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
  21 # DEBUG WITH PDB
  22 import os
  23 import pdb
  24
  25 import asyncio
  26 import yaml
  27 import logging
  28 import logging.handlers
  29 import getopt
  30 import sys
  31
  32 from osm_lcm import ns, vim_sdn, netslice
  33 from osm_lcm.ng_ro import NgRoException, NgRoClient
  34 from osm_lcm.ROclient import ROClient, ROClientException
  35
  36 from time import time
  37 from osm_lcm.lcm_utils import versiontuple, LcmException, TaskRegistry, LcmExceptionExit
  38 from osm_lcm import version as lcm_version, version_date as lcm_version_date
  39
  40 from osm_common import msglocal, msgkafka
  41 from osm_common import version as common_version
  42 from osm_common.dbbase import DbException
  43 from osm_common.fsbase import FsException
  44 from osm_common.msgbase import MsgException
  45 from osm_lcm.data_utils.database.database import Database
  46 from osm_lcm.data_utils.filesystem.filesystem import Filesystem
  47 from os import environ, path
  48 from random import choice as random_choice
  49 from n2vc import version as n2vc_version
  50 import traceback
  51
  52 if os.getenv("OSMLCM_PDB_DEBUG", None) is not None:
  53     pdb.set_trace()
  54
  55
  56 __author__ = "Alfonso Tierno"
  57 min_RO_version = "6.0.2"
  58 min_n2vc_version = "0.0.2"
  59
  60 min_common_version = "0.1.19"
  61 health_check_file = (
  62     path.expanduser("~") + "/time_last_ping"
  63 )  # TODO find better location for this file
  64
  65
  66 class Lcm:
  67
  68     ping_interval_pace = (
  69         120  # how many time ping is send once is confirmed all is running
  70     )
  71     ping_interval_boot = 5  # how many time ping is sent when booting
  72     cfg_logger_name = {
  73         "message": "lcm.msg",
  74         "database": "lcm.db",
  75         "storage": "lcm.fs",
  76         "tsdb": "lcm.prometheus",
  77     }
  78     # ^ contains for each section at lcm.cfg the used logger name
  79
  80     def __init__(self, config_file, loop=None):
  81         """
  82         Init, Connect to database, filesystem storage, and messaging
  83         :param config: two level dictionary with configuration. Top level should contain 'database', 'storage',
  84         :return: None
  85         """
  86         self.db = None
  87         self.msg = None
  88         self.msg_admin = None
  89         self.fs = None
  90         self.pings_not_received = 1
  91         self.consecutive_errors = 0
  92         self.first_start = False
  93
  94         # logging
  95         self.logger = logging.getLogger("lcm")
  96         # get id
  97         self.worker_id = self.get_process_id()
  98         # load configuration
  99         config = self.read_config_file(config_file)
 100         self.config = config
 101         self.config["ro_config"] = {
 102             "ng": config["RO"].get("ng", False),
 103             "uri": config["RO"].get("uri"),
 104             "tenant": config.get("tenant", "osm"),
 105             "logger_name": "lcm.roclient",
 106             "loglevel": config["RO"].get("loglevel", "ERROR"),
 107         }
 108         if not self.config["ro_config"]["uri"]:
 109             self.config["ro_config"]["uri"] = "http://{}:{}/".format(
 110                 config["RO"]["host"], config["RO"]["port"]
 111             )
 112         elif (
 113             "/ro" in self.config["ro_config"]["uri"][-4:]
 114             or "/openmano" in self.config["ro_config"]["uri"][-10:]
 115         ):
 116             # uri ends with '/ro', '/ro/', '/openmano', '/openmano/'
 117             index = self.config["ro_config"]["uri"][-1].rfind("/")
 118             self.config["ro_config"]["uri"] = self.config["ro_config"]["uri"][index + 1]
 119
 120         self.loop = loop or asyncio.get_event_loop()
 121         self.ns = (
 122             self.netslice
 123         ) = (
 124             self.vim
 125         ) = self.wim = self.sdn = self.k8scluster = self.vca = self.k8srepo = None
 126
 127         # logging
 128         log_format_simple = (
 129             "%(asctime)s %(levelname)s %(name)s %(filename)s:%(lineno)s %(message)s"
 130         )
 131         log_formatter_simple = logging.Formatter(
 132             log_format_simple, datefmt="%Y-%m-%dT%H:%M:%S"
 133         )
 134         config["database"]["logger_name"] = "lcm.db"
 135         config["storage"]["logger_name"] = "lcm.fs"
 136         config["message"]["logger_name"] = "lcm.msg"
 137         if config["global"].get("logfile"):
 138             file_handler = logging.handlers.RotatingFileHandler(
 139                 config["global"]["logfile"], maxBytes=100e6, backupCount=9, delay=0
 140             )
 141             file_handler.setFormatter(log_formatter_simple)
 142             self.logger.addHandler(file_handler)
 143         if not config["global"].get("nologging"):
 144             str_handler = logging.StreamHandler()
 145             str_handler.setFormatter(log_formatter_simple)
 146             self.logger.addHandler(str_handler)
 147
 148         if config["global"].get("loglevel"):
 149             self.logger.setLevel(config["global"]["loglevel"])
 150
 151         # logging other modules
 152         for k1, logname in self.cfg_logger_name.items():
 153             config[k1]["logger_name"] = logname
 154             logger_module = logging.getLogger(logname)
 155             if config[k1].get("logfile"):
 156                 file_handler = logging.handlers.RotatingFileHandler(
 157                     config[k1]["logfile"], maxBytes=100e6, backupCount=9, delay=0
 158                 )
 159                 file_handler.setFormatter(log_formatter_simple)
 160                 logger_module.addHandler(file_handler)
 161             if config[k1].get("loglevel"):
 162                 logger_module.setLevel(config[k1]["loglevel"])
 163         self.logger.critical(
 164             "starting osm/lcm version {} {}".format(lcm_version, lcm_version_date)
 165         )
 166
 167         # check version of N2VC
 168         # TODO enhance with int conversion or from distutils.version import LooseVersion
 169         # or with list(map(int, version.split(".")))
 170         if versiontuple(n2vc_version) < versiontuple(min_n2vc_version):
 171             raise LcmException(
 172                 "Not compatible osm/N2VC version '{}'. Needed '{}' or higher".format(
 173                     n2vc_version, min_n2vc_version
 174                 )
 175             )
 176         # check version of common
 177         if versiontuple(common_version) < versiontuple(min_common_version):
 178             raise LcmException(
 179                 "Not compatible osm/common version '{}'. Needed '{}' or higher".format(
 180                     common_version, min_common_version
 181                 )
 182             )
 183
 184         try:
 185             self.db = Database(config).instance.db
 186
 187             self.fs = Filesystem(config).instance.fs
 188             self.fs.sync()
 189
 190             # copy message configuration in order to remove 'group_id' for msg_admin
 191             config_message = config["message"].copy()
 192             config_message["loop"] = self.loop
 193             if config_message["driver"] == "local":
 194                 self.msg = msglocal.MsgLocal()
 195                 self.msg.connect(config_message)
 196                 self.msg_admin = msglocal.MsgLocal()
 197                 config_message.pop("group_id", None)
 198                 self.msg_admin.connect(config_message)
 199             elif config_message["driver"] == "kafka":
 200                 self.msg = msgkafka.MsgKafka()
 201                 self.msg.connect(config_message)
 202                 self.msg_admin = msgkafka.MsgKafka()
 203                 config_message.pop("group_id", None)
 204                 self.msg_admin.connect(config_message)
 205             else:
 206                 raise LcmException(
 207                     "Invalid configuration param '{}' at '[message]':'driver'".format(
 208                         config["message"]["driver"]
 209                     )
 210                 )
 211         except (DbException, FsException, MsgException) as e:
 212             self.logger.critical(str(e), exc_info=True)
 213             raise LcmException(str(e))
 214
 215         # contains created tasks/futures to be able to cancel
 216         self.lcm_tasks = TaskRegistry(self.worker_id, self.logger)
 217
 218     async def check_RO_version(self):
 219         tries = 14
 220         last_error = None
 221         while True:
 222             ro_uri = self.config["ro_config"]["uri"]
 223             try:
 224                 # try new  RO, if fail old RO
 225                 try:
 226                     self.config["ro_config"]["uri"] = ro_uri + "ro"
 227                     ro_server = NgRoClient(self.loop, **self.config["ro_config"])
 228                     ro_version = await ro_server.get_version()
 229                     self.config["ro_config"]["ng"] = True
 230                 except Exception:
 231                     self.config["ro_config"]["uri"] = ro_uri + "openmano"
 232                     ro_server = ROClient(self.loop, **self.config["ro_config"])
 233                     ro_version = await ro_server.get_version()
 234                     self.config["ro_config"]["ng"] = False
 235                 if versiontuple(ro_version) < versiontuple(min_RO_version):
 236                     raise LcmException(
 237                         "Not compatible osm/RO version '{}'. Needed '{}' or higher".format(
 238                             ro_version, min_RO_version
 239                         )
 240                     )
 241                 self.logger.info(
 242                     "Connected to RO version {} new-generation version {}".format(
 243                         ro_version, self.config["ro_config"]["ng"]
 244                     )
 245                 )
 246                 return
 247             except (ROClientException, NgRoException) as e:
 248                 self.config["ro_config"]["uri"] = ro_uri
 249                 tries -= 1
 250                 traceback.print_tb(e.__traceback__)
 251                 error_text = "Error while connecting to RO on {}: {}".format(
 252                     self.config["ro_config"]["uri"], e
 253                 )
 254                 if tries <= 0:
 255                     self.logger.critical(error_text)
 256                     raise LcmException(error_text)
 257                 if last_error != error_text:
 258                     last_error = error_text
 259                     self.logger.error(
 260                         error_text + ". Waiting until {} seconds".format(5 * tries)
 261                     )
 262                 await asyncio.sleep(5)
 263
 264     async def test(self, param=None):
 265         self.logger.debug("Starting/Ending test task: {}".format(param))
 266
 267     async def kafka_ping(self):
 268         self.logger.debug("Task kafka_ping Enter")
 269         consecutive_errors = 0
 270         first_start = True
 271         kafka_has_received = False
 272         self.pings_not_received = 1
 273         while True:
 274             try:
 275                 await self.msg_admin.aiowrite(
 276                     "admin",
 277                     "ping",
 278                     {
 279                         "from": "lcm",
 280                         "to": "lcm",
 281                         "worker_id": self.worker_id,
 282                         "version": lcm_version,
 283                     },
 284                     self.loop,
 285                 )
 286                 # time between pings are low when it is not received and at starting
 287                 wait_time = (
 288                     self.ping_interval_boot
 289                     if not kafka_has_received
 290                     else self.ping_interval_pace
 291                 )
 292                 if not self.pings_not_received:
 293                     kafka_has_received = True
 294                 self.pings_not_received += 1
 295                 await asyncio.sleep(wait_time, loop=self.loop)
 296                 if self.pings_not_received > 10:
 297                     raise LcmException("It is not receiving pings from Kafka bus")
 298                 consecutive_errors = 0
 299                 first_start = False
 300             except LcmException:
 301                 raise
 302             except Exception as e:
 303                 # if not first_start is the first time after starting. So leave more time and wait
 304                 # to allow kafka starts
 305                 if consecutive_errors == 8 if not first_start else 30:
 306                     self.logger.error(
 307                         "Task kafka_read task exit error too many errors. Exception: {}".format(
 308                             e
 309                         )
 310                     )
 311                     raise
 312                 consecutive_errors += 1
 313                 self.logger.error(
 314                     "Task kafka_read retrying after Exception {}".format(e)
 315                 )
 316                 wait_time = 2 if not first_start else 5
 317                 await asyncio.sleep(wait_time, loop=self.loop)
 318
 319     def kafka_read_callback(self, topic, command, params):
 320         order_id = 1
 321
 322         if topic != "admin" and command != "ping":
 323             self.logger.debug(
 324                 "Task kafka_read receives {} {}: {}".format(topic, command, params)
 325             )
 326         self.consecutive_errors = 0
 327         self.first_start = False
 328         order_id += 1
 329         if command == "exit":
 330             raise LcmExceptionExit
 331         elif command.startswith("#"):
 332             return
 333         elif command == "echo":
 334             # just for test
 335             print(params)
 336             sys.stdout.flush()
 337             return
 338         elif command == "test":
 339             asyncio.Task(self.test(params), loop=self.loop)
 340             return
 341
 342         if topic == "admin":
 343             if command == "ping" and params["to"] == "lcm" and params["from"] == "lcm":
 344                 if params.get("worker_id") != self.worker_id:
 345                     return
 346                 self.pings_not_received = 0
 347                 try:
 348                     with open(health_check_file, "w") as f:
 349                         f.write(str(time()))
 350                 except Exception as e:
 351                     self.logger.error(
 352                         "Cannot write into '{}' for healthcheck: {}".format(
 353                             health_check_file, e
 354                         )
 355                     )
 356             return
 357         elif topic == "pla":
 358             if command == "placement":
 359                 self.ns.update_nsrs_with_pla_result(params)
 360             return
 361         elif topic == "k8scluster":
 362             if command == "create" or command == "created":
 363                 k8scluster_id = params.get("_id")
 364                 task = asyncio.ensure_future(self.k8scluster.create(params, order_id))
 365                 self.lcm_tasks.register(
 366                     "k8scluster", k8scluster_id, order_id, "k8scluster_create", task
 367                 )
 368                 return
 369             elif command == "delete" or command == "deleted":
 370                 k8scluster_id = params.get("_id")
 371                 task = asyncio.ensure_future(self.k8scluster.delete(params, order_id))
 372                 self.lcm_tasks.register(
 373                     "k8scluster", k8scluster_id, order_id, "k8scluster_delete", task
 374                 )
 375                 return
 376         elif topic == "vca":
 377             if command == "create" or command == "created":
 378                 vca_id = params.get("_id")
 379                 task = asyncio.ensure_future(self.vca.create(params, order_id))
 380                 self.lcm_tasks.register("vca", vca_id, order_id, "vca_create", task)
 381                 return
 382             elif command == "delete" or command == "deleted":
 383                 vca_id = params.get("_id")
 384                 task = asyncio.ensure_future(self.vca.delete(params, order_id))
 385                 self.lcm_tasks.register("vca", vca_id, order_id, "vca_delete", task)
 386                 return
 387         elif topic == "k8srepo":
 388             if command == "create" or command == "created":
 389                 k8srepo_id = params.get("_id")
 390                 self.logger.debug("k8srepo_id = {}".format(k8srepo_id))
 391                 task = asyncio.ensure_future(self.k8srepo.create(params, order_id))
 392                 self.lcm_tasks.register(
 393                     "k8srepo", k8srepo_id, order_id, "k8srepo_create", task
 394                 )
 395                 return
 396             elif command == "delete" or command == "deleted":
 397                 k8srepo_id = params.get("_id")
 398                 task = asyncio.ensure_future(self.k8srepo.delete(params, order_id))
 399                 self.lcm_tasks.register(
 400                     "k8srepo", k8srepo_id, order_id, "k8srepo_delete", task
 401                 )
 402                 return
 403         elif topic == "ns":
 404             if command == "instantiate":
 405                 # self.logger.debug("Deploying NS {}".format(nsr_id))
 406                 nslcmop = params
 407                 nslcmop_id = nslcmop["_id"]
 408                 nsr_id = nslcmop["nsInstanceId"]
 409                 task = asyncio.ensure_future(self.ns.instantiate(nsr_id, nslcmop_id))
 410                 self.lcm_tasks.register(
 411                     "ns", nsr_id, nslcmop_id, "ns_instantiate", task
 412                 )
 413                 return
 414             elif command == "terminate":
 415                 # self.logger.debug("Deleting NS {}".format(nsr_id))
 416                 nslcmop = params
 417                 nslcmop_id = nslcmop["_id"]
 418                 nsr_id = nslcmop["nsInstanceId"]
 419                 self.lcm_tasks.cancel(topic, nsr_id)
 420                 task = asyncio.ensure_future(self.ns.terminate(nsr_id, nslcmop_id))
 421                 self.lcm_tasks.register("ns", nsr_id, nslcmop_id, "ns_terminate", task)
 422                 return
 423             elif command == "vca_status_refresh":
 424                 nslcmop = params
 425                 nslcmop_id = nslcmop["_id"]
 426                 nsr_id = nslcmop["nsInstanceId"]
 427                 task = asyncio.ensure_future(
 428                     self.ns.vca_status_refresh(nsr_id, nslcmop_id)
 429                 )
 430                 self.lcm_tasks.register(
 431                     "ns", nsr_id, nslcmop_id, "ns_vca_status_refresh", task
 432                 )
 433                 return
 434             elif command == "action":
 435                 # self.logger.debug("Update NS {}".format(nsr_id))
 436                 nslcmop = params
 437                 nslcmop_id = nslcmop["_id"]
 438                 nsr_id = nslcmop["nsInstanceId"]
 439                 task = asyncio.ensure_future(self.ns.action(nsr_id, nslcmop_id))
 440                 self.lcm_tasks.register("ns", nsr_id, nslcmop_id, "ns_action", task)
 441                 return
 442             elif command == "update":
 443                 # self.logger.debug("Update NS {}".format(nsr_id))
 444                 nslcmop = params
 445                 nslcmop_id = nslcmop["_id"]
 446                 nsr_id = nslcmop["nsInstanceId"]
 447                 task = asyncio.ensure_future(self.ns.update(nsr_id, nslcmop_id))
 448                 self.lcm_tasks.register("ns", nsr_id, nslcmop_id, "ns_update", task)
 449                 return
 450             elif command == "scale":
 451                 # self.logger.debug("Update NS {}".format(nsr_id))
 452                 nslcmop = params
 453                 nslcmop_id = nslcmop["_id"]
 454                 nsr_id = nslcmop["nsInstanceId"]
 455                 task = asyncio.ensure_future(self.ns.scale(nsr_id, nslcmop_id))
 456                 self.lcm_tasks.register("ns", nsr_id, nslcmop_id, "ns_scale", task)
 457                 return
 458             elif command == "migrate":
 459                 nslcmop = params
 460                 nslcmop_id = nslcmop["_id"]
 461                 nsr_id = nslcmop["nsInstanceId"]
 462                 task = asyncio.ensure_future(self.ns.migrate(nsr_id, nslcmop_id))
 463                 self.lcm_tasks.register("ns", nsr_id, nslcmop_id, "ns_migrate", task)
 464                 return
 465             elif command == "show":
 466                 nsr_id = params
 467                 try:
 468                     db_nsr = self.db.get_one("nsrs", {"_id": nsr_id})
 469                     print(
 470                         "nsr:\n    _id={}\n    operational-status: {}\n    config-status: {}"
 471                         "\n    detailed-status: {}\n    deploy: {}\n    tasks: {}"
 472                         "".format(
 473                             nsr_id,
 474                             db_nsr["operational-status"],
 475                             db_nsr["config-status"],
 476                             db_nsr["detailed-status"],
 477                             db_nsr["_admin"]["deployed"],
 478                             self.lcm_ns_tasks.get(nsr_id),
 479                         )
 480                     )
 481                 except Exception as e:
 482                     print("nsr {} not found: {}".format(nsr_id, e))
 483                 sys.stdout.flush()
 484                 return
 485             elif command == "deleted":
 486                 return  # TODO cleaning of task just in case should be done
 487             elif command in (
 488                 "vnf_terminated",
 489                 "policy_updated",
 490                 "terminated",
 491                 "instantiated",
 492                 "scaled",
 493                 "actioned",
 494                 "updated",
 495                 "migrated",
 496             ):  # "scaled-cooldown-time"
 497                 return
 498
 499         elif topic == "nsi":  # netslice LCM processes (instantiate, terminate, etc)
 500             if command == "instantiate":
 501                 # self.logger.debug("Instantiating Network Slice {}".format(nsilcmop["netsliceInstanceId"]))
 502                 nsilcmop = params
 503                 nsilcmop_id = nsilcmop["_id"]  # slice operation id
 504                 nsir_id = nsilcmop["netsliceInstanceId"]  # slice record id
 505                 task = asyncio.ensure_future(
 506                     self.netslice.instantiate(nsir_id, nsilcmop_id)
 507                 )
 508                 self.lcm_tasks.register(
 509                     "nsi", nsir_id, nsilcmop_id, "nsi_instantiate", task
 510                 )
 511                 return
 512             elif command == "terminate":
 513                 # self.logger.debug("Terminating Network Slice NS {}".format(nsilcmop["netsliceInstanceId"]))
 514                 nsilcmop = params
 515                 nsilcmop_id = nsilcmop["_id"]  # slice operation id
 516                 nsir_id = nsilcmop["netsliceInstanceId"]  # slice record id
 517                 self.lcm_tasks.cancel(topic, nsir_id)
 518                 task = asyncio.ensure_future(
 519                     self.netslice.terminate(nsir_id, nsilcmop_id)
 520                 )
 521                 self.lcm_tasks.register(
 522                     "nsi", nsir_id, nsilcmop_id, "nsi_terminate", task
 523                 )
 524                 return
 525             elif command == "show":
 526                 nsir_id = params
 527                 try:
 528                     db_nsir = self.db.get_one("nsirs", {"_id": nsir_id})
 529                     print(
 530                         "nsir:\n    _id={}\n    operational-status: {}\n    config-status: {}"
 531                         "\n    detailed-status: {}\n    deploy: {}\n    tasks: {}"
 532                         "".format(
 533                             nsir_id,
 534                             db_nsir["operational-status"],
 535                             db_nsir["config-status"],
 536                             db_nsir["detailed-status"],
 537                             db_nsir["_admin"]["deployed"],
 538                             self.lcm_netslice_tasks.get(nsir_id),
 539                         )
 540                     )
 541                 except Exception as e:
 542                     print("nsir {} not found: {}".format(nsir_id, e))
 543                 sys.stdout.flush()
 544                 return
 545             elif command == "deleted":
 546                 return  # TODO cleaning of task just in case should be done
 547             elif command in (
 548                 "terminated",
 549                 "instantiated",
 550                 "scaled",
 551                 "actioned",
 552             ):  # "scaled-cooldown-time"
 553                 return
 554         elif topic == "vim_account":
 555             vim_id = params["_id"]
 556             if command in ("create", "created"):
 557                 if not self.config["ro_config"].get("ng"):
 558                     task = asyncio.ensure_future(self.vim.create(params, order_id))
 559                     self.lcm_tasks.register(
 560                         "vim_account", vim_id, order_id, "vim_create", task
 561                     )
 562                 return
 563             elif command == "delete" or command == "deleted":
 564                 self.lcm_tasks.cancel(topic, vim_id)
 565                 task = asyncio.ensure_future(self.vim.delete(params, order_id))
 566                 self.lcm_tasks.register(
 567                     "vim_account", vim_id, order_id, "vim_delete", task
 568                 )
 569                 return
 570             elif command == "show":
 571                 print("not implemented show with vim_account")
 572                 sys.stdout.flush()
 573                 return
 574             elif command in ("edit", "edited"):
 575                 if not self.config["ro_config"].get("ng"):
 576                     task = asyncio.ensure_future(self.vim.edit(params, order_id))
 577                     self.lcm_tasks.register(
 578                         "vim_account", vim_id, order_id, "vim_edit", task
 579                     )
 580                 return
 581             elif command == "deleted":
 582                 return  # TODO cleaning of task just in case should be done
 583         elif topic == "wim_account":
 584             wim_id = params["_id"]
 585             if command in ("create", "created"):
 586                 if not self.config["ro_config"].get("ng"):
 587                     task = asyncio.ensure_future(self.wim.create(params, order_id))
 588                     self.lcm_tasks.register(
 589                         "wim_account", wim_id, order_id, "wim_create", task
 590                     )
 591                 return
 592             elif command == "delete" or command == "deleted":
 593                 self.lcm_tasks.cancel(topic, wim_id)
 594                 task = asyncio.ensure_future(self.wim.delete(params, order_id))
 595                 self.lcm_tasks.register(
 596                     "wim_account", wim_id, order_id, "wim_delete", task
 597                 )
 598                 return
 599             elif command == "show":
 600                 print("not implemented show with wim_account")
 601                 sys.stdout.flush()
 602                 return
 603             elif command in ("edit", "edited"):
 604                 task = asyncio.ensure_future(self.wim.edit(params, order_id))
 605                 self.lcm_tasks.register(
 606                     "wim_account", wim_id, order_id, "wim_edit", task
 607                 )
 608                 return
 609             elif command == "deleted":
 610                 return  # TODO cleaning of task just in case should be done
 611         elif topic == "sdn":
 612             _sdn_id = params["_id"]
 613             if command in ("create", "created"):
 614                 if not self.config["ro_config"].get("ng"):
 615                     task = asyncio.ensure_future(self.sdn.create(params, order_id))
 616                     self.lcm_tasks.register(
 617                         "sdn", _sdn_id, order_id, "sdn_create", task
 618                     )
 619                 return
 620             elif command == "delete" or command == "deleted":
 621                 self.lcm_tasks.cancel(topic, _sdn_id)
 622                 task = asyncio.ensure_future(self.sdn.delete(params, order_id))
 623                 self.lcm_tasks.register("sdn", _sdn_id, order_id, "sdn_delete", task)
 624                 return
 625             elif command in ("edit", "edited"):
 626                 task = asyncio.ensure_future(self.sdn.edit(params, order_id))
 627                 self.lcm_tasks.register("sdn", _sdn_id, order_id, "sdn_edit", task)
 628                 return
 629             elif command == "deleted":
 630                 return  # TODO cleaning of task just in case should be done
 631         self.logger.critical("unknown topic {} and command '{}'".format(topic, command))
 632
 633     async def kafka_read(self):
 634         self.logger.debug(
 635             "Task kafka_read Enter with worker_id={}".format(self.worker_id)
 636         )
 637         # future = asyncio.Future()
 638         self.consecutive_errors = 0
 639         self.first_start = True
 640         while self.consecutive_errors < 10:
 641             try:
 642                 topics = (
 643                     "ns",
 644                     "vim_account",
 645                     "wim_account",
 646                     "sdn",
 647                     "nsi",
 648                     "k8scluster",
 649                     "vca",
 650                     "k8srepo",
 651                     "pla",
 652                 )
 653                 topics_admin = ("admin",)
 654                 await asyncio.gather(
 655                     self.msg.aioread(
 656                         topics, self.loop, self.kafka_read_callback, from_beginning=True
 657                     ),
 658                     self.msg_admin.aioread(
 659                         topics_admin,
 660                         self.loop,
 661                         self.kafka_read_callback,
 662                         group_id=False,
 663                     ),
 664                 )
 665
 666             except LcmExceptionExit:
 667                 self.logger.debug("Bye!")
 668                 break
 669             except Exception as e:
 670                 # if not first_start is the first time after starting. So leave more time and wait
 671                 # to allow kafka starts
 672                 if self.consecutive_errors == 8 if not self.first_start else 30:
 673                     self.logger.error(
 674                         "Task kafka_read task exit error too many errors. Exception: {}".format(
 675                             e
 676                         )
 677                     )
 678                     raise
 679                 self.consecutive_errors += 1
 680                 self.logger.error(
 681                     "Task kafka_read retrying after Exception {}".format(e)
 682                 )
 683                 wait_time = 2 if not self.first_start else 5
 684                 await asyncio.sleep(wait_time, loop=self.loop)
 685
 686         # self.logger.debug("Task kafka_read terminating")
 687         self.logger.debug("Task kafka_read exit")
 688
 689     def start(self):
 690
 691         # check RO version
 692         self.loop.run_until_complete(self.check_RO_version())
 693
 694         self.ns = ns.NsLcm(self.msg, self.lcm_tasks, self.config, self.loop)
 695         self.netslice = netslice.NetsliceLcm(
 696             self.msg, self.lcm_tasks, self.config, self.loop, self.ns
 697         )
 698         self.vim = vim_sdn.VimLcm(self.msg, self.lcm_tasks, self.config, self.loop)
 699         self.wim = vim_sdn.WimLcm(self.msg, self.lcm_tasks, self.config, self.loop)
 700         self.sdn = vim_sdn.SdnLcm(self.msg, self.lcm_tasks, self.config, self.loop)
 701         self.k8scluster = vim_sdn.K8sClusterLcm(
 702             self.msg, self.lcm_tasks, self.config, self.loop
 703         )
 704         self.vca = vim_sdn.VcaLcm(self.msg, self.lcm_tasks, self.config, self.loop)
 705         self.k8srepo = vim_sdn.K8sRepoLcm(
 706             self.msg, self.lcm_tasks, self.config, self.loop
 707         )
 708
 709         self.loop.run_until_complete(
 710             asyncio.gather(self.kafka_read(), self.kafka_ping())
 711         )
 712
 713         # TODO
 714         # self.logger.debug("Terminating cancelling creation tasks")
 715         # self.lcm_tasks.cancel("ALL", "create")
 716         # timeout = 200
 717         # while self.is_pending_tasks():
 718         #     self.logger.debug("Task kafka_read terminating. Waiting for tasks termination")
 719         #     await asyncio.sleep(2, loop=self.loop)
 720         #     timeout -= 2
 721         #     if not timeout:
 722         #         self.lcm_tasks.cancel("ALL", "ALL")
 723         self.loop.close()
 724         self.loop = None
 725         if self.db:
 726             self.db.db_disconnect()
 727         if self.msg:
 728             self.msg.disconnect()
 729         if self.msg_admin:
 730             self.msg_admin.disconnect()
 731         if self.fs:
 732             self.fs.fs_disconnect()
 733
 734     def read_config_file(self, config_file):
 735         # TODO make a [ini] + yaml inside parser
 736         # the configparser library is not suitable, because it does not admit comments at the end of line,
 737         # and not parse integer or boolean
 738         try:
 739             # read file as yaml format
 740             with open(config_file) as f:
 741                 conf = yaml.load(f, Loader=yaml.Loader)
 742             # Ensure all sections are not empty
 743             for k in (
 744                 "global",
 745                 "timeout",
 746                 "RO",
 747                 "VCA",
 748                 "database",
 749                 "storage",
 750                 "message",
 751             ):
 752                 if not conf.get(k):
 753                     conf[k] = {}
 754
 755             # read all environ that starts with OSMLCM_
 756             for k, v in environ.items():
 757                 if not k.startswith("OSMLCM_"):
 758                     continue
 759                 subject, _, item = k[7:].lower().partition("_")
 760                 if not item:
 761                     continue
 762                 if subject in ("ro", "vca"):
 763                     # put in capital letter
 764                     subject = subject.upper()
 765                 try:
 766                     if item == "port" or subject == "timeout":
 767                         conf[subject][item] = int(v)
 768                     else:
 769                         conf[subject][item] = v
 770                 except Exception as e:
 771                     self.logger.warning(
 772                         "skipping environ '{}' on exception '{}'".format(k, e)
 773                     )
 774
 775             # backward compatibility of VCA parameters
 776
 777             if "pubkey" in conf["VCA"]:
 778                 conf["VCA"]["public_key"] = conf["VCA"].pop("pubkey")
 779             if "cacert" in conf["VCA"]:
 780                 conf["VCA"]["ca_cert"] = conf["VCA"].pop("cacert")
 781             if "apiproxy" in conf["VCA"]:
 782                 conf["VCA"]["api_proxy"] = conf["VCA"].pop("apiproxy")
 783
 784             if "enableosupgrade" in conf["VCA"]:
 785                 conf["VCA"]["enable_os_upgrade"] = conf["VCA"].pop("enableosupgrade")
 786             if isinstance(conf["VCA"].get("enable_os_upgrade"), str):
 787                 if conf["VCA"]["enable_os_upgrade"].lower() == "false":
 788                     conf["VCA"]["enable_os_upgrade"] = False
 789                 elif conf["VCA"]["enable_os_upgrade"].lower() == "true":
 790                     conf["VCA"]["enable_os_upgrade"] = True
 791
 792             if "aptmirror" in conf["VCA"]:
 793                 conf["VCA"]["apt_mirror"] = conf["VCA"].pop("aptmirror")
 794
 795             return conf
 796         except Exception as e:
 797             self.logger.critical("At config file '{}': {}".format(config_file, e))
 798             exit(1)
 799
 800     @staticmethod
 801     def get_process_id():
 802         """
 803         Obtain a unique ID for this process. If running from inside docker, it will get docker ID. If not it
 804         will provide a random one
 805         :return: Obtained ID
 806         """
 807         # Try getting docker id. If fails, get pid
 808         try:
 809             with open("/proc/self/cgroup", "r") as f:
 810                 text_id_ = f.readline()
 811                 _, _, text_id = text_id_.rpartition("/")
 812                 text_id = text_id.replace("\n", "")[:12]
 813                 if text_id:
 814                     return text_id
 815         except Exception:
 816             pass
 817         # Return a random id
 818         return "".join(random_choice("0123456789abcdef") for _ in range(12))
 819
 820
 821 def usage():
 822     print(
 823         """Usage: {} [options]
 824         -c|--config [configuration_file]: loads the configuration file (default: ./lcm.cfg)
 825         --health-check: do not run lcm, but inspect kafka bus to determine if lcm is healthy
 826         -h|--help: shows this help
 827         """.format(
 828             sys.argv[0]
 829         )
 830     )
 831     # --log-socket-host HOST: send logs to this host")
 832     # --log-socket-port PORT: send logs using this port (default: 9022)")
 833
 834
 835 if __name__ == "__main__":
 836
 837     try:
 838         # print("SYS.PATH='{}'".format(sys.path))
 839         # load parameters and configuration
 840         # -h
 841         # -c value
 842         # --config value
 843         # --help
 844         # --health-check
 845         opts, args = getopt.getopt(
 846             sys.argv[1:], "hc:", ["config=", "help", "health-check"]
 847         )
 848         # TODO add  "log-socket-host=", "log-socket-port=", "log-file="
 849         config_file = None
 850         for o, a in opts:
 851             if o in ("-h", "--help"):
 852                 usage()
 853                 sys.exit()
 854             elif o in ("-c", "--config"):
 855                 config_file = a
 856             elif o == "--health-check":
 857                 from osm_lcm.lcm_hc import health_check
 858
 859                 health_check(health_check_file, Lcm.ping_interval_pace)
 860             # elif o == "--log-socket-port":
 861             #     log_socket_port = a
 862             # elif o == "--log-socket-host":
 863             #     log_socket_host = a
 864             # elif o == "--log-file":
 865             #     log_file = a
 866             else:
 867                 assert False, "Unhandled option"
 868
 869         if config_file:
 870             if not path.isfile(config_file):
 871                 print(
 872                     "configuration file '{}' does not exist".format(config_file),
 873                     file=sys.stderr,
 874                 )
 875                 exit(1)
 876         else:
 877             for config_file in (
 878                 __file__[: __file__.rfind(".")] + ".cfg",
 879                 "./lcm.cfg",
 880                 "/etc/osm/lcm.cfg",
 881             ):
 882                 if path.isfile(config_file):
 883                     break
 884             else:
 885                 print(
 886                     "No configuration file 'lcm.cfg' found neither at local folder nor at /etc/osm/",
 887                     file=sys.stderr,
 888                 )
 889                 exit(1)
 890         lcm = Lcm(config_file)
 891         lcm.start()
 892     except (LcmException, getopt.GetoptError) as e:
 893         print(str(e), file=sys.stderr)
 894         # usage()
 895         exit(1)