osm_lcm/lcm.py

   1 #!/usr/bin/python3
   2 # -*- coding: utf-8 -*-
   3
   4 import asyncio
   5 import yaml
   6 import logging
   7 import logging.handlers
   8 import getopt
   9 import sys
  10 import ROclient
  11 import ns
  12 import vim_sdn
  13 from lcm_utils import versiontuple, LcmException, TaskRegistry
  14
  15 # from osm_lcm import version as lcm_version, version_date as lcm_version_date, ROclient
  16 from osm_common import dbmemory, dbmongo, fslocal, msglocal, msgkafka
  17 from osm_common import version as common_version
  18 from osm_common.dbbase import DbException
  19 from osm_common.fsbase import FsException
  20 from osm_common.msgbase import MsgException
  21 from os import environ, path
  22 from n2vc import version as n2vc_version
  23
  24
  25 __author__ = "Alfonso Tierno"
  26 min_RO_version = [0, 5, 72]
  27 min_n2vc_version = "0.0.2"
  28 min_common_version = "0.1.11"
  29 # uncomment if LCM is installed as library and installed, and get them from __init__.py
  30 lcm_version = '0.1.23'
  31 lcm_version_date = '2018-11-13'
  32
  33
  34 class Lcm:
  35
  36     ping_interval_pace = 120  # how many time ping is send once is confirmed all is running
  37     ping_interval_boot = 5    # how many time ping is sent when booting
  38
  39     def __init__(self, config_file, loop=None):
  40         """
  41         Init, Connect to database, filesystem storage, and messaging
  42         :param config: two level dictionary with configuration. Top level should contain 'database', 'storage',
  43         :return: None
  44         """
  45
  46         self.db = None
  47         self.msg = None
  48         self.fs = None
  49         self.pings_not_received = 1
  50
  51         # contains created tasks/futures to be able to cancel
  52         self.lcm_tasks = TaskRegistry()
  53         # logging
  54         self.logger = logging.getLogger('lcm')
  55         # load configuration
  56         config = self.read_config_file(config_file)
  57         self.config = config
  58         self.ro_config = {
  59             "endpoint_url": "http://{}:{}/openmano".format(config["RO"]["host"], config["RO"]["port"]),
  60             "tenant": config.get("tenant", "osm"),
  61             "logger_name": "lcm.ROclient",
  62             "loglevel": "ERROR",
  63         }
  64
  65         self.vca_config = config["VCA"]
  66
  67         self.loop = loop or asyncio.get_event_loop()
  68
  69         # logging
  70         log_format_simple = "%(asctime)s %(levelname)s %(name)s %(filename)s:%(lineno)s %(message)s"
  71         log_formatter_simple = logging.Formatter(log_format_simple, datefmt='%Y-%m-%dT%H:%M:%S')
  72         config["database"]["logger_name"] = "lcm.db"
  73         config["storage"]["logger_name"] = "lcm.fs"
  74         config["message"]["logger_name"] = "lcm.msg"
  75         if config["global"].get("logfile"):
  76             file_handler = logging.handlers.RotatingFileHandler(config["global"]["logfile"],
  77                                                                 maxBytes=100e6, backupCount=9, delay=0)
  78             file_handler.setFormatter(log_formatter_simple)
  79             self.logger.addHandler(file_handler)
  80         if not config["global"].get("nologging"):
  81             str_handler = logging.StreamHandler()
  82             str_handler.setFormatter(log_formatter_simple)
  83             self.logger.addHandler(str_handler)
  84
  85         if config["global"].get("loglevel"):
  86             self.logger.setLevel(config["global"]["loglevel"])
  87
  88         # logging other modules
  89         for k1, logname in {"message": "lcm.msg", "database": "lcm.db", "storage": "lcm.fs"}.items():
  90             config[k1]["logger_name"] = logname
  91             logger_module = logging.getLogger(logname)
  92             if config[k1].get("logfile"):
  93                 file_handler = logging.handlers.RotatingFileHandler(config[k1]["logfile"],
  94                                                                     maxBytes=100e6, backupCount=9, delay=0)
  95                 file_handler.setFormatter(log_formatter_simple)
  96                 logger_module.addHandler(file_handler)
  97             if config[k1].get("loglevel"):
  98                 logger_module.setLevel(config[k1]["loglevel"])
  99         self.logger.critical("starting osm/lcm version {} {}".format(lcm_version, lcm_version_date))
 100
 101         # check version of N2VC
 102         # TODO enhance with int conversion or from distutils.version import LooseVersion
 103         # or with list(map(int, version.split(".")))
 104         if versiontuple(n2vc_version) < versiontuple(min_n2vc_version):
 105             raise LcmException("Not compatible osm/N2VC version '{}'. Needed '{}' or higher".format(
 106                 n2vc_version, min_n2vc_version))
 107         # check version of common
 108         if versiontuple(common_version) < versiontuple(min_common_version):
 109             raise LcmException("Not compatible osm/common version '{}'. Needed '{}' or higher".format(
 110                 common_version, min_common_version))
 111
 112         try:
 113             # TODO check database version
 114             if config["database"]["driver"] == "mongo":
 115                 self.db = dbmongo.DbMongo()
 116                 self.db.db_connect(config["database"])
 117             elif config["database"]["driver"] == "memory":
 118                 self.db = dbmemory.DbMemory()
 119                 self.db.db_connect(config["database"])
 120             else:
 121                 raise LcmException("Invalid configuration param '{}' at '[database]':'driver'".format(
 122                     config["database"]["driver"]))
 123
 124             if config["storage"]["driver"] == "local":
 125                 self.fs = fslocal.FsLocal()
 126                 self.fs.fs_connect(config["storage"])
 127             else:
 128                 raise LcmException("Invalid configuration param '{}' at '[storage]':'driver'".format(
 129                     config["storage"]["driver"]))
 130
 131             if config["message"]["driver"] == "local":
 132                 self.msg = msglocal.MsgLocal()
 133                 self.msg.connect(config["message"])
 134             elif config["message"]["driver"] == "kafka":
 135                 self.msg = msgkafka.MsgKafka()
 136                 self.msg.connect(config["message"])
 137             else:
 138                 raise LcmException("Invalid configuration param '{}' at '[message]':'driver'".format(
 139                     config["storage"]["driver"]))
 140         except (DbException, FsException, MsgException) as e:
 141             self.logger.critical(str(e), exc_info=True)
 142             raise LcmException(str(e))
 143
 144         self.ns = ns.NsLcm(self.db, self.msg, self.fs, self.lcm_tasks, self.ro_config, self.vca_config, self.loop)
 145         self.vim = vim_sdn.VimLcm(self.db, self.msg, self.fs, self.lcm_tasks, self.ro_config, self.loop)
 146         self.sdn = vim_sdn.SdnLcm(self.db, self.msg, self.fs, self.lcm_tasks, self.ro_config, self.loop)
 147
 148     async def check_RO_version(self):
 149         try:
 150             RO = ROclient.ROClient(self.loop, **self.ro_config)
 151             RO_version = await RO.get_version()
 152             if RO_version < min_RO_version:
 153                 raise LcmException("Not compatible osm/RO version '{}.{}.{}'. Needed '{}.{}.{}' or higher".format(
 154                     *RO_version, *min_RO_version
 155                 ))
 156         except ROclient.ROClientException as e:
 157             error_text = "Error while conneting to osm/RO " + str(e)
 158             self.logger.critical(error_text, exc_info=True)
 159             raise LcmException(error_text)
 160
 161     async def test(self, param=None):
 162         self.logger.debug("Starting/Ending test task: {}".format(param))
 163
 164     async def kafka_ping(self):
 165         self.logger.debug("Task kafka_ping Enter")
 166         consecutive_errors = 0
 167         first_start = True
 168         kafka_has_received = False
 169         self.pings_not_received = 1
 170         while True:
 171             try:
 172                 await self.msg.aiowrite("admin", "ping", {"from": "lcm", "to": "lcm"}, self.loop)
 173                 # time between pings are low when it is not received and at starting
 174                 wait_time = self.ping_interval_boot if not kafka_has_received else self.ping_interval_pace
 175                 if not self.pings_not_received:
 176                     kafka_has_received = True
 177                 self.pings_not_received += 1
 178                 await asyncio.sleep(wait_time, loop=self.loop)
 179                 if self.pings_not_received > 10:
 180                     raise LcmException("It is not receiving pings from Kafka bus")
 181                 consecutive_errors = 0
 182                 first_start = False
 183             except LcmException:
 184                 raise
 185             except Exception as e:
 186                 # if not first_start is the first time after starting. So leave more time and wait
 187                 # to allow kafka starts
 188                 if consecutive_errors == 8 if not first_start else 30:
 189                     self.logger.error("Task kafka_read task exit error too many errors. Exception: {}".format(e))
 190                     raise
 191                 consecutive_errors += 1
 192                 self.logger.error("Task kafka_read retrying after Exception {}".format(e))
 193                 wait_time = 1 if not first_start else 5
 194                 await asyncio.sleep(wait_time, loop=self.loop)
 195
 196     async def kafka_read(self):
 197         self.logger.debug("Task kafka_read Enter")
 198         order_id = 1
 199         # future = asyncio.Future()
 200         consecutive_errors = 0
 201         first_start = True
 202         while consecutive_errors < 10:
 203             try:
 204                 topics = ("admin", "ns", "vim_account", "sdn")
 205                 topic, command, params = await self.msg.aioread(topics, self.loop)
 206                 if topic != "admin" and command != "ping":
 207                     self.logger.debug("Task kafka_read receives {} {}: {}".format(topic, command, params))
 208                 consecutive_errors = 0
 209                 first_start = False
 210                 order_id += 1
 211                 if command == "exit":
 212                     print("Bye!")
 213                     break
 214                 elif command.startswith("#"):
 215                     continue
 216                 elif command == "echo":
 217                     # just for test
 218                     print(params)
 219                     sys.stdout.flush()
 220                     continue
 221                 elif command == "test":
 222                     asyncio.Task(self.test(params), loop=self.loop)
 223                     continue
 224
 225                 if topic == "admin":
 226                     if command == "ping" and params["to"] == "lcm" and params["from"] == "lcm":
 227                         self.pings_not_received = 0
 228                     continue
 229                 elif topic == "ns":
 230                     if command == "instantiate":
 231                         # self.logger.debug("Deploying NS {}".format(nsr_id))
 232                         nslcmop = params
 233                         nslcmop_id = nslcmop["_id"]
 234                         nsr_id = nslcmop["nsInstanceId"]
 235                         task = asyncio.ensure_future(self.ns.instantiate(nsr_id, nslcmop_id))
 236                         self.lcm_tasks.register("ns", nsr_id, nslcmop_id, "ns_instantiate", task)
 237                         continue
 238                     elif command == "terminate":
 239                         # self.logger.debug("Deleting NS {}".format(nsr_id))
 240                         nslcmop = params
 241                         nslcmop_id = nslcmop["_id"]
 242                         nsr_id = nslcmop["nsInstanceId"]
 243                         self.lcm_tasks.cancel(topic, nsr_id)
 244                         task = asyncio.ensure_future(self.ns.terminate(nsr_id, nslcmop_id))
 245                         self.lcm_tasks.register("ns", nsr_id, nslcmop_id, "ns_terminate", task)
 246                         continue
 247                     elif command == "action":
 248                         # self.logger.debug("Update NS {}".format(nsr_id))
 249                         nslcmop = params
 250                         nslcmop_id = nslcmop["_id"]
 251                         nsr_id = nslcmop["nsInstanceId"]
 252                         task = asyncio.ensure_future(self.ns.action(nsr_id, nslcmop_id))
 253                         self.lcm_tasks.register("ns", nsr_id, nslcmop_id, "ns_action", task)
 254                         continue
 255                     elif command == "scale":
 256                         # self.logger.debug("Update NS {}".format(nsr_id))
 257                         nslcmop = params
 258                         nslcmop_id = nslcmop["_id"]
 259                         nsr_id = nslcmop["nsInstanceId"]
 260                         task = asyncio.ensure_future(self.ns.scale(nsr_id, nslcmop_id))
 261                         self.lcm_tasks.register("ns", nsr_id, nslcmop_id, "ns_scale", task)
 262                         continue
 263                     elif command == "show":
 264                         try:
 265                             db_nsr = self.db.get_one("nsrs", {"_id": nsr_id})
 266                             print("nsr:\n    _id={}\n    operational-status: {}\n    config-status: {}"
 267                                   "\n    detailed-status: {}\n    deploy: {}\n    tasks: {}"
 268                                   "".format(nsr_id, db_nsr["operational-status"], db_nsr["config-status"],
 269                                             db_nsr["detailed-status"],
 270                                             db_nsr["_admin"]["deployed"], self.lcm_ns_tasks.get(nsr_id)))
 271                         except Exception as e:
 272                             print("nsr {} not found: {}".format(nsr_id, e))
 273                         sys.stdout.flush()
 274                         continue
 275                     elif command == "deleted":
 276                         continue  # TODO cleaning of task just in case should be done
 277                     elif command in ("terminated", "instantiated", "scaled", "actioned"):  # "scaled-cooldown-time"
 278                         continue
 279                 elif topic == "vim_account":
 280                     vim_id = params["_id"]
 281                     if command == "create":
 282                         task = asyncio.ensure_future(self.vim.create(params, order_id))
 283                         self.lcm_tasks.register("vim_account", vim_id, order_id, "vim_create", task)
 284                         continue
 285                     elif command == "delete":
 286                         self.lcm_tasks.cancel(topic, vim_id)
 287                         task = asyncio.ensure_future(self.vim.delete(vim_id, order_id))
 288                         self.lcm_tasks.register("vim_account", vim_id, order_id, "vim_delete", task)
 289                         continue
 290                     elif command == "show":
 291                         print("not implemented show with vim_account")
 292                         sys.stdout.flush()
 293                         continue
 294                     elif command == "edit":
 295                         task = asyncio.ensure_future(self.vim.edit(params, order_id))
 296                         self.lcm_tasks.register("vim_account", vim_id, order_id, "vim_edit", task)
 297                         continue
 298                 elif topic == "sdn":
 299                     _sdn_id = params["_id"]
 300                     if command == "create":
 301                         task = asyncio.ensure_future(self.sdn.create(params, order_id))
 302                         self.lcm_tasks.register("sdn", _sdn_id, order_id, "sdn_create", task)
 303                         continue
 304                     elif command == "delete":
 305                         self.lcm_tasks.cancel(topic, _sdn_id)
 306                         task = asyncio.ensure_future(self.sdn.delete(_sdn_id, order_id))
 307                         self.lcm_tasks.register("sdn", _sdn_id, order_id, "sdn_delete", task)
 308                         continue
 309                     elif command == "edit":
 310                         task = asyncio.ensure_future(self.sdn.edit(params, order_id))
 311                         self.lcm_tasks.register("sdn", _sdn_id, order_id, "sdn_edit", task)
 312                         continue
 313                 self.logger.critical("unknown topic {} and command '{}'".format(topic, command))
 314             except Exception as e:
 315                 # if not first_start is the first time after starting. So leave more time and wait
 316                 # to allow kafka starts
 317                 if consecutive_errors == 8 if not first_start else 30:
 318                     self.logger.error("Task kafka_read task exit error too many errors. Exception: {}".format(e))
 319                     raise
 320                 consecutive_errors += 1
 321                 self.logger.error("Task kafka_read retrying after Exception {}".format(e))
 322                 wait_time = 2 if not first_start else 5
 323                 await asyncio.sleep(wait_time, loop=self.loop)
 324
 325         # self.logger.debug("Task kafka_read terminating")
 326         self.logger.debug("Task kafka_read exit")
 327
 328     def health_check(self):
 329
 330         global exit_code
 331         task = None
 332         exit_code = 1
 333
 334         def health_check_callback(topic, command, params):
 335             global exit_code
 336             print("receiving callback {} {} {}".format(topic, command, params))
 337             if topic == "admin" and command == "ping" and params["to"] == "lcm" and params["from"] == "lcm":
 338                 # print("received LCM ping")
 339                 exit_code = 0
 340                 task.cancel()
 341
 342         try:
 343             task = asyncio.ensure_future(self.msg.aioread(("admin",), self.loop, health_check_callback))
 344             self.loop.run_until_complete(task)
 345         except Exception:
 346             pass
 347         exit(exit_code)
 348
 349     def start(self):
 350
 351         # check RO version
 352         self.loop.run_until_complete(self.check_RO_version())
 353
 354         self.loop.run_until_complete(asyncio.gather(
 355             self.kafka_read(),
 356             self.kafka_ping()
 357         ))
 358         # TODO
 359         # self.logger.debug("Terminating cancelling creation tasks")
 360         # self.lcm_tasks.cancel("ALL", "create")
 361         # timeout = 200
 362         # while self.is_pending_tasks():
 363         #     self.logger.debug("Task kafka_read terminating. Waiting for tasks termination")
 364         #     await asyncio.sleep(2, loop=self.loop)
 365         #     timeout -= 2
 366         #     if not timeout:
 367         #         self.lcm_tasks.cancel("ALL", "ALL")
 368         self.loop.close()
 369         self.loop = None
 370         if self.db:
 371             self.db.db_disconnect()
 372         if self.msg:
 373             self.msg.disconnect()
 374         if self.fs:
 375             self.fs.fs_disconnect()
 376
 377     def read_config_file(self, config_file):
 378         # TODO make a [ini] + yaml inside parser
 379         # the configparser library is not suitable, because it does not admit comments at the end of line,
 380         # and not parse integer or boolean
 381         try:
 382             with open(config_file) as f:
 383                 conf = yaml.load(f)
 384             for k, v in environ.items():
 385                 if not k.startswith("OSMLCM_"):
 386                     continue
 387                 k_items = k.lower().split("_")
 388                 if len(k_items) < 3:
 389                     continue
 390                 if k_items[1] in ("ro", "vca"):
 391                     # put in capital letter
 392                     k_items[1] = k_items[1].upper()
 393                 c = conf
 394                 try:
 395                     for k_item in k_items[1:-1]:
 396                         c = c[k_item]
 397                     if k_items[-1] == "port":
 398                         c[k_items[-1]] = int(v)
 399                     else:
 400                         c[k_items[-1]] = v
 401                 except Exception as e:
 402                     self.logger.warn("skipping environ '{}' on exception '{}'".format(k, e))
 403
 404             return conf
 405         except Exception as e:
 406             self.logger.critical("At config file '{}': {}".format(config_file, e))
 407             exit(1)
 408
 409
 410 def usage():
 411     print("""Usage: {} [options]
 412         -c|--config [configuration_file]: loads the configuration file (default: ./nbi.cfg)
 413         --health-check: do not run lcm, but inspect kafka bus to determine if lcm is healthy
 414         -h|--help: shows this help
 415         """.format(sys.argv[0]))
 416     # --log-socket-host HOST: send logs to this host")
 417     # --log-socket-port PORT: send logs using this port (default: 9022)")
 418
 419
 420 if __name__ == '__main__':
 421     try:
 422         # load parameters and configuration
 423         opts, args = getopt.getopt(sys.argv[1:], "hc:", ["config=", "help", "health-check"])
 424         # TODO add  "log-socket-host=", "log-socket-port=", "log-file="
 425         config_file = None
 426         health_check = None
 427         for o, a in opts:
 428             if o in ("-h", "--help"):
 429                 usage()
 430                 sys.exit()
 431             elif o in ("-c", "--config"):
 432                 config_file = a
 433             elif o == "--health-check":
 434                 health_check = True
 435             # elif o == "--log-socket-port":
 436             #     log_socket_port = a
 437             # elif o == "--log-socket-host":
 438             #     log_socket_host = a
 439             # elif o == "--log-file":
 440             #     log_file = a
 441             else:
 442                 assert False, "Unhandled option"
 443         if config_file:
 444             if not path.isfile(config_file):
 445                 print("configuration file '{}' not exist".format(config_file), file=sys.stderr)
 446                 exit(1)
 447         else:
 448             for config_file in (__file__[:__file__.rfind(".")] + ".cfg", "./lcm.cfg", "/etc/osm/lcm.cfg"):
 449                 if path.isfile(config_file):
 450                     break
 451             else:
 452                 print("No configuration file 'lcm.cfg' found neither at local folder nor at /etc/osm/", file=sys.stderr)
 453                 exit(1)
 454         lcm = Lcm(config_file)
 455         if health_check:
 456             lcm.health_check()
 457         else:
 458             lcm.start()
 459     except (LcmException, getopt.GetoptError) as e:
 460         print(str(e), file=sys.stderr)
 461         # usage()
 462         exit(1)