X-Git-Url: https://osm.etsi.org/gitweb/?a=blobdiff_plain;f=osm_lcm%2Flcm.py;h=274154b16834e89b08c815d02264adfe8b022713;hb=f210c1c8a326d56fea613c7c7c1ff3f17dd5901c;hp=a01d1883a65d1bc29c914afdb5cfbfca70f6828c;hpb=8a5188754662c19e2471e5e7a1172a3bccea669e;p=osm%2FLCM.git diff --git a/osm_lcm/lcm.py b/osm_lcm/lcm.py index a01d188..274154b 100644 --- a/osm_lcm/lcm.py +++ b/osm_lcm/lcm.py @@ -23,29 +23,30 @@ import logging import logging.handlers import getopt import sys -import ROclient -import ns -import vim_sdn -import netslice -from lcm_utils import versiontuple, LcmException, TaskRegistry, LcmExceptionExit -# from osm_lcm import version as lcm_version, version_date as lcm_version_date, ROclient +from osm_lcm import ROclient, ns, vim_sdn, netslice +from time import time, sleep +from osm_lcm.lcm_utils import versiontuple, LcmException, TaskRegistry, LcmExceptionExit +from osm_lcm import version as lcm_version, version_date as lcm_version_date + from osm_common import dbmemory, dbmongo, fslocal, msglocal, msgkafka from osm_common import version as common_version from osm_common.dbbase import DbException from osm_common.fsbase import FsException from osm_common.msgbase import MsgException from os import environ, path +from random import choice as random_choice from n2vc import version as n2vc_version __author__ = "Alfonso Tierno" -min_RO_version = [0, 6, 0] +min_RO_version = "6.0.2" min_n2vc_version = "0.0.2" -min_common_version = "0.1.11" +min_common_version = "0.1.19" # uncomment if LCM is installed as library and installed, and get them from __init__.py -lcm_version = '0.1.30' -lcm_version_date = '2019-01-10' +# lcm_version = '0.1.41' +# lcm_version_date = '2019-06-19' +health_check_file = path.expanduser("~") + "/time_last_ping" # TODO find better location for this file class Lcm: @@ -62,13 +63,16 @@ class Lcm: self.db = None self.msg = None + self.msg_admin = None self.fs = None self.pings_not_received = 1 + self.consecutive_errors = 0 + self.first_start = False - # contains created tasks/futures to be able to cancel - self.lcm_tasks = TaskRegistry() # logging self.logger = logging.getLogger('lcm') + # get id + self.worker_id = self.get_process_id() # load configuration config = self.read_config_file(config_file) self.config = config @@ -145,19 +149,30 @@ class Lcm: raise LcmException("Invalid configuration param '{}' at '[storage]':'driver'".format( config["storage"]["driver"])) - if config["message"]["driver"] == "local": + config_message = config["message"].copy() + config_message["loop"] = self.loop + if config_message["driver"] == "local": self.msg = msglocal.MsgLocal() - self.msg.connect(config["message"]) - elif config["message"]["driver"] == "kafka": + self.msg.connect(config_message) + self.msg_admin = msglocal.MsgLocal() + config_message.pop("group_id", None) + self.msg_admin.connect(config_message) + elif config_message["driver"] == "kafka": self.msg = msgkafka.MsgKafka() - self.msg.connect(config["message"]) + self.msg.connect(config_message) + self.msg_admin = msgkafka.MsgKafka() + config_message.pop("group_id", None) + self.msg_admin.connect(config_message) else: raise LcmException("Invalid configuration param '{}' at '[message]':'driver'".format( - config["storage"]["driver"])) + config["message"]["driver"])) except (DbException, FsException, MsgException) as e: self.logger.critical(str(e), exc_info=True) raise LcmException(str(e)) + # contains created tasks/futures to be able to cancel + self.lcm_tasks = TaskRegistry(self.worker_id, self.db, self.logger) + self.ns = ns.NsLcm(self.db, self.msg, self.fs, self.lcm_tasks, self.ro_config, self.vca_config, self.loop) self.netslice = netslice.NetsliceLcm(self.db, self.msg, self.fs, self.lcm_tasks, self.ro_config, self.vca_config, self.loop) @@ -166,17 +181,27 @@ class Lcm: self.sdn = vim_sdn.SdnLcm(self.db, self.msg, self.fs, self.lcm_tasks, self.ro_config, self.loop) async def check_RO_version(self): - try: - RO = ROclient.ROClient(self.loop, **self.ro_config) - RO_version = await RO.get_version() - if RO_version < min_RO_version: - raise LcmException("Not compatible osm/RO version '{}.{}.{}'. Needed '{}.{}.{}' or higher".format( - *RO_version, *min_RO_version - )) - except ROclient.ROClientException as e: - error_text = "Error while conneting to osm/RO " + str(e) - self.logger.critical(error_text, exc_info=True) - raise LcmException(error_text) + tries = 14 + last_error = None + while True: + try: + ro_server = ROclient.ROClient(self.loop, **self.ro_config) + ro_version = await ro_server.get_version() + if versiontuple(ro_version) < versiontuple(min_RO_version): + raise LcmException("Not compatible osm/RO version '{}'. Needed '{}' or higher".format( + ro_version, min_RO_version)) + self.logger.info("Connected to RO version {}".format(ro_version)) + return + except ROclient.ROClientException as e: + tries -= 1 + error_text = "Error while connecting to RO on {}: {}".format(self.ro_config["endpoint_url"], e) + if tries <= 0: + self.logger.critical(error_text) + raise LcmException(error_text) + if last_error != error_text: + last_error = error_text + self.logger.error(error_text + ". Waiting until {} seconds".format(5*tries)) + await asyncio.sleep(5) async def test(self, param=None): self.logger.debug("Starting/Ending test task: {}".format(param)) @@ -189,7 +214,10 @@ class Lcm: self.pings_not_received = 1 while True: try: - await self.msg.aiowrite("admin", "ping", {"from": "lcm", "to": "lcm"}, self.loop) + await self.msg_admin.aiowrite( + "admin", "ping", + {"from": "lcm", "to": "lcm", "worker_id": self.worker_id, "version": lcm_version}, + self.loop) # time between pings are low when it is not received and at starting wait_time = self.ping_interval_boot if not kafka_has_received else self.ping_interval_pace if not self.pings_not_received: @@ -210,7 +238,7 @@ class Lcm: raise consecutive_errors += 1 self.logger.error("Task kafka_read retrying after Exception {}".format(e)) - wait_time = 1 if not first_start else 5 + wait_time = 2 if not first_start else 5 await asyncio.sleep(wait_time, loop=self.loop) def kafka_read_callback(self, topic, command, params): @@ -236,7 +264,14 @@ class Lcm: if topic == "admin": if command == "ping" and params["to"] == "lcm" and params["from"] == "lcm": + if params.get("worker_id") != self.worker_id: + return self.pings_not_received = 0 + try: + with open(health_check_file, "w") as f: + f.write(str(time())) + except Exception as e: + self.logger.error("Cannot write into '{}' for healthcheck: {}".format(health_check_file, e)) return elif topic == "ns": if command == "instantiate": @@ -273,6 +308,7 @@ class Lcm: self.lcm_tasks.register("ns", nsr_id, nslcmop_id, "ns_scale", task) return elif command == "show": + nsr_id = params try: db_nsr = self.db.get_one("nsrs", {"_id": nsr_id}) print("nsr:\n _id={}\n operational-status: {}\n config-status: {}" @@ -307,6 +343,7 @@ class Lcm: self.lcm_tasks.register("nsi", nsir_id, nsilcmop_id, "nsi_terminate", task) return elif command == "show": + nsir_id = params try: db_nsir = self.db.get_one("nsirs", {"_id": nsir_id}) print("nsir:\n _id={}\n operational-status: {}\n config-status: {}" @@ -324,68 +361,78 @@ class Lcm: return elif topic == "vim_account": vim_id = params["_id"] - if command == "create": + if command in ("create", "created"): task = asyncio.ensure_future(self.vim.create(params, order_id)) self.lcm_tasks.register("vim_account", vim_id, order_id, "vim_create", task) return elif command == "delete": self.lcm_tasks.cancel(topic, vim_id) - task = asyncio.ensure_future(self.vim.delete(vim_id, order_id)) + task = asyncio.ensure_future(self.vim.delete(params, order_id)) self.lcm_tasks.register("vim_account", vim_id, order_id, "vim_delete", task) return elif command == "show": print("not implemented show with vim_account") sys.stdout.flush() return - elif command == "edit": + elif command in ("edit", "edited"): task = asyncio.ensure_future(self.vim.edit(params, order_id)) self.lcm_tasks.register("vim_account", vim_id, order_id, "vim_edit", task) return + elif command == "deleted": + return # TODO cleaning of task just in case should be done elif topic == "wim_account": wim_id = params["_id"] - if command == "create": + if command in ("create", "created"): task = asyncio.ensure_future(self.wim.create(params, order_id)) self.lcm_tasks.register("wim_account", wim_id, order_id, "wim_create", task) return elif command == "delete": self.lcm_tasks.cancel(topic, wim_id) - task = asyncio.ensure_future(self.wim.delete(wim_id, order_id)) + task = asyncio.ensure_future(self.wim.delete(params, order_id)) self.lcm_tasks.register("wim_account", wim_id, order_id, "wim_delete", task) return elif command == "show": print("not implemented show with wim_account") sys.stdout.flush() return - elif command == "edit": + elif command in ("edit", "edited"): task = asyncio.ensure_future(self.wim.edit(params, order_id)) self.lcm_tasks.register("wim_account", wim_id, order_id, "wim_edit", task) return + elif command == "deleted": + return # TODO cleaning of task just in case should be done elif topic == "sdn": _sdn_id = params["_id"] - if command == "create": + if command in ("create", "created"): task = asyncio.ensure_future(self.sdn.create(params, order_id)) self.lcm_tasks.register("sdn", _sdn_id, order_id, "sdn_create", task) return elif command == "delete": self.lcm_tasks.cancel(topic, _sdn_id) - task = asyncio.ensure_future(self.sdn.delete(_sdn_id, order_id)) + task = asyncio.ensure_future(self.sdn.delete(params, order_id)) self.lcm_tasks.register("sdn", _sdn_id, order_id, "sdn_delete", task) return - elif command == "edit": + elif command in ("edit", "edited"): task = asyncio.ensure_future(self.sdn.edit(params, order_id)) self.lcm_tasks.register("sdn", _sdn_id, order_id, "sdn_edit", task) return + elif command == "deleted": + return # TODO cleaning of task just in case should be done self.logger.critical("unknown topic {} and command '{}'".format(topic, command)) async def kafka_read(self): - self.logger.debug("Task kafka_read Enter") + self.logger.debug("Task kafka_read Enter with worker_id={}".format(self.worker_id)) # future = asyncio.Future() self.consecutive_errors = 0 self.first_start = True while self.consecutive_errors < 10: try: - topics = ("admin", "ns", "vim_account", "wim_account", "sdn", "nsi") - await self.msg.aioread(topics, self.loop, self.kafka_read_callback) + topics = ("ns", "vim_account", "wim_account", "sdn", "nsi") + topics_admin = ("admin", ) + await asyncio.gather( + self.msg.aioread(topics, self.loop, self.kafka_read_callback), + self.msg_admin.aioread(topics_admin, self.loop, self.kafka_read_callback, group_id=False) + ) except LcmExceptionExit: self.logger.debug("Bye!") @@ -404,27 +451,6 @@ class Lcm: # self.logger.debug("Task kafka_read terminating") self.logger.debug("Task kafka_read exit") - def health_check(self): - - global exit_code - task = None - exit_code = 1 - - def health_check_callback(topic, command, params): - global exit_code - print("receiving callback {} {} {}".format(topic, command, params)) - if topic == "admin" and command == "ping" and params["to"] == "lcm" and params["from"] == "lcm": - # print("received LCM ping") - exit_code = 0 - task.cancel() - - try: - task = asyncio.ensure_future(self.msg.aioread(("admin",), self.loop, health_check_callback)) - self.loop.run_until_complete(task) - except Exception: - pass - exit(exit_code) - def start(self): # check RO version @@ -450,6 +476,8 @@ class Lcm: self.db.db_disconnect() if self.msg: self.msg.disconnect() + if self.msg_admin: + self.msg_admin.disconnect() if self.fs: self.fs.fs_disconnect() @@ -485,6 +513,26 @@ class Lcm: self.logger.critical("At config file '{}': {}".format(config_file, e)) exit(1) + @staticmethod + def get_process_id(): + """ + Obtain a unique ID for this process. If running from inside docker, it will get docker ID. If not it + will provide a random one + :return: Obtained ID + """ + # Try getting docker id. If fails, get pid + try: + with open("/proc/self/cgroup", "r") as f: + text_id_ = f.readline() + _, _, text_id = text_id_.rpartition("/") + text_id = text_id.replace('\n', '')[:12] + if text_id: + return text_id + except Exception: + pass + # Return a random id + return ''.join(random_choice("0123456789abcdef") for _ in range(12)) + def usage(): print("""Usage: {} [options] @@ -496,13 +544,29 @@ def usage(): # --log-socket-port PORT: send logs using this port (default: 9022)") +def health_check(): + retry = 2 + while retry: + retry -= 1 + try: + with open(health_check_file, "r") as f: + last_received_ping = f.read() + + if time() - float(last_received_ping) < Lcm.ping_interval_pace + 10: + exit(0) + except Exception: + pass + if retry: + sleep(6) + exit(1) + + if __name__ == '__main__': try: # load parameters and configuration opts, args = getopt.getopt(sys.argv[1:], "hc:", ["config=", "help", "health-check"]) # TODO add "log-socket-host=", "log-socket-port=", "log-file=" config_file = None - health_check = None for o, a in opts: if o in ("-h", "--help"): usage() @@ -510,7 +574,7 @@ if __name__ == '__main__': elif o in ("-c", "--config"): config_file = a elif o == "--health-check": - health_check = True + health_check() # elif o == "--log-socket-port": # log_socket_port = a # elif o == "--log-socket-host": @@ -531,10 +595,7 @@ if __name__ == '__main__': print("No configuration file 'lcm.cfg' found neither at local folder nor at /etc/osm/", file=sys.stderr) exit(1) lcm = Lcm(config_file) - if health_check: - lcm.health_check() - else: - lcm.start() + lcm.start() except (LcmException, getopt.GetoptError) as e: print(str(e), file=sys.stderr) # usage()