osm_lcm/lcm.py

   1 #!/usr/bin/python3
   2 # -*- coding: utf-8 -*-
   3
   4 ##
   5 # Copyright 2018 Telefonica S.A.
   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
  20 import asyncio
  21 import yaml
  22 import logging
  23 import logging.handlers
  24 import getopt
  25 import sys
  26 import ROclient
  27 import ns
  28 import vim_sdn
  29 import netslice
  30 from lcm_utils import versiontuple, LcmException, TaskRegistry, LcmExceptionExit
  31
  32 # from osm_lcm import version as lcm_version, version_date as lcm_version_date, ROclient
  33 from osm_common import dbmemory, dbmongo, fslocal, msglocal, msgkafka
  34 from osm_common import version as common_version
  35 from osm_common.dbbase import DbException
  36 from osm_common.fsbase import FsException
  37 from osm_common.msgbase import MsgException
  38 from os import environ, path
  39 from n2vc import version as n2vc_version
  40
  41
  42 __author__ = "Alfonso Tierno"
  43 min_RO_version = [0, 6, 0]
  44 min_n2vc_version = "0.0.2"
  45 min_common_version = "0.1.11"
  46 # uncomment if LCM is installed as library and installed, and get them from __init__.py
  47 lcm_version = '0.1.29'
  48 lcm_version_date = '2018-12-14'
  49
  50
  51 class Lcm:
  52
  53     ping_interval_pace = 120  # how many time ping is send once is confirmed all is running
  54     ping_interval_boot = 5    # how many time ping is sent when booting
  55
  56     def __init__(self, config_file, loop=None):
  57         """
  58         Init, Connect to database, filesystem storage, and messaging
  59         :param config: two level dictionary with configuration. Top level should contain 'database', 'storage',
  60         :return: None
  61         """
  62
  63         self.db = None
  64         self.msg = None
  65         self.fs = None
  66         self.pings_not_received = 1
  67
  68         # contains created tasks/futures to be able to cancel
  69         self.lcm_tasks = TaskRegistry()
  70         # logging
  71         self.logger = logging.getLogger('lcm')
  72         # load configuration
  73         config = self.read_config_file(config_file)
  74         self.config = config
  75         self.ro_config = {
  76             "endpoint_url": "http://{}:{}/openmano".format(config["RO"]["host"], config["RO"]["port"]),
  77             "tenant": config.get("tenant", "osm"),
  78             "logger_name": "lcm.ROclient",
  79             "loglevel": "ERROR",
  80         }
  81
  82         self.vca_config = config["VCA"]
  83
  84         self.loop = loop or asyncio.get_event_loop()
  85
  86         # logging
  87         log_format_simple = "%(asctime)s %(levelname)s %(name)s %(filename)s:%(lineno)s %(message)s"
  88         log_formatter_simple = logging.Formatter(log_format_simple, datefmt='%Y-%m-%dT%H:%M:%S')
  89         config["database"]["logger_name"] = "lcm.db"
  90         config["storage"]["logger_name"] = "lcm.fs"
  91         config["message"]["logger_name"] = "lcm.msg"
  92         if config["global"].get("logfile"):
  93             file_handler = logging.handlers.RotatingFileHandler(config["global"]["logfile"],
  94                                                                 maxBytes=100e6, backupCount=9, delay=0)
  95             file_handler.setFormatter(log_formatter_simple)
  96             self.logger.addHandler(file_handler)
  97         if not config["global"].get("nologging"):
  98             str_handler = logging.StreamHandler()
  99             str_handler.setFormatter(log_formatter_simple)
 100             self.logger.addHandler(str_handler)
 101
 102         if config["global"].get("loglevel"):
 103             self.logger.setLevel(config["global"]["loglevel"])
 104
 105         # logging other modules
 106         for k1, logname in {"message": "lcm.msg", "database": "lcm.db", "storage": "lcm.fs"}.items():
 107             config[k1]["logger_name"] = logname
 108             logger_module = logging.getLogger(logname)
 109             if config[k1].get("logfile"):
 110                 file_handler = logging.handlers.RotatingFileHandler(config[k1]["logfile"],
 111                                                                     maxBytes=100e6, backupCount=9, delay=0)
 112                 file_handler.setFormatter(log_formatter_simple)
 113                 logger_module.addHandler(file_handler)
 114             if config[k1].get("loglevel"):
 115                 logger_module.setLevel(config[k1]["loglevel"])
 116         self.logger.critical("starting osm/lcm version {} {}".format(lcm_version, lcm_version_date))
 117
 118         # check version of N2VC
 119         # TODO enhance with int conversion or from distutils.version import LooseVersion
 120         # or with list(map(int, version.split(".")))
 121         if versiontuple(n2vc_version) < versiontuple(min_n2vc_version):
 122             raise LcmException("Not compatible osm/N2VC version '{}'. Needed '{}' or higher".format(
 123                 n2vc_version, min_n2vc_version))
 124         # check version of common
 125         if versiontuple(common_version) < versiontuple(min_common_version):
 126             raise LcmException("Not compatible osm/common version '{}'. Needed '{}' or higher".format(
 127                 common_version, min_common_version))
 128
 129         try:
 130             # TODO check database version
 131             if config["database"]["driver"] == "mongo":
 132                 self.db = dbmongo.DbMongo()
 133                 self.db.db_connect(config["database"])
 134             elif config["database"]["driver"] == "memory":
 135                 self.db = dbmemory.DbMemory()
 136                 self.db.db_connect(config["database"])
 137             else:
 138                 raise LcmException("Invalid configuration param '{}' at '[database]':'driver'".format(
 139                     config["database"]["driver"]))
 140
 141             if config["storage"]["driver"] == "local":
 142                 self.fs = fslocal.FsLocal()
 143                 self.fs.fs_connect(config["storage"])
 144             else:
 145                 raise LcmException("Invalid configuration param '{}' at '[storage]':'driver'".format(
 146                     config["storage"]["driver"]))
 147
 148             if config["message"]["driver"] == "local":
 149                 self.msg = msglocal.MsgLocal()
 150                 self.msg.connect(config["message"])
 151             elif config["message"]["driver"] == "kafka":
 152                 self.msg = msgkafka.MsgKafka()
 153                 self.msg.connect(config["message"])
 154             else:
 155                 raise LcmException("Invalid configuration param '{}' at '[message]':'driver'".format(
 156                     config["storage"]["driver"]))
 157         except (DbException, FsException, MsgException) as e:
 158             self.logger.critical(str(e), exc_info=True)
 159             raise LcmException(str(e))
 160
 161         self.ns = ns.NsLcm(self.db, self.msg, self.fs, self.lcm_tasks, self.ro_config, self.vca_config, self.loop)
 162         self.netslice = netslice.NetsliceLcm(self.db, self.msg, self.fs, self.lcm_tasks, self.ro_config,
 163                                              self.vca_config, self.loop)
 164         self.vim = vim_sdn.VimLcm(self.db, self.msg, self.fs, self.lcm_tasks, self.ro_config, self.loop)
 165         self.wim = vim_sdn.WimLcm(self.db, self.msg, self.fs, self.lcm_tasks, self.ro_config, self.loop)
 166         self.sdn = vim_sdn.SdnLcm(self.db, self.msg, self.fs, self.lcm_tasks, self.ro_config, self.loop)
 167
 168     async def check_RO_version(self):
 169         try:
 170             RO = ROclient.ROClient(self.loop, **self.ro_config)
 171             RO_version = await RO.get_version()
 172             if RO_version < min_RO_version:
 173                 raise LcmException("Not compatible osm/RO version '{}.{}.{}'. Needed '{}.{}.{}' or higher".format(
 174                     *RO_version, *min_RO_version
 175                 ))
 176         except ROclient.ROClientException as e:
 177             error_text = "Error while conneting to osm/RO " + str(e)
 178             self.logger.critical(error_text, exc_info=True)
 179             raise LcmException(error_text)
 180
 181     async def test(self, param=None):
 182         self.logger.debug("Starting/Ending test task: {}".format(param))
 183
 184     async def kafka_ping(self):
 185         self.logger.debug("Task kafka_ping Enter")
 186         consecutive_errors = 0
 187         first_start = True
 188         kafka_has_received = False
 189         self.pings_not_received = 1
 190         while True:
 191             try:
 192                 await self.msg.aiowrite("admin", "ping", {"from": "lcm", "to": "lcm"}, self.loop)
 193                 # time between pings are low when it is not received and at starting
 194                 wait_time = self.ping_interval_boot if not kafka_has_received else self.ping_interval_pace
 195                 if not self.pings_not_received:
 196                     kafka_has_received = True
 197                 self.pings_not_received += 1
 198                 await asyncio.sleep(wait_time, loop=self.loop)
 199                 if self.pings_not_received > 10:
 200                     raise LcmException("It is not receiving pings from Kafka bus")
 201                 consecutive_errors = 0
 202                 first_start = False
 203             except LcmException:
 204                 raise
 205             except Exception as e:
 206                 # if not first_start is the first time after starting. So leave more time and wait
 207                 # to allow kafka starts
 208                 if consecutive_errors == 8 if not first_start else 30:
 209                     self.logger.error("Task kafka_read task exit error too many errors. Exception: {}".format(e))
 210                     raise
 211                 consecutive_errors += 1
 212                 self.logger.error("Task kafka_read retrying after Exception {}".format(e))
 213                 wait_time = 1 if not first_start else 5
 214                 await asyncio.sleep(wait_time, loop=self.loop)
 215
 216     def kafka_read_callback(self, topic, command, params):
 217         order_id = 1
 218
 219         if topic != "admin" and command != "ping":
 220             self.logger.debug("Task kafka_read receives {} {}: {}".format(topic, command, params))
 221         self.consecutive_errors = 0
 222         self.first_start = False
 223         order_id += 1
 224         if command == "exit":
 225             raise LcmExceptionExit
 226         elif command.startswith("#"):
 227             return
 228         elif command == "echo":
 229             # just for test
 230             print(params)
 231             sys.stdout.flush()
 232             return
 233         elif command == "test":
 234             asyncio.Task(self.test(params), loop=self.loop)
 235             return
 236
 237         if topic == "admin":
 238             if command == "ping" and params["to"] == "lcm" and params["from"] == "lcm":
 239                 self.pings_not_received = 0
 240             return
 241         elif topic == "ns":
 242             if command == "instantiate":
 243                 # self.logger.debug("Deploying NS {}".format(nsr_id))
 244                 nslcmop = params
 245                 nslcmop_id = nslcmop["_id"]
 246                 nsr_id = nslcmop["nsInstanceId"]
 247                 task = asyncio.ensure_future(self.ns.instantiate(nsr_id, nslcmop_id))
 248                 self.lcm_tasks.register("ns", nsr_id, nslcmop_id, "ns_instantiate", task)
 249                 return
 250             elif command == "terminate":
 251                 # self.logger.debug("Deleting NS {}".format(nsr_id))
 252                 nslcmop = params
 253                 nslcmop_id = nslcmop["_id"]
 254                 nsr_id = nslcmop["nsInstanceId"]
 255                 self.lcm_tasks.cancel(topic, nsr_id)
 256                 task = asyncio.ensure_future(self.ns.terminate(nsr_id, nslcmop_id))
 257                 self.lcm_tasks.register("ns", nsr_id, nslcmop_id, "ns_terminate", task)
 258                 return
 259             elif command == "action":
 260                 # self.logger.debug("Update NS {}".format(nsr_id))
 261                 nslcmop = params
 262                 nslcmop_id = nslcmop["_id"]
 263                 nsr_id = nslcmop["nsInstanceId"]
 264                 task = asyncio.ensure_future(self.ns.action(nsr_id, nslcmop_id))
 265                 self.lcm_tasks.register("ns", nsr_id, nslcmop_id, "ns_action", task)
 266                 return
 267             elif command == "scale":
 268                 # self.logger.debug("Update NS {}".format(nsr_id))
 269                 nslcmop = params
 270                 nslcmop_id = nslcmop["_id"]
 271                 nsr_id = nslcmop["nsInstanceId"]
 272                 task = asyncio.ensure_future(self.ns.scale(nsr_id, nslcmop_id))
 273                 self.lcm_tasks.register("ns", nsr_id, nslcmop_id, "ns_scale", task)
 274                 return
 275             elif command == "show":
 276                 try:
 277                     db_nsr = self.db.get_one("nsrs", {"_id": nsr_id})
 278                     print("nsr:\n    _id={}\n    operational-status: {}\n    config-status: {}"
 279                           "\n    detailed-status: {}\n    deploy: {}\n    tasks: {}"
 280                           "".format(nsr_id, db_nsr["operational-status"], db_nsr["config-status"],
 281                                     db_nsr["detailed-status"],
 282                                     db_nsr["_admin"]["deployed"], self.lcm_ns_tasks.get(nsr_id)))
 283                 except Exception as e:
 284                     print("nsr {} not found: {}".format(nsr_id, e))
 285                 sys.stdout.flush()
 286                 return
 287             elif command == "deleted":
 288                 return  # TODO cleaning of task just in case should be done
 289             elif command in ("terminated", "instantiated", "scaled", "actioned"):  # "scaled-cooldown-time"
 290                 return
 291         elif topic == "nsi":  # netslice LCM processes (instantiate, terminate, etc)
 292             if command == "instantiate":
 293                 # self.logger.debug("Instantiating Network Slice {}".format(nsilcmop["netsliceInstanceId"]))
 294                 nsilcmop = params
 295                 nsilcmop_id = nsilcmop["_id"]  # slice operation id
 296                 nsir_id = nsilcmop["netsliceInstanceId"]  # slice record id
 297                 task = asyncio.ensure_future(self.netslice.instantiate(nsir_id, nsilcmop_id))
 298                 self.lcm_tasks.register("nsi", nsir_id, nsilcmop_id, "nsi_instantiate", task)
 299                 return
 300             elif command == "terminate":
 301                 # self.logger.debug("Terminating Network Slice NS {}".format(nsilcmop["netsliceInstanceId"]))
 302                 nsilcmop = params
 303                 nsilcmop_id = nsilcmop["_id"]  # slice operation id
 304                 nsir_id = nsilcmop["netsliceInstanceId"]  # slice record id
 305                 self.lcm_tasks.cancel(topic, nsir_id)
 306                 task = asyncio.ensure_future(self.netslice.terminate(nsir_id, nsilcmop_id))
 307                 self.lcm_tasks.register("nsi", nsir_id, nsilcmop_id, "nsi_terminate", task)
 308                 return
 309             elif command == "show":
 310                 try:
 311                     db_nsir = self.db.get_one("nsirs", {"_id": nsir_id})
 312                     print("nsir:\n    _id={}\n    operational-status: {}\n    config-status: {}"
 313                           "\n    detailed-status: {}\n    deploy: {}\n    tasks: {}"
 314                           "".format(nsir_id, db_nsir["operational-status"], db_nsir["config-status"],
 315                                     db_nsir["detailed-status"],
 316                                     db_nsir["_admin"]["deployed"], self.lcm_netslice_tasks.get(nsir_id)))
 317                 except Exception as e:
 318                     print("nsir {} not found: {}".format(nsir_id, e))
 319                 sys.stdout.flush()
 320                 return
 321             elif command == "deleted":
 322                 return  # TODO cleaning of task just in case should be done
 323             elif command in ("terminated", "instantiated", "scaled", "actioned"):  # "scaled-cooldown-time"
 324                 return
 325         elif topic == "vim_account":
 326             vim_id = params["_id"]
 327             if command == "create":
 328                 task = asyncio.ensure_future(self.vim.create(params, order_id))
 329                 self.lcm_tasks.register("vim_account", vim_id, order_id, "vim_create", task)
 330                 return
 331             elif command == "delete":
 332                 self.lcm_tasks.cancel(topic, vim_id)
 333                 task = asyncio.ensure_future(self.vim.delete(vim_id, order_id))
 334                 self.lcm_tasks.register("vim_account", vim_id, order_id, "vim_delete", task)
 335                 return
 336             elif command == "show":
 337                 print("not implemented show with vim_account")
 338                 sys.stdout.flush()
 339                 return
 340             elif command == "edit":
 341                 task = asyncio.ensure_future(self.vim.edit(params, order_id))
 342                 self.lcm_tasks.register("vim_account", vim_id, order_id, "vim_edit", task)
 343                 return
 344         elif topic == "wim_account":
 345             wim_id = params["_id"]
 346             if command == "create":
 347                 task = asyncio.ensure_future(self.wim.create(params, order_id))
 348                 self.lcm_tasks.register("wim_account", wim_id, order_id, "wim_create", task)
 349                 return
 350             elif command == "delete":
 351                 self.lcm_tasks.cancel(topic, wim_id)
 352                 task = asyncio.ensure_future(self.wim.delete(wim_id, order_id))
 353                 self.lcm_tasks.register("wim_account", wim_id, order_id, "wim_delete", task)
 354                 return
 355             elif command == "show":
 356                 print("not implemented show with wim_account")
 357                 sys.stdout.flush()
 358                 return
 359             elif command == "edit":
 360                 task = asyncio.ensure_future(self.wim.edit(params, order_id))
 361                 self.lcm_tasks.register("wim_account", wim_id, order_id, "wim_edit", task)
 362                 return
 363         elif topic == "sdn":
 364             _sdn_id = params["_id"]
 365             if command == "create":
 366                 task = asyncio.ensure_future(self.sdn.create(params, order_id))
 367                 self.lcm_tasks.register("sdn", _sdn_id, order_id, "sdn_create", task)
 368                 return
 369             elif command == "delete":
 370                 self.lcm_tasks.cancel(topic, _sdn_id)
 371                 task = asyncio.ensure_future(self.sdn.delete(_sdn_id, order_id))
 372                 self.lcm_tasks.register("sdn", _sdn_id, order_id, "sdn_delete", task)
 373                 return
 374             elif command == "edit":
 375                 task = asyncio.ensure_future(self.sdn.edit(params, order_id))
 376                 self.lcm_tasks.register("sdn", _sdn_id, order_id, "sdn_edit", task)
 377                 return
 378         self.logger.critical("unknown topic {} and command '{}'".format(topic, command))
 379
 380     async def kafka_read(self):
 381         self.logger.debug("Task kafka_read Enter")
 382         # future = asyncio.Future()
 383         self.consecutive_errors = 0
 384         self.first_start = True
 385         while self.consecutive_errors < 10:
 386             try:
 387                 topics = ("admin", "ns", "vim_account", "wim_account", "sdn", "nsi")
 388                 await self.msg.aioread(topics, self.loop, self.kafka_read_callback)
 389
 390             except LcmExceptionExit:
 391                 self.logger.debug("Bye!")
 392                 break
 393             except Exception as e:
 394                 # if not first_start is the first time after starting. So leave more time and wait
 395                 # to allow kafka starts
 396                 if self.consecutive_errors == 8 if not self.first_start else 30:
 397                     self.logger.error("Task kafka_read task exit error too many errors. Exception: {}".format(e))
 398                     raise
 399                 self.consecutive_errors += 1
 400                 self.logger.error("Task kafka_read retrying after Exception {}".format(e))
 401                 wait_time = 2 if not self.first_start else 5
 402                 await asyncio.sleep(wait_time, loop=self.loop)
 403
 404         # self.logger.debug("Task kafka_read terminating")
 405         self.logger.debug("Task kafka_read exit")
 406
 407     def health_check(self):
 408
 409         global exit_code
 410         task = None
 411         exit_code = 1
 412
 413         def health_check_callback(topic, command, params):
 414             global exit_code
 415             print("receiving callback {} {} {}".format(topic, command, params))
 416             if topic == "admin" and command == "ping" and params["to"] == "lcm" and params["from"] == "lcm":
 417                 # print("received LCM ping")
 418                 exit_code = 0
 419                 task.cancel()
 420
 421         try:
 422             task = asyncio.ensure_future(self.msg.aioread(("admin",), self.loop, health_check_callback))
 423             self.loop.run_until_complete(task)
 424         except Exception:
 425             pass
 426         exit(exit_code)
 427
 428     def start(self):
 429
 430         # check RO version
 431         self.loop.run_until_complete(self.check_RO_version())
 432
 433         self.loop.run_until_complete(asyncio.gather(
 434             self.kafka_read(),
 435             self.kafka_ping()
 436         ))
 437         # TODO
 438         # self.logger.debug("Terminating cancelling creation tasks")
 439         # self.lcm_tasks.cancel("ALL", "create")
 440         # timeout = 200
 441         # while self.is_pending_tasks():
 442         #     self.logger.debug("Task kafka_read terminating. Waiting for tasks termination")
 443         #     await asyncio.sleep(2, loop=self.loop)
 444         #     timeout -= 2
 445         #     if not timeout:
 446         #         self.lcm_tasks.cancel("ALL", "ALL")
 447         self.loop.close()
 448         self.loop = None
 449         if self.db:
 450             self.db.db_disconnect()
 451         if self.msg:
 452             self.msg.disconnect()
 453         if self.fs:
 454             self.fs.fs_disconnect()
 455
 456     def read_config_file(self, config_file):
 457         # TODO make a [ini] + yaml inside parser
 458         # the configparser library is not suitable, because it does not admit comments at the end of line,
 459         # and not parse integer or boolean
 460         try:
 461             with open(config_file) as f:
 462                 conf = yaml.load(f)
 463             for k, v in environ.items():
 464                 if not k.startswith("OSMLCM_"):
 465                     continue
 466                 k_items = k.lower().split("_")
 467                 if len(k_items) < 3:
 468                     continue
 469                 if k_items[1] in ("ro", "vca"):
 470                     # put in capital letter
 471                     k_items[1] = k_items[1].upper()
 472                 c = conf
 473                 try:
 474                     for k_item in k_items[1:-1]:
 475                         c = c[k_item]
 476                     if k_items[-1] == "port":
 477                         c[k_items[-1]] = int(v)
 478                     else:
 479                         c[k_items[-1]] = v
 480                 except Exception as e:
 481                     self.logger.warn("skipping environ '{}' on exception '{}'".format(k, e))
 482
 483             return conf
 484         except Exception as e:
 485             self.logger.critical("At config file '{}': {}".format(config_file, e))
 486             exit(1)
 487
 488
 489 def usage():
 490     print("""Usage: {} [options]
 491         -c|--config [configuration_file]: loads the configuration file (default: ./nbi.cfg)
 492         --health-check: do not run lcm, but inspect kafka bus to determine if lcm is healthy
 493         -h|--help: shows this help
 494         """.format(sys.argv[0]))
 495     # --log-socket-host HOST: send logs to this host")
 496     # --log-socket-port PORT: send logs using this port (default: 9022)")
 497
 498
 499 if __name__ == '__main__':
 500     try:
 501         # load parameters and configuration
 502         opts, args = getopt.getopt(sys.argv[1:], "hc:", ["config=", "help", "health-check"])
 503         # TODO add  "log-socket-host=", "log-socket-port=", "log-file="
 504         config_file = None
 505         health_check = None
 506         for o, a in opts:
 507             if o in ("-h", "--help"):
 508                 usage()
 509                 sys.exit()
 510             elif o in ("-c", "--config"):
 511                 config_file = a
 512             elif o == "--health-check":
 513                 health_check = True
 514             # elif o == "--log-socket-port":
 515             #     log_socket_port = a
 516             # elif o == "--log-socket-host":
 517             #     log_socket_host = a
 518             # elif o == "--log-file":
 519             #     log_file = a
 520             else:
 521                 assert False, "Unhandled option"
 522         if config_file:
 523             if not path.isfile(config_file):
 524                 print("configuration file '{}' not exist".format(config_file), file=sys.stderr)
 525                 exit(1)
 526         else:
 527             for config_file in (__file__[:__file__.rfind(".")] + ".cfg", "./lcm.cfg", "/etc/osm/lcm.cfg"):
 528                 if path.isfile(config_file):
 529                     break
 530             else:
 531                 print("No configuration file 'lcm.cfg' found neither at local folder nor at /etc/osm/", file=sys.stderr)
 532                 exit(1)
 533         lcm = Lcm(config_file)
 534         if health_check:
 535             lcm.health_check()
 536         else:
 537             lcm.start()
 538     except (LcmException, getopt.GetoptError) as e:
 539         print(str(e), file=sys.stderr)
 540         # usage()
 541         exit(1)