
 232                     cor_id=alarm_details['correlation_id'],
 233                     status=status)
 234                 log.info("Response message: %s", resp_message)
 235                 self._producer.update_alarm_response(
 236                     'update_alarm_response', resp_message)
 237             except Exception:
 238                 log.exception("Failed to send an update response: ")
 239
 240         else:
 241             log.debug("Unknown key, no action will be performed")
 242
 243         return
 244
 245     def delete_alarm(self, endpoint, auth_token, alarm_id):
 246         """Delete alarm function."""
 247         url = "{}/v2/alarms/%s".format(endpoint) % alarm_id
 248
 249         try:
 250             result = Common.perform_request(
 251                 url, auth_token, req_type="delete")
 252             if str(result.status_code) == "404":
 253                 log.info("Alarm doesn't exist: %s", result.status_code)
 254                 # If status code is 404 alarm did not exist
 255                 return False
 256             else:
 257                 return True
 258
 259         except Exception:
 260             log.exception("Failed to delete alarm %s :", alarm_id)
 261         return False
 262
 263     def list_alarms(self, endpoint, auth_token, list_details):
 264         """Generate the requested list of alarms."""
 265         url = "{}/v2/alarms/".format(endpoint)
 266         a_list, name_list, sev_list, res_list = [], [], [], []
 267
 268         # TODO(mcgoughh): for now resource_id is a mandatory field
 269         # Check for a resource id
 270         try:
 271             resource = list_details['resource_uuid']
 272         except KeyError as exc:
 273             log.warning("Resource id not specified for list request: %s", exc)
 274             return None
 275
 276         # Checking what fields are specified for a list request
 277         try:
 278             name = list_details['alarm_name'].lower()
 279             if name not in ALARM_NAMES.keys():
 280                 log.warning("This alarm is not supported, won't be used!")
 281                 name = None
 282         except KeyError as exc:
 283             log.info("Alarm name isn't specified.")
 284             name = None
 285
 286         try:
 287             severity = list_details['severity'].lower()
 288             sev = SEVERITIES[severity]
 289         except KeyError as exc:
 290             log.info("Severity is unspecified/incorrectly configured")
 291             sev = None
 292
 293         # Perform the request to get the desired list
 294         try:
 295             result = Common.perform_request(
 296                 url, auth_token, req_type="get")
 297
 298             if result is not None:
 299                 # Get list based on resource id
 300                 for alarm in json.loads(result.text):
 301                     rule = alarm['gnocchi_resources_threshold_rule']
 302                     if resource == rule['resource_id']:
 303                         res_list.append(alarm)
 304                     if not res_list:
 305                         log.info("No alarms for this resource")
 306                         return a_list
 307
 308                 # Generate specified listed if requested
 309                 if name is not None and sev is not None:
 310                     log.info("Return a list of %s alarms with %s severity.",
 311                              name, sev)
 312                     for alarm in json.loads(result.text):
 313                         if name == alarm['name']:
 314                             name_list.append(alarm)
 315                     for alarm in json.loads(result.text):
 316                         if sev == alarm['severity']:
 317                             sev_list.append(alarm)
 318                     name_sev_list = list(set(name_list).intersection(sev_list))
 319                     a_list = list(set(name_sev_list).intersection(res_list))
 320                 elif name is not None:
 321                     log.info("Returning a %s list of alarms.", name)
 322                     for alarm in json.loads(result.text):
 323                         if name == alarm['name']:
 324                             name_list.append(alarm)
 325                     a_list = list(set(name_list).intersection(res_list))
 326                 elif sev is not None:
 327                     log.info("Returning %s severity alarm list.", sev)
 328                     for alarm in json.loads(result.text):
 329                         if sev == alarm['severity']:
 330                             sev_list.append(alarm)
 331                     a_list = list(set(sev_list).intersection(res_list))
 332                 else:
 333                     log.info("Returning an entire list of alarms.")
 334                     a_list = res_list
 335             else:
 336                 log.info("There are no alarms!")
 337
 338         except Exception as exc:
 339             log.info("Failed to generate required list: %s", exc)
 340             return None
 341
 342         return a_list
 343
 344     def update_alarm_state(self, endpoint, auth_token, alarm_id):
 345         """Set the state of an alarm to ok when ack message is received."""
 346         url = "{}/v2/alarms/%s/state".format(endpoint) % alarm_id
 347         payload = json.dumps("ok")
 348
 349         try:
 350             Common.perform_request(
 351                 url, auth_token, req_type="put", payload=payload)
 352             return True
 353         except Exception:
 354             log.exception("Unable to update alarm state: ")
 355         return False
 356
 357     def update_alarm(self, endpoint, auth_token, values, vim_config):
 358         """Get alarm name for an alarm configuration update."""
 359         # Get already existing alarm details
 360         url = "{}/v2/alarms/%s".format(endpoint) % values['alarm_uuid']
 361
 362         # Gets current configurations about the alarm
 363         try:
 364             result = Common.perform_request(
 365                 url, auth_token, req_type="get")
 366             alarm_name = json.loads(result.text)['name']
 367             rule = json.loads(result.text)['gnocchi_resources_threshold_rule']
 368             alarm_state = json.loads(result.text)['state']
 369             resource_id = rule['resource_id']
 370             metric_name = [key for key, value in six.iteritems(METRIC_MAPPINGS) if value == rule['metric']][0]
 371         except Exception as exc:
 372             log.warning("Failed to retrieve existing alarm info: %s.\
 373                      Can only update OSM alarms.", exc)
 374             return None, False
 375
 376         # Generates and check payload configuration for alarm update
 377         if 'granularity' in vim_config and 'granularity' not in values:
 378             values['granularity'] = vim_config['granularity']
 379         payload = self.check_payload(values, metric_name, resource_id,
 380                                      alarm_name, alarm_state=alarm_state)
 381
 382         # Updates the alarm configurations with the valid payload
 383         if payload is not None:
 384             try:
 385                 update_alarm = Common.perform_request(
 386                     url, auth_token, req_type="put", payload=payload)
 387
 388                 return json.loads(update_alarm.text)['alarm_id'], True
 389             except Exception as exc:
 390                 log.warning("Alarm update could not be performed: %s", exc)
 391                 return None, False
 392         return None, False
 393
 394     def check_payload(self, values, metric_name, resource_id,
 395                       alarm_name, alarm_state=None):
 396         """Check that the payload is configuration for update/create alarm."""
 397         try:
 398             cfg = Config.instance()
 399             # Check state and severity
 400
 401             severity = 'critical'
 402             if 'severity' in values:
 403                 severity = values['severity'].lower()
 404
 405             if severity == "indeterminate":
 406                 alarm_state = "insufficient data"
 407             if alarm_state is None:
 408                 alarm_state = "ok"
 409
 410             statistic = values['statistic'].lower()
 411
 412             granularity = cfg.OS_DEFAULT_GRANULARITY
 413             if 'granularity' in values:
 414                 granularity = values['granularity']
 415
 416             resource_type = 'generic'
 417             if 'resource_type' in values:
 418                 resource_type = values['resource_type'].lower()
 419
 420             # Try to configure the payload for the update/create request
 421             # Can only update: threshold, operation, statistic and
 422             # the severity of the alarm
 423             rule = {'threshold': values['threshold_value'],
 424                     'comparison_operator': values['operation'].lower(),
 425                     'metric': METRIC_MAPPINGS[metric_name],
 426                     'resource_id': resource_id,
 427                     'resource_type': resource_type,
 428                     'aggregation_method': STATISTICS[statistic],
 429                     'granularity': granularity, }
 430             payload = json.dumps({'state': alarm_state,
 431                                   'name': alarm_name,
 432                                   'severity': SEVERITIES[severity],
 433                                   'type': 'gnocchi_resources_threshold',
 434                                   'gnocchi_resources_threshold_rule': rule,
 435                                   'alarm_actions': [cfg.OS_NOTIFIER_URI], })
 436             return payload
 437         except KeyError as exc:
 438             log.warning("Alarm is not configured correctly: %s", exc)
 439         return None
 440
 441     def get_alarm_state(self, endpoint, auth_token, alarm_id):
 442         """Get the state of the alarm."""
 443         url = "{}/v2/alarms/%s/state".format(endpoint) % alarm_id
 444
 445         try:
 446             alarm_state = Common.perform_request(
 447                 url, auth_token, req_type="get")
 448             return json.loads(alarm_state.text)
 449         except Exception as exc:
 450             log.warning("Failed to get the state of the alarm:%s", exc)
 451         return None
 452
 453     def check_for_metric(self, auth_token, metric_endpoint, m_name, r_id):
 454         """Check for the alarm metric."""
 455         try:
 456             url = "{}/v1/metric?sort=name:asc".format(metric_endpoint)
 457             result = Common.perform_request(
 458                 url, auth_token, req_type="get")
 459             metric_list = []
 460             metrics_partial = json.loads(result.text)
 461             for metric in metrics_partial:
 462                 metric_list.append(metric)
 463
 464             while len(json.loads(result.text)) > 0:
 465                 last_metric_id = metrics_partial[-1]['id']
 466                 url = "{}/v1/metric?sort=name:asc&marker={}".format(metric_endpoint, last_metric_id)
 467                 result = Common.perform_request(
 468                     url, auth_token, req_type="get")
 469                 if len(json.loads(result.text)) > 0:
 470                     metrics_partial = json.loads(result.text)
 471                     for metric in metrics_partial:
 472                         metric_list.append(metric)
 473             metric_id = None
 474             for metric in metric_list:
 475                 name = metric['name']
 476                 resource = metric['resource_id']
 477                 if name == METRIC_MAPPINGS[m_name] and resource == r_id:
 478                     metric_id = metric['id']
 479             log.info("The required metric exists, an alarm will be created.")
 480             return metric_id
 481         except Exception as exc:
 482             log.info("Desired Gnocchi metric not found:%s", exc)
 483         return None