osm_lcm/lcm_helm_conn.py

   1 ##
   2 # Copyright 2020 Telefonica Investigacion y Desarrollo, S.A.U.
   3 #
   4 # Licensed under the Apache License, Version 2.0 (the "License");
   5 # you may not use this file except in compliance with the License.
   6 # You may obtain a copy of the License at
   7 #
   8 #    http://www.apache.org/licenses/LICENSE-2.0
   9 #
  10 # Unless required by applicable law or agreed to in writing, software
  11 # distributed under the License is distributed on an "AS IS" BASIS,
  12 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
  13 # implied.
  14 # See the License for the specific language governing permissions and
  15 # limitations under the License.
  16 #
  17 ##
  18 import functools
  19 import yaml
  20 import asyncio
  21 import socket
  22 import uuid
  23 import os
  24
  25 from grpclib.client import Channel
  26
  27 from osm_lcm.frontend_pb2 import PrimitiveRequest
  28 from osm_lcm.frontend_pb2 import SshKeyRequest, SshKeyReply
  29 from osm_lcm.frontend_grpc import FrontendExecutorStub
  30 from osm_lcm.lcm_utils import LcmBase
  31
  32 from osm_lcm.data_utils.database.database import Database
  33 from osm_lcm.data_utils.filesystem.filesystem import Filesystem
  34
  35 from n2vc.n2vc_conn import N2VCConnector
  36 from n2vc.k8s_helm_conn import K8sHelmConnector
  37 from n2vc.k8s_helm3_conn import K8sHelm3Connector
  38 from n2vc.exceptions import N2VCBadArgumentsException, N2VCException, N2VCExecutionException
  39
  40 from osm_lcm.lcm_utils import deep_get
  41
  42
  43 def retryer(max_wait_time_var="_initial_retry_time", delay_time_var="_retry_delay"):
  44     def wrapper(func):
  45         retry_exceptions = (
  46             ConnectionRefusedError
  47         )
  48
  49         @functools.wraps(func)
  50         async def wrapped(*args, **kwargs):
  51             # default values for wait time and delay_time
  52             delay_time = 10
  53             max_wait_time = 300
  54
  55             # obtain arguments from variable names
  56             self = args[0]
  57             if self.__dict__.get(max_wait_time_var):
  58                 max_wait_time = self.__dict__.get(max_wait_time_var)
  59             if self.__dict__.get(delay_time_var):
  60                 delay_time = self.__dict__.get(delay_time_var)
  61
  62             wait_time = max_wait_time
  63             while wait_time > 0:
  64                 try:
  65                     return await func(*args, **kwargs)
  66                 except retry_exceptions:
  67                     wait_time = wait_time - delay_time
  68                     await asyncio.sleep(delay_time)
  69                     continue
  70             else:
  71                 return ConnectionRefusedError
  72         return wrapped
  73     return wrapper
  74
  75
  76 class LCMHelmConn(N2VCConnector, LcmBase):
  77     _KUBECTL_OSM_NAMESPACE = "osm"
  78     _KUBECTL_OSM_CLUSTER_NAME = "_system-osm-k8s"
  79     _EE_SERVICE_PORT = 50050
  80
  81     # Initial max retry time
  82     _MAX_INITIAL_RETRY_TIME = 600
  83     # Max retry time for normal operations
  84     _MAX_RETRY_TIME = 30
  85     # Time beetween retries, retry time after a connection error is raised
  86     _EE_RETRY_DELAY = 10
  87
  88     def __init__(self,
  89                  log: object = None,
  90                  loop: object = None,
  91                  vca_config: dict = None,
  92                  on_update_db=None, ):
  93         """
  94         Initialize EE helm connector.
  95         """
  96
  97         self.db = Database().instance.db
  98         self.fs = Filesystem().instance.fs
  99
 100         # parent class constructor
 101         N2VCConnector.__init__(
 102             self,
 103             log=log,
 104             loop=loop,
 105             on_update_db=on_update_db,
 106             db=self.db,
 107             fs=self.fs
 108         )
 109
 110         self.vca_config = vca_config
 111         self.log.debug("Initialize helm N2VC connector")
 112         self.log.debug("initial vca_config: {}".format(vca_config))
 113
 114         # TODO - Obtain data from configuration
 115         self._ee_service_port = self._EE_SERVICE_PORT
 116
 117         self._retry_delay = self._EE_RETRY_DELAY
 118
 119         if self.vca_config and self.vca_config.get("eegrpcinittimeout"):
 120             self._initial_retry_time = self.vca_config.get("eegrpcinittimeout")
 121             self.log.debug("Initial retry time: {}".format(self._initial_retry_time))
 122         else:
 123             self._initial_retry_time = self._MAX_INITIAL_RETRY_TIME
 124             self.log.debug("Applied default retry time: {}".format(self._initial_retry_time))
 125
 126         if self.vca_config and self.vca_config.get("eegrpctimeout"):
 127             self._max_retry_time = self.vca_config.get("eegrpctimeout")
 128             self.log.debug("Retry time: {}".format(self._max_retry_time))
 129         else:
 130             self._max_retry_time = self._MAX_RETRY_TIME
 131             self.log.debug("Applied default retry time: {}".format(self._max_retry_time))
 132
 133         # initialize helm connector for helmv2 and helmv3
 134         self._k8sclusterhelm2 = K8sHelmConnector(
 135             kubectl_command=self.vca_config.get("kubectlpath"),
 136             helm_command=self.vca_config.get("helmpath"),
 137             fs=self.fs,
 138             db=self.db,
 139             log=self.log,
 140             on_update_db=None,
 141         )
 142
 143         self._k8sclusterhelm3 = K8sHelm3Connector(
 144             kubectl_command=self.vca_config.get("kubectlpath"),
 145             helm_command=self.vca_config.get("helm3path"),
 146             fs=self.fs,
 147             log=self.log,
 148             db=self.db,
 149             on_update_db=None,
 150         )
 151
 152         self._system_cluster_id = None
 153         self.log.info("Helm N2VC connector initialized")
 154
 155     # TODO - ¿reuse_ee_id?
 156     async def create_execution_environment(self,
 157                                            namespace: str,
 158                                            db_dict: dict,
 159                                            reuse_ee_id: str = None,
 160                                            progress_timeout: float = None,
 161                                            total_timeout: float = None,
 162                                            config: dict = None,
 163                                            artifact_path: str = None,
 164                                            vca_type: str = None,
 165                                            *kargs, **kwargs) -> (str, dict):
 166         """
 167         Creates a new helm execution environment deploying the helm-chat indicated in the
 168         attifact_path
 169         :param str namespace: This param is not used, all helm charts are deployed in the osm
 170         system namespace
 171         :param dict db_dict: where to write to database when the status changes.
 172             It contains a dictionary with {collection: str, filter: {},  path: str},
 173                 e.g. {collection: "nsrs", filter: {_id: <nsd-id>, path:
 174                 "_admin.deployed.VCA.3"}
 175         :param str reuse_ee_id: ee id from an older execution. TODO - right now this params is not used
 176         :param float progress_timeout:
 177         :param float total_timeout:
 178         :param dict config:  General variables to instantiate KDU
 179         :param str artifact_path:  path of package content
 180         :param str vca_type:  Type of vca, must be type helm or helm-v3
 181         :returns str, dict: id of the new execution environment including namespace.helm_id
 182         and credentials object set to None as all credentials should be osm kubernetes .kubeconfig
 183         """
 184
 185         self.log.info(
 186             "create_execution_environment: namespace: {}, artifact_path: {}, db_dict: {}, "
 187             "reuse_ee_id: {}".format(
 188                 namespace, artifact_path, db_dict, reuse_ee_id)
 189         )
 190
 191         # Validate artifact-path is provided
 192         if artifact_path is None or len(artifact_path) == 0:
 193             raise N2VCBadArgumentsException(
 194                 message="artifact_path is mandatory", bad_args=["artifact_path"]
 195             )
 196
 197         # Validate artifact-path exists and sync path
 198         from_path = os.path.split(artifact_path)[0]
 199         self.fs.sync(from_path)
 200
 201         # remove / in charm path
 202         while artifact_path.find("//") >= 0:
 203             artifact_path = artifact_path.replace("//", "/")
 204
 205         # check charm path
 206         if self.fs.file_exists(artifact_path):
 207             helm_chart_path = artifact_path
 208         else:
 209             msg = "artifact path does not exist: {}".format(artifact_path)
 210             raise N2VCBadArgumentsException(message=msg, bad_args=["artifact_path"])
 211
 212         if artifact_path.startswith("/"):
 213             full_path = self.fs.path + helm_chart_path
 214         else:
 215             full_path = self.fs.path + "/" + helm_chart_path
 216
 217         while full_path.find("//") >= 0:
 218             full_path = full_path.replace("//", "/")
 219
 220         try:
 221             # Call helm conn install
 222             # Obtain system cluster id from database
 223             system_cluster_uuid = await self._get_system_cluster_id()
 224             # Add parameter osm if exist to global
 225             if config and config.get("osm"):
 226                 if not config.get("global"):
 227                     config["global"] = {}
 228                 config["global"]["osm"] = config.get("osm")
 229
 230             self.log.debug("install helm chart: {}".format(full_path))
 231             if vca_type == "helm":
 232                 helm_id = self._k8sclusterhelm2.generate_kdu_instance_name(
 233                     db_dict=db_dict,
 234                     kdu_model=full_path,
 235                 )
 236                 await self._k8sclusterhelm2.install(system_cluster_uuid, kdu_model=full_path,
 237                                                     kdu_instance=helm_id,
 238                                                     namespace=self._KUBECTL_OSM_NAMESPACE,
 239                                                     params=config,
 240                                                     db_dict=db_dict,
 241                                                     timeout=progress_timeout)
 242             else:
 243                 helm_id = self._k8sclusterhelm2.generate_kdu_instance_name(
 244                     db_dict=db_dict,
 245                     kdu_model=full_path,
 246                 )
 247                 await self._k8sclusterhelm3.install(system_cluster_uuid, kdu_model=full_path,
 248                                                     kdu_instance=helm_id,
 249                                                     namespace=self._KUBECTL_OSM_NAMESPACE,
 250                                                     params=config,
 251                                                     db_dict=db_dict,
 252                                                     timeout=progress_timeout)
 253
 254             ee_id = "{}:{}.{}".format(vca_type, self._KUBECTL_OSM_NAMESPACE, helm_id)
 255             return ee_id, None
 256         except N2VCException:
 257             raise
 258         except Exception as e:
 259             self.log.error("Error deploying chart ee: {}".format(e), exc_info=True)
 260             raise N2VCException("Error deploying chart ee: {}".format(e))
 261
 262     async def register_execution_environment(self, namespace: str, credentials: dict, db_dict: dict,
 263                                              progress_timeout: float = None, total_timeout: float = None,
 264                                              *kargs, **kwargs) -> str:
 265         # nothing to do
 266         pass
 267
 268     async def install_configuration_sw(self, *args, **kwargs):
 269         # nothing to do
 270         pass
 271
 272     async def add_relation(self, *args, **kwargs):
 273         # nothing to do
 274         pass
 275
 276     async def remove_relation(self):
 277         # nothing to to
 278         pass
 279
 280     async def get_status(self, *args, **kwargs):
 281         # not used for this connector
 282         pass
 283
 284     async def get_ee_ssh_public__key(
 285         self,
 286         ee_id: str,
 287         db_dict: dict,
 288         progress_timeout: float = None,
 289         total_timeout: float = None,
 290         **kwargs,
 291     ) -> str:
 292         """
 293         Obtains ssh-public key from ee executing GetSShKey method from the ee.
 294
 295         :param str ee_id: the id of the execution environment returned by
 296             create_execution_environment or register_execution_environment
 297         :param dict db_dict:
 298         :param float progress_timeout:
 299         :param float total_timeout:
 300         :returns: public key of the execution environment
 301         """
 302
 303         self.log.info(
 304             "get_ee_ssh_public_key: ee_id: {}, db_dict: {}".format(
 305                 ee_id, db_dict)
 306         )
 307
 308         # check arguments
 309         if ee_id is None or len(ee_id) == 0:
 310             raise N2VCBadArgumentsException(
 311                 message="ee_id is mandatory", bad_args=["ee_id"]
 312             )
 313
 314         try:
 315             # Obtain ip_addr for the ee service, it is resolved by dns from the ee name by kubernetes
 316             version, namespace, helm_id = self._get_ee_id_parts(ee_id)
 317             ip_addr = socket.gethostbyname(helm_id)
 318
 319             # Obtain ssh_key from the ee, this method will implement retries to allow the ee
 320             # install libraries and start successfully
 321             ssh_key = await self._get_ssh_key(ip_addr)
 322             return ssh_key
 323         except Exception as e:
 324             self.log.error("Error obtaining ee ssh_key: {}".format(e), exc_info=True)
 325             raise N2VCException("Error obtaining ee ssh_ke: {}".format(e))
 326
 327     async def exec_primitive(
 328             self,
 329             ee_id: str,
 330             primitive_name: str,
 331             params_dict: dict,
 332             db_dict: dict = None,
 333             progress_timeout: float = None,
 334             total_timeout: float = None,
 335             **kwargs,
 336     ) -> str:
 337         """
 338         Execute a primitive in the execution environment
 339
 340         :param str ee_id: the one returned by create_execution_environment or
 341             register_execution_environment with the format namespace.helm_id
 342         :param str primitive_name: must be one defined in the software. There is one
 343             called 'config', where, for the proxy case, the 'credentials' of VM are
 344             provided
 345         :param dict params_dict: parameters of the action
 346         :param dict db_dict: where to write into database when the status changes.
 347                         It contains a dict with
 348                             {collection: <str>, filter: {},  path: <str>},
 349                             e.g. {collection: "nslcmops", filter:
 350                                 {_id: <nslcmop_id>, path: "_admin.VCA"}
 351                         It will be used to store information about intermediate notifications
 352         :param float progress_timeout:
 353         :param float total_timeout:
 354         :returns str: primitive result, if ok. It raises exceptions in case of fail
 355         """
 356
 357         self.log.info("exec primitive for ee_id : {}, primitive_name: {}, params_dict: {}, db_dict: {}".format(
 358             ee_id, primitive_name, params_dict, db_dict
 359         ))
 360
 361         # check arguments
 362         if ee_id is None or len(ee_id) == 0:
 363             raise N2VCBadArgumentsException(
 364                 message="ee_id is mandatory", bad_args=["ee_id"]
 365             )
 366         if primitive_name is None or len(primitive_name) == 0:
 367             raise N2VCBadArgumentsException(
 368                 message="action_name is mandatory", bad_args=["action_name"]
 369             )
 370         if params_dict is None:
 371             params_dict = dict()
 372
 373         try:
 374             version, namespace, helm_id = self._get_ee_id_parts(ee_id)
 375             ip_addr = socket.gethostbyname(helm_id)
 376         except Exception as e:
 377             self.log.error("Error getting ee ip ee: {}".format(e))
 378             raise N2VCException("Error getting ee ip ee: {}".format(e))
 379
 380         if primitive_name == "config":
 381             try:
 382                 # Execute config primitive, higher timeout to check the case ee is starting
 383                 status, detailed_message = await self._execute_config_primitive(ip_addr, params_dict, db_dict=db_dict)
 384                 self.log.debug("Executed config primitive ee_id_ {}, status: {}, message: {}".format(
 385                     ee_id, status, detailed_message))
 386                 if status != "OK":
 387                     self.log.error("Error configuring helm ee, status: {}, message: {}".format(
 388                         status, detailed_message))
 389                     raise N2VCExecutionException(
 390                         message="Error configuring helm ee_id: {}, status: {}, message: {}: ".format(
 391                             ee_id, status, detailed_message
 392                         ),
 393                         primitive_name=primitive_name,
 394                     )
 395             except Exception as e:
 396                 self.log.error("Error configuring helm ee: {}".format(e))
 397                 raise N2VCExecutionException(
 398                     message="Error configuring helm ee_id: {}, {}".format(
 399                         ee_id, e
 400                     ),
 401                     primitive_name=primitive_name,
 402                 )
 403             return "CONFIG OK"
 404         else:
 405             try:
 406                 # Execute primitive
 407                 status, detailed_message = await self._execute_primitive(ip_addr, primitive_name,
 408                                                                          params_dict, db_dict=db_dict)
 409                 self.log.debug("Executed primitive {} ee_id_ {}, status: {}, message: {}".format(
 410                     primitive_name, ee_id, status, detailed_message))
 411                 if status != "OK" and status != "PROCESSING":
 412                     self.log.error(
 413                         "Execute primitive {} returned not ok status: {}, message: {}".format(
 414                             primitive_name, status, detailed_message)
 415                     )
 416                     raise N2VCExecutionException(
 417                         message="Execute primitive {} returned not ok status: {}, message: {}".format(
 418                             primitive_name, status, detailed_message
 419                         ),
 420                         primitive_name=primitive_name,
 421                     )
 422             except Exception as e:
 423                 self.log.error(
 424                     "Error executing primitive {}: {}".format(primitive_name, e)
 425                 )
 426                 raise N2VCExecutionException(
 427                     message="Error executing primitive {} into ee={} : {}".format(
 428                         primitive_name, ee_id, e
 429                     ),
 430                     primitive_name=primitive_name,
 431                 )
 432             return detailed_message
 433
 434     async def deregister_execution_environments(self):
 435         # nothing to be done
 436         pass
 437
 438     async def delete_execution_environment(
 439         self,
 440         ee_id: str,
 441         db_dict: dict = None,
 442         total_timeout: float = None,
 443         **kwargs,
 444     ):
 445         """
 446         Delete an execution environment
 447         :param str ee_id: id of the execution environment to delete, included namespace.helm_id
 448         :param dict db_dict: where to write into database when the status changes.
 449                         It contains a dict with
 450                             {collection: <str>, filter: {},  path: <str>},
 451                             e.g. {collection: "nsrs", filter:
 452                                 {_id: <nsd-id>, path: "_admin.deployed.VCA.3"}
 453         :param float total_timeout:
 454         """
 455
 456         self.log.info("ee_id: {}".format(ee_id))
 457
 458         # check arguments
 459         if ee_id is None:
 460             raise N2VCBadArgumentsException(
 461                 message="ee_id is mandatory", bad_args=["ee_id"]
 462             )
 463
 464         try:
 465
 466             # Obtain cluster_uuid
 467             system_cluster_uuid = await self._get_system_cluster_id()
 468
 469             # Get helm_id
 470             version, namespace, helm_id = self._get_ee_id_parts(ee_id)
 471
 472             # Uninstall chart, for backward compatibility we must assume that if there is no
 473             # version it is helm-v2
 474             if version == "helm-v3":
 475                 await self._k8sclusterhelm3.uninstall(system_cluster_uuid, helm_id)
 476             else:
 477                 await self._k8sclusterhelm2.uninstall(system_cluster_uuid, helm_id)
 478             self.log.info("ee_id: {} deleted".format(ee_id))
 479         except N2VCException:
 480             raise
 481         except Exception as e:
 482             self.log.error("Error deleting ee id: {}: {}".format(ee_id, e), exc_info=True)
 483             raise N2VCException("Error deleting ee id {}: {}".format(ee_id, e))
 484
 485     async def delete_namespace(self, namespace: str, db_dict: dict = None, total_timeout: float = None):
 486         # method not implemented for this connector, execution environments must be deleted individually
 487         pass
 488
 489     async def install_k8s_proxy_charm(
 490         self,
 491         charm_name: str,
 492         namespace: str,
 493         artifact_path: str,
 494         db_dict: dict,
 495         progress_timeout: float = None,
 496         total_timeout: float = None,
 497         config: dict = None,
 498         *kargs, **kwargs
 499     ) -> str:
 500         pass
 501
 502     @retryer(max_wait_time_var="_initial_retry_time", delay_time_var="_retry_delay")
 503     async def _get_ssh_key(self, ip_addr):
 504         channel = Channel(ip_addr, self._ee_service_port)
 505         try:
 506             stub = FrontendExecutorStub(channel)
 507             self.log.debug("get ssh key, ip_addr: {}".format(ip_addr))
 508             reply: SshKeyReply = await stub.GetSshKey(SshKeyRequest())
 509             return reply.message
 510         finally:
 511             channel.close()
 512
 513     @retryer(max_wait_time_var="_initial_retry_time", delay_time_var="_retry_delay")
 514     async def _execute_config_primitive(self, ip_addr, params, db_dict=None):
 515         return await self._execute_primitive_internal(ip_addr, "config", params, db_dict=db_dict)
 516
 517     @retryer(max_wait_time_var="_max_retry_time", delay_time_var="_retry_delay")
 518     async def _execute_primitive(self, ip_addr, primitive_name, params, db_dict=None):
 519         return await self._execute_primitive_internal(ip_addr, primitive_name, params, db_dict=db_dict)
 520
 521     async def _execute_primitive_internal(self, ip_addr, primitive_name, params, db_dict=None):
 522
 523         channel = Channel(ip_addr, self._ee_service_port)
 524         try:
 525             stub = FrontendExecutorStub(channel)
 526             async with stub.RunPrimitive.open() as stream:
 527                 primitive_id = str(uuid.uuid1())
 528                 result = None
 529                 self.log.debug("Execute primitive internal: id:{}, name:{}, params: {}".
 530                                format(primitive_id, primitive_name, params))
 531                 await stream.send_message(
 532                     PrimitiveRequest(id=primitive_id, name=primitive_name, params=yaml.dump(params)), end=True)
 533                 async for reply in stream:
 534                     self.log.debug("Received reply: {}".format(reply))
 535                     result = reply
 536                     # If db_dict provided write notifs in database
 537                     if db_dict:
 538                         self._write_op_detailed_status(db_dict, reply.status, reply.detailed_message)
 539                 if result:
 540                     return reply.status, reply.detailed_message
 541                 else:
 542                     return "ERROR", "No result received"
 543         finally:
 544             channel.close()
 545
 546     def _write_op_detailed_status(self, db_dict, status, detailed_message):
 547
 548         # write ee_id to database: _admin.deployed.VCA.x
 549         try:
 550             the_table = db_dict["collection"]
 551             the_filter = db_dict["filter"]
 552             update_dict = {"detailed-status": "{}: {}".format(status, detailed_message)}
 553             # self.log.debug('Writing ee_id to database: {}'.format(the_path))
 554             self.db.set_one(
 555                 table=the_table,
 556                 q_filter=the_filter,
 557                 update_dict=update_dict,
 558                 fail_on_empty=True,
 559             )
 560         except asyncio.CancelledError:
 561             raise
 562         except Exception as e:
 563             self.log.error("Error writing detailedStatus to database: {}".format(e))
 564
 565     async def _get_system_cluster_id(self):
 566         if not self._system_cluster_id:
 567             db_k8cluster = self.db.get_one("k8sclusters", {"name": self._KUBECTL_OSM_CLUSTER_NAME})
 568             k8s_hc_id = deep_get(db_k8cluster, ("_admin", "helm-chart-v3", "id"))
 569             if not k8s_hc_id:
 570                 try:
 571                     # backward compatibility for existing clusters that have not been initialized for helm v3
 572                     cluster_id = db_k8cluster.get("_id")
 573                     k8s_credentials = yaml.safe_dump(db_k8cluster.get("credentials"))
 574                     k8s_hc_id, uninstall_sw = await self._k8sclusterhelm3.init_env(k8s_credentials,
 575                                                                                    reuse_cluster_uuid=cluster_id)
 576                     db_k8scluster_update = {"_admin.helm-chart-v3.error_msg": None,
 577                                             "_admin.helm-chart-v3.id": k8s_hc_id,
 578                                             "_admin.helm-chart-v3}.created": uninstall_sw,
 579                                             "_admin.helm-chart-v3.operationalState": "ENABLED"}
 580                     self.update_db_2("k8sclusters", cluster_id, db_k8scluster_update)
 581                 except Exception as e:
 582                     self.log.error("error initializing helm-v3 cluster: {}".format(str(e)))
 583                     raise N2VCException("K8s system cluster '{}' has not been initialized for helm-chart-v3".format(
 584                         cluster_id))
 585             self._system_cluster_id = k8s_hc_id
 586         return self._system_cluster_id
 587
 588     def _get_ee_id_parts(self, ee_id):
 589         """
 590         Parses ee_id stored at database that can be either 'version:namespace.helm_id' or only
 591         namespace.helm_id for backward compatibility
 592         If exists helm version can be helm-v3 or helm (helm-v2 old version)
 593         """
 594         version, _, part_id = ee_id.rpartition(':')
 595         namespace, _, helm_id = part_id.rpartition('.')
 596         return version, namespace, helm_id