X-Git-Url: https://osm.etsi.org/gitweb/?a=blobdiff_plain;f=osm_lcm%2Fns.py;h=fad397290657b38d38d57e0949e8ec387e593f2d;hb=4c0e6805c44f9ed1d0bb35161bf69645f5b84151;hp=1bcf4c7a196c9004431f780a36a2acdc4a8d5a75;hpb=1900848f8ab8e046f5184da5a2763e22ab7ba648;p=osm%2FLCM.git diff --git a/osm_lcm/ns.py b/osm_lcm/ns.py index 1bcf4c7..fad3972 100644 --- a/osm_lcm/ns.py +++ b/osm_lcm/ns.py @@ -17,25 +17,86 @@ ## import asyncio +import shutil +from typing import Any, Dict, List import yaml import logging import logging.handlers import traceback import json -from jinja2 import Environment, TemplateError, TemplateNotFound, StrictUndefined, UndefinedError +from jinja2 import ( + Environment, + TemplateError, + TemplateNotFound, + StrictUndefined, + UndefinedError, + select_autoescape, +) from osm_lcm import ROclient +from osm_lcm.data_utils.lcm_config import LcmCfg +from osm_lcm.data_utils.nsr import ( + get_deployed_kdu, + get_deployed_vca, + get_deployed_vca_list, + get_nsd, +) +from osm_lcm.data_utils.vca import ( + DeployedComponent, + DeployedK8sResource, + DeployedVCA, + EELevel, + Relation, + EERelation, + safe_get_ee_relation, +) from osm_lcm.ng_ro import NgRoClient, NgRoException -from osm_lcm.lcm_utils import LcmException, LcmExceptionNoMgmtIP, LcmBase, deep_get, get_iterable, populate_dict -from osm_lcm.data_utils.nsd import get_vnf_profiles -from osm_lcm.data_utils.vnfd import get_vdu_list, get_vdu_profile, \ - get_ee_sorted_initial_config_primitive_list, get_ee_sorted_terminate_config_primitive_list, \ - get_kdu_list, get_virtual_link_profiles, get_vdu, get_configuration, \ - get_vdu_index, get_scaling_aspect, get_number_of_instances, get_juju_ee_ref +from osm_lcm.lcm_utils import ( + LcmException, + LcmBase, + deep_get, + get_iterable, + populate_dict, + check_juju_bundle_existence, + get_charm_artifact_path, + get_ee_id_parts, + vld_to_ro_ip_profile, +) +from osm_lcm.data_utils.nsd import ( + get_ns_configuration_relation_list, + get_vnf_profile, + get_vnf_profiles, +) +from osm_lcm.data_utils.vnfd import ( + get_kdu, + get_kdu_services, + get_relation_list, + get_vdu_list, + get_vdu_profile, + get_ee_sorted_initial_config_primitive_list, + get_ee_sorted_terminate_config_primitive_list, + get_kdu_list, + get_virtual_link_profiles, + get_vdu, + get_configuration, + get_vdu_index, + get_scaling_aspect, + get_number_of_instances, + get_juju_ee_ref, + get_kdu_resource_profile, + find_software_version, + check_helm_ee_in_ns, +) from osm_lcm.data_utils.list_utils import find_in_list -from osm_lcm.data_utils.vnfr import get_osm_params, get_vdur_index +from osm_lcm.data_utils.vnfr import ( + get_osm_params, + get_vdur_index, + get_kdur, + get_volumes_from_instantiation_params, +) from osm_lcm.data_utils.dict_utils import parse_yaml_strings from osm_lcm.data_utils.database.vim_account import VimAccountDB +from n2vc.definitions import RelationEndpoint from n2vc.k8s_helm_conn import K8sHelmConnector from n2vc.k8s_helm3_conn import K8sHelm3Connector from n2vc.k8s_juju_conn import K8sJujuConnector @@ -45,87 +106,84 @@ from osm_common.fsbase import FsException from osm_lcm.data_utils.database.database import Database from osm_lcm.data_utils.filesystem.filesystem import Filesystem +from osm_lcm.data_utils.wim import ( + get_sdn_ports, + get_target_wim_attrs, + select_feasible_wim_account, +) from n2vc.n2vc_juju_conn import N2VCJujuConnector from n2vc.exceptions import N2VCException, N2VCNotFound, K8sException from osm_lcm.lcm_helm_conn import LCMHelmConn +from osm_lcm.osm_config import OsmConfigBuilder +from osm_lcm.prometheus import parse_job from copy import copy, deepcopy from time import time from uuid import uuid4 -from random import randint +from random import SystemRandom __author__ = "Alfonso Tierno " class NsLcm(LcmBase): - timeout_vca_on_error = 5 * 60 # Time for charm from first time at blocked,error status to mark as failed - timeout_ns_deploy = 2 * 3600 # default global timeout for deployment a ns - timeout_ns_terminate = 1800 # default global timeout for un deployment a ns - timeout_charm_delete = 10 * 60 - timeout_primitive = 30 * 60 # timeout for primitive execution - timeout_progress_primitive = 10 * 60 # timeout for some progress in a primitive execution - SUBOPERATION_STATUS_NOT_FOUND = -1 SUBOPERATION_STATUS_NEW = -2 SUBOPERATION_STATUS_SKIP = -3 + EE_TLS_NAME = "ee-tls" task_name_deploy_vca = "Deploying VCA" - - def __init__(self, msg, lcm_tasks, config, loop, prometheus=None): + rel_operation_types = { + "GE": ">=", + "LE": "<=", + "GT": ">", + "LT": "<", + "EQ": "==", + "NE": "!=", + } + + def __init__(self, msg, lcm_tasks, config: LcmCfg): """ Init, Connect to database, filesystem storage, and messaging :param config: two level dictionary with configuration. Top level should contain 'database', 'storage', :return: None """ - super().__init__( - msg=msg, - logger=logging.getLogger('lcm.ns') - ) + super().__init__(msg=msg, logger=logging.getLogger("lcm.ns")) self.db = Database().instance.db self.fs = Filesystem().instance.fs - self.loop = loop self.lcm_tasks = lcm_tasks - self.timeout = config["timeout"] - self.ro_config = config["ro_config"] - self.ng_ro = config["ro_config"].get("ng") - self.vca_config = config["VCA"].copy() + self.timeout = config.timeout + self.ro_config = config.RO + self.vca_config = config.VCA # create N2VC connector self.n2vc = N2VCJujuConnector( log=self.logger, - loop=self.loop, - url='{}:{}'.format(self.vca_config['host'], self.vca_config['port']), - username=self.vca_config.get('user', None), - vca_config=self.vca_config, on_update_db=self._on_update_n2vc_db, fs=self.fs, - db=self.db + db=self.db, ) self.conn_helm_ee = LCMHelmConn( log=self.logger, - loop=self.loop, - url=None, - username=None, vca_config=self.vca_config, - on_update_db=self._on_update_n2vc_db + on_update_db=self._on_update_n2vc_db, ) self.k8sclusterhelm2 = K8sHelmConnector( - kubectl_command=self.vca_config.get("kubectlpath"), - helm_command=self.vca_config.get("helmpath"), + kubectl_command=self.vca_config.kubectlpath, + helm_command=self.vca_config.helmpath, log=self.logger, on_update_db=None, fs=self.fs, - db=self.db + db=self.db, ) self.k8sclusterhelm3 = K8sHelm3Connector( - kubectl_command=self.vca_config.get("kubectlpath"), - helm_command=self.vca_config.get("helm3path"), + kubectl_command=self.vca_config.kubectlpath, + helm_command=self.vca_config.helm3path, fs=self.fs, log=self.logger, db=self.db, @@ -133,14 +191,12 @@ class NsLcm(LcmBase): ) self.k8sclusterjuju = K8sJujuConnector( - kubectl_command=self.vca_config.get("kubectlpath"), - juju_command=self.vca_config.get("jujupath"), + kubectl_command=self.vca_config.kubectlpath, + juju_command=self.vca_config.jujupath, log=self.logger, - loop=self.loop, on_update_db=self._on_update_k8s_db, - vca_config=self.vca_config, fs=self.fs, - db=self.db + db=self.db, ) self.k8scluster_map = { @@ -156,13 +212,20 @@ class NsLcm(LcmBase): "native_charm": self.n2vc, "k8s_proxy_charm": self.n2vc, "helm": self.conn_helm_ee, - "helm-v3": self.conn_helm_ee + "helm-v3": self.conn_helm_ee, } - self.prometheus = prometheus - # create RO client - self.RO = NgRoClient(self.loop, **self.ro_config) + self.RO = NgRoClient(**self.ro_config.to_dict()) + + self.op_status_map = { + "instantiation": self.RO.status, + "termination": self.RO.status, + "migrate": self.RO.status, + "healing": self.RO.recreate_status, + "verticalscale": self.RO.status, + "start_stop_rebuild": self.RO.status, + } @staticmethod def increment_ip_mac(ip_mac, vm_index=1): @@ -179,107 +242,104 @@ class NsLcm(LcmBase): if i > 0: i += 1 # format in hex, len can be 2 for mac or 4 for ipv6 - return ("{}{:0" + str(len(ip_mac) - i) + "x}").format(ip_mac[:i], int(ip_mac[i:], 16) + vm_index) + return ("{}{:0" + str(len(ip_mac) - i) + "x}").format( + ip_mac[:i], int(ip_mac[i:], 16) + vm_index + ) except Exception: pass return None - def _on_update_ro_db(self, nsrs_id, ro_descriptor): - - # self.logger.debug('_on_update_ro_db(nsrs_id={}'.format(nsrs_id)) - - try: - # TODO filter RO descriptor fields... - - # write to database - db_dict = dict() - # db_dict['deploymentStatus'] = yaml.dump(ro_descriptor, default_flow_style=False, indent=2) - db_dict['deploymentStatus'] = ro_descriptor - self.update_db_2("nsrs", nsrs_id, db_dict) - - except Exception as e: - self.logger.warn('Cannot write database RO deployment for ns={} -> {}'.format(nsrs_id, e)) - - async def _on_update_n2vc_db(self, table, filter, path, updated_data): - + async def _on_update_n2vc_db(self, table, filter, path, updated_data, vca_id=None): # remove last dot from path (if exists) - if path.endswith('.'): + if path.endswith("."): path = path[:-1] # self.logger.debug('_on_update_n2vc_db(table={}, filter={}, path={}, updated_data={}' # .format(table, filter, path, updated_data)) try: - - nsr_id = filter.get('_id') + nsr_id = filter.get("_id") # read ns record from database - nsr = self.db.get_one(table='nsrs', q_filter=filter) - current_ns_status = nsr.get('nsState') + nsr = self.db.get_one(table="nsrs", q_filter=filter) + current_ns_status = nsr.get("nsState") # get vca status for NS - status_dict = await self.n2vc.get_status(namespace='.' + nsr_id, yaml_format=False) + status_dict = await self.n2vc.get_status( + namespace="." + nsr_id, yaml_format=False, vca_id=vca_id + ) # vcaStatus db_dict = dict() - db_dict['vcaStatus'] = status_dict - await self.n2vc.update_vca_status(db_dict['vcaStatus']) + db_dict["vcaStatus"] = status_dict # update configurationStatus for this VCA try: - vca_index = int(path[path.rfind(".")+1:]) + vca_index = int(path[path.rfind(".") + 1 :]) - vca_list = deep_get(target_dict=nsr, key_list=('_admin', 'deployed', 'VCA')) - vca_status = vca_list[vca_index].get('status') + vca_list = deep_get( + target_dict=nsr, key_list=("_admin", "deployed", "VCA") + ) + vca_status = vca_list[vca_index].get("status") - configuration_status_list = nsr.get('configurationStatus') - config_status = configuration_status_list[vca_index].get('status') + configuration_status_list = nsr.get("configurationStatus") + config_status = configuration_status_list[vca_index].get("status") - if config_status == 'BROKEN' and vca_status != 'failed': - db_dict['configurationStatus'][vca_index] = 'READY' - elif config_status != 'BROKEN' and vca_status == 'failed': - db_dict['configurationStatus'][vca_index] = 'BROKEN' + if config_status == "BROKEN" and vca_status != "failed": + db_dict["configurationStatus"][vca_index] = "READY" + elif config_status != "BROKEN" and vca_status == "failed": + db_dict["configurationStatus"][vca_index] = "BROKEN" except Exception as e: # not update configurationStatus - self.logger.debug('Error updating vca_index (ignore): {}'.format(e)) + self.logger.debug("Error updating vca_index (ignore): {}".format(e)) # if nsState = 'READY' check if juju is reporting some error => nsState = 'DEGRADED' # if nsState = 'DEGRADED' check if all is OK is_degraded = False - if current_ns_status in ('READY', 'DEGRADED'): - error_description = '' + if current_ns_status in ("READY", "DEGRADED"): + error_description = "" # check machines - if status_dict.get('machines'): - for machine_id in status_dict.get('machines'): - machine = status_dict.get('machines').get(machine_id) + if status_dict.get("machines"): + for machine_id in status_dict.get("machines"): + machine = status_dict.get("machines").get(machine_id) # check machine agent-status - if machine.get('agent-status'): - s = machine.get('agent-status').get('status') - if s != 'started': + if machine.get("agent-status"): + s = machine.get("agent-status").get("status") + if s != "started": is_degraded = True - error_description += 'machine {} agent-status={} ; '.format(machine_id, s) + error_description += ( + "machine {} agent-status={} ; ".format( + machine_id, s + ) + ) # check machine instance status - if machine.get('instance-status'): - s = machine.get('instance-status').get('status') - if s != 'running': + if machine.get("instance-status"): + s = machine.get("instance-status").get("status") + if s != "running": is_degraded = True - error_description += 'machine {} instance-status={} ; '.format(machine_id, s) + error_description += ( + "machine {} instance-status={} ; ".format( + machine_id, s + ) + ) # check applications - if status_dict.get('applications'): - for app_id in status_dict.get('applications'): - app = status_dict.get('applications').get(app_id) + if status_dict.get("applications"): + for app_id in status_dict.get("applications"): + app = status_dict.get("applications").get(app_id) # check application status - if app.get('status'): - s = app.get('status').get('status') - if s != 'active': + if app.get("status"): + s = app.get("status").get("status") + if s != "active": is_degraded = True - error_description += 'application {} status={} ; '.format(app_id, s) + error_description += ( + "application {} status={} ; ".format(app_id, s) + ) if error_description: - db_dict['errorDescription'] = error_description - if current_ns_status == 'READY' and is_degraded: - db_dict['nsState'] = 'DEGRADED' - if current_ns_status == 'DEGRADED' and not is_degraded: - db_dict['nsState'] = 'READY' + db_dict["errorDescription"] = error_description + if current_ns_status == "READY" and is_degraded: + db_dict["nsState"] = "DEGRADED" + if current_ns_status == "DEGRADED" and not is_degraded: + db_dict["nsState"] = "READY" # write to database self.update_db_2("nsrs", nsr_id, db_dict) @@ -287,63 +347,86 @@ class NsLcm(LcmBase): except (asyncio.CancelledError, asyncio.TimeoutError): raise except Exception as e: - self.logger.warn('Error updating NS state for ns={}: {}'.format(nsr_id, e)) + self.logger.warn("Error updating NS state for ns={}: {}".format(nsr_id, e)) - async def _on_update_k8s_db(self, cluster_uuid, kdu_instance, filter=None): + async def _on_update_k8s_db( + self, cluster_uuid, kdu_instance, filter=None, vca_id=None, cluster_type="juju" + ): """ Updating vca status in NSR record :param cluster_uuid: UUID of a k8s cluster :param kdu_instance: The unique name of the KDU instance :param filter: To get nsr_id + :cluster_type: The cluster type (juju, k8s) :return: none """ # self.logger.debug("_on_update_k8s_db(cluster_uuid={}, kdu_instance={}, filter={}" # .format(cluster_uuid, kdu_instance, filter)) + nsr_id = filter.get("_id") try: - nsr_id = filter.get('_id') + vca_status = await self.k8scluster_map[cluster_type].status_kdu( + cluster_uuid=cluster_uuid, + kdu_instance=kdu_instance, + yaml_format=False, + complete_status=True, + vca_id=vca_id, + ) - # get vca status for NS - vca_status = await self.k8sclusterjuju.status_kdu(cluster_uuid, - kdu_instance, - complete_status=True, - yaml_format=False) # vcaStatus db_dict = dict() - db_dict['vcaStatus'] = {nsr_id: vca_status} + db_dict["vcaStatus"] = {nsr_id: vca_status} - await self.k8sclusterjuju.update_vca_status(db_dict['vcaStatus'], kdu_instance) + self.logger.debug( + f"Obtained VCA status for cluster type '{cluster_type}': {vca_status}" + ) # write to database self.update_db_2("nsrs", nsr_id, db_dict) - except (asyncio.CancelledError, asyncio.TimeoutError): raise except Exception as e: - self.logger.warn('Error updating NS state for ns={}: {}'.format(nsr_id, e)) + self.logger.warn("Error updating NS state for ns={}: {}".format(nsr_id, e)) @staticmethod def _parse_cloud_init(cloud_init_text, additional_params, vnfd_id, vdu_id): try: - env = Environment(undefined=StrictUndefined) + env = Environment( + undefined=StrictUndefined, + autoescape=select_autoescape(default_for_string=True, default=True), + ) template = env.from_string(cloud_init_text) return template.render(additional_params or {}) except UndefinedError as e: - raise LcmException("Variable {} at vnfd[id={}]:vdu[id={}]:cloud-init/cloud-init-" - "file, must be provided in the instantiation parameters inside the " - "'additionalParamsForVnf/Vdu' block".format(e, vnfd_id, vdu_id)) + raise LcmException( + "Variable {} at vnfd[id={}]:vdu[id={}]:cloud-init/cloud-init-" + "file, must be provided in the instantiation parameters inside the " + "'additionalParamsForVnf/Vdu' block".format(e, vnfd_id, vdu_id) + ) except (TemplateError, TemplateNotFound) as e: - raise LcmException("Error parsing Jinja2 to cloud-init content at vnfd[id={}]:vdu[id={}]: {}". - format(vnfd_id, vdu_id, e)) + raise LcmException( + "Error parsing Jinja2 to cloud-init content at vnfd[id={}]:vdu[id={}]: {}".format( + vnfd_id, vdu_id, e + ) + ) def _get_vdu_cloud_init_content(self, vdu, vnfd): cloud_init_content = cloud_init_file = None try: if vdu.get("cloud-init-file"): base_folder = vnfd["_admin"]["storage"] - cloud_init_file = "{}/{}/cloud_init/{}".format(base_folder["folder"], base_folder["pkg-dir"], - vdu["cloud-init-file"]) + if base_folder["pkg-dir"]: + cloud_init_file = "{}/{}/cloud_init/{}".format( + base_folder["folder"], + base_folder["pkg-dir"], + vdu["cloud-init-file"], + ) + else: + cloud_init_file = "{}/Scripts/cloud_init/{}".format( + base_folder["folder"], + vdu["cloud-init-file"], + ) with self.fs.file_open(cloud_init_file, "r") as ci_file: cloud_init_content = ci_file.read() elif vdu.get("cloud-init"): @@ -351,40 +434,19 @@ class NsLcm(LcmBase): return cloud_init_content except FsException as e: - raise LcmException("Error reading vnfd[id={}]:vdu[id={}]:cloud-init-file={}: {}". - format(vnfd["id"], vdu["id"], cloud_init_file, e)) + raise LcmException( + "Error reading vnfd[id={}]:vdu[id={}]:cloud-init-file={}: {}".format( + vnfd["id"], vdu["id"], cloud_init_file, e + ) + ) def _get_vdu_additional_params(self, db_vnfr, vdu_id): - vdur = next(vdur for vdur in db_vnfr.get("vdur") if vdu_id == vdur["vdu-id-ref"]) + vdur = next( + (vdur for vdur in db_vnfr.get("vdur") if vdu_id == vdur["vdu-id-ref"]), {} + ) additional_params = vdur.get("additionalParams") return parse_yaml_strings(additional_params) - def vnfd2RO(self, vnfd, new_id=None, additionalParams=None, nsrId=None): - """ - Converts creates a new vnfd descriptor for RO base on input OSM IM vnfd - :param vnfd: input vnfd - :param new_id: overrides vnf id if provided - :param additionalParams: Instantiation params for VNFs provided - :param nsrId: Id of the NSR - :return: copy of vnfd - """ - vnfd_RO = deepcopy(vnfd) - # remove unused by RO configuration, monitoring, scaling and internal keys - vnfd_RO.pop("_id", None) - vnfd_RO.pop("_admin", None) - vnfd_RO.pop("monitoring-param", None) - vnfd_RO.pop("scaling-group-descriptor", None) - vnfd_RO.pop("kdu", None) - vnfd_RO.pop("k8s-cluster", None) - if new_id: - vnfd_RO["id"] = new_id - - # parse cloud-init or cloud-init-file with the provided variables using Jinja2 - for vdu in get_iterable(vnfd_RO, "vdu"): - vdu.pop("cloud-init-file", None) - vdu.pop("cloud-init", None) - return vnfd_RO - @staticmethod def ip_profile_2_RO(ip_profile): RO_ip_profile = deepcopy(ip_profile) @@ -392,7 +454,7 @@ class NsLcm(LcmBase): if isinstance(RO_ip_profile["dns-server"], list): RO_ip_profile["dns-address"] = [] for ds in RO_ip_profile.pop("dns-server"): - RO_ip_profile["dns-address"].append(ds['address']) + RO_ip_profile["dns-address"].append(ds["address"]) else: RO_ip_profile["dns-address"] = RO_ip_profile.pop("dns-server") if RO_ip_profile.get("ip-version") == "ipv4": @@ -403,68 +465,112 @@ class NsLcm(LcmBase): RO_ip_profile["dhcp"] = RO_ip_profile.pop("dhcp-params") return RO_ip_profile - def _get_ro_vim_id_for_vim_account(self, vim_account): - db_vim = self.db.get_one("vim_accounts", {"_id": vim_account}) - if db_vim["_admin"]["operationalState"] != "ENABLED": - raise LcmException("VIM={} is not available. operationalState={}".format( - vim_account, db_vim["_admin"]["operationalState"])) - RO_vim_id = db_vim["_admin"]["deployed"]["RO"] - return RO_vim_id - - def get_ro_wim_id_for_wim_account(self, wim_account): - if isinstance(wim_account, str): - db_wim = self.db.get_one("wim_accounts", {"_id": wim_account}) - if db_wim["_admin"]["operationalState"] != "ENABLED": - raise LcmException("WIM={} is not available. operationalState={}".format( - wim_account, db_wim["_admin"]["operationalState"])) - RO_wim_id = db_wim["_admin"]["deployed"]["RO-account"] - return RO_wim_id - else: - return wim_account - def scale_vnfr(self, db_vnfr, vdu_create=None, vdu_delete=None, mark_delete=False): - db_vdu_push_list = [] + template_vdur = [] db_update = {"_admin.modified": time()} if vdu_create: for vdu_id, vdu_count in vdu_create.items(): - vdur = next((vdur for vdur in reversed(db_vnfr["vdur"]) if vdur["vdu-id-ref"] == vdu_id), None) + vdur = next( + ( + vdur + for vdur in reversed(db_vnfr["vdur"]) + if vdur["vdu-id-ref"] == vdu_id + ), + None, + ) if not vdur: - raise LcmException("Error scaling OUT VNFR for {}. There is not any existing vnfr. Scaled to 0?". - format(vdu_id)) - + # Read the template saved in the db: + self.logger.debug( + "No vdur in the database. Using the vdur-template to scale" + ) + vdur_template = db_vnfr.get("vdur-template") + if not vdur_template: + raise LcmException( + "Error scaling OUT VNFR for {}. No vnfr or template exists".format( + vdu_id + ) + ) + vdur = vdur_template[0] + # Delete a template from the database after using it + self.db.set_one( + "vnfrs", + {"_id": db_vnfr["_id"]}, + None, + pull={"vdur-template": {"_id": vdur["_id"]}}, + ) for count in range(vdu_count): vdur_copy = deepcopy(vdur) vdur_copy["status"] = "BUILD" vdur_copy["status-detailed"] = None - vdur_copy["ip-address"]: None + vdur_copy["ip-address"] = None vdur_copy["_id"] = str(uuid4()) vdur_copy["count-index"] += count + 1 - vdur_copy["id"] = "{}-{}".format(vdur_copy["vdu-id-ref"], vdur_copy["count-index"]) + vdur_copy["id"] = "{}-{}".format( + vdur_copy["vdu-id-ref"], vdur_copy["count-index"] + ) vdur_copy.pop("vim_info", None) for iface in vdur_copy["interfaces"]: if iface.get("fixed-ip"): - iface["ip-address"] = self.increment_ip_mac(iface["ip-address"], count+1) + iface["ip-address"] = self.increment_ip_mac( + iface["ip-address"], count + 1 + ) else: iface.pop("ip-address", None) if iface.get("fixed-mac"): - iface["mac-address"] = self.increment_ip_mac(iface["mac-address"], count+1) + iface["mac-address"] = self.increment_ip_mac( + iface["mac-address"], count + 1 + ) else: iface.pop("mac-address", None) - iface.pop("mgmt_vnf", None) # only first vdu can be managment of vnf + if db_vnfr["vdur"]: + iface.pop( + "mgmt_vnf", None + ) # only first vdu can be managment of vnf db_vdu_push_list.append(vdur_copy) # self.logger.debug("scale out, adding vdu={}".format(vdur_copy)) if vdu_delete: + if len(db_vnfr["vdur"]) == 1: + # The scale will move to 0 instances + self.logger.debug( + "Scaling to 0 !, creating the template with the last vdur" + ) + template_vdur = [db_vnfr["vdur"][0]] for vdu_id, vdu_count in vdu_delete.items(): if mark_delete: - indexes_to_delete = [iv[0] for iv in enumerate(db_vnfr["vdur"]) if iv[1]["vdu-id-ref"] == vdu_id] - db_update.update({"vdur.{}.status".format(i): "DELETING" for i in indexes_to_delete[-vdu_count:]}) + indexes_to_delete = [ + iv[0] + for iv in enumerate(db_vnfr["vdur"]) + if iv[1]["vdu-id-ref"] == vdu_id + ] + db_update.update( + { + "vdur.{}.status".format(i): "DELETING" + for i in indexes_to_delete[-vdu_count:] + } + ) else: # it must be deleted one by one because common.db does not allow otherwise - vdus_to_delete = [v for v in reversed(db_vnfr["vdur"]) if v["vdu-id-ref"] == vdu_id] + vdus_to_delete = [ + v + for v in reversed(db_vnfr["vdur"]) + if v["vdu-id-ref"] == vdu_id + ] for vdu in vdus_to_delete[:vdu_count]: - self.db.set_one("vnfrs", {"_id": db_vnfr["_id"]}, None, pull={"vdur": {"_id": vdu["_id"]}}) - db_push = {"vdur": db_vdu_push_list} if db_vdu_push_list else None + self.db.set_one( + "vnfrs", + {"_id": db_vnfr["_id"]}, + None, + pull={"vdur": {"_id": vdu["_id"]}}, + ) + db_push = {} + if db_vdu_push_list: + db_push["vdur"] = db_vdu_push_list + if template_vdur: + db_push["vdur-template"] = template_vdur + if not db_push: + db_push = None + db_vnfr["vdur-template"] = template_vdur self.db.set_one("vnfrs", {"_id": db_vnfr["_id"]}, db_update, push_list=db_push) # modify passed dictionary db_vnfr db_vnfr_ = self.db.get_one("vnfrs", {"_id": db_vnfr["_id"]}) @@ -490,7 +596,9 @@ class NsLcm(LcmBase): ns_update_nsr["vld.{}".format(vld_index)] = vld break else: - raise LcmException("ns_update_nsr: Not found vld={} at RO info".format(vld["id"])) + raise LcmException( + "ns_update_nsr: Not found vld={} at RO info".format(vld["id"]) + ) def set_vnfr_at_error(self, db_vnfrs, error_text): try: @@ -502,84 +610,13 @@ class NsLcm(LcmBase): vnfr_update["vdur.{}.status".format(vdu_index)] = "ERROR" if error_text: vdur["status-detailed"] = str(error_text) - vnfr_update["vdur.{}.status-detailed".format(vdu_index)] = "ERROR" + vnfr_update[ + "vdur.{}.status-detailed".format(vdu_index) + ] = "ERROR" self.update_db_2("vnfrs", db_vnfr["_id"], vnfr_update) except DbException as e: self.logger.error("Cannot update vnf. {}".format(e)) - def ns_update_vnfr(self, db_vnfrs, nsr_desc_RO): - """ - Updates database vnfr with the RO info, e.g. ip_address, vim_id... Descriptor db_vnfrs is also updated - :param db_vnfrs: dictionary with member-vnf-index: vnfr-content - :param nsr_desc_RO: nsr descriptor from RO - :return: Nothing, LcmException is raised on errors - """ - for vnf_index, db_vnfr in db_vnfrs.items(): - for vnf_RO in nsr_desc_RO["vnfs"]: - if vnf_RO["member_vnf_index"] != vnf_index: - continue - vnfr_update = {} - if vnf_RO.get("ip_address"): - db_vnfr["ip-address"] = vnfr_update["ip-address"] = vnf_RO["ip_address"].split(";")[0] - elif not db_vnfr.get("ip-address"): - if db_vnfr.get("vdur"): # if not VDUs, there is not ip_address - raise LcmExceptionNoMgmtIP("ns member_vnf_index '{}' has no IP address".format(vnf_index)) - - for vdu_index, vdur in enumerate(get_iterable(db_vnfr, "vdur")): - vdur_RO_count_index = 0 - if vdur.get("pdu-type"): - continue - for vdur_RO in get_iterable(vnf_RO, "vms"): - if vdur["vdu-id-ref"] != vdur_RO["vdu_osm_id"]: - continue - if vdur["count-index"] != vdur_RO_count_index: - vdur_RO_count_index += 1 - continue - vdur["vim-id"] = vdur_RO.get("vim_vm_id") - if vdur_RO.get("ip_address"): - vdur["ip-address"] = vdur_RO["ip_address"].split(";")[0] - else: - vdur["ip-address"] = None - vdur["vdu-id-ref"] = vdur_RO.get("vdu_osm_id") - vdur["name"] = vdur_RO.get("vim_name") - vdur["status"] = vdur_RO.get("status") - vdur["status-detailed"] = vdur_RO.get("error_msg") - for ifacer in get_iterable(vdur, "interfaces"): - for interface_RO in get_iterable(vdur_RO, "interfaces"): - if ifacer["name"] == interface_RO.get("internal_name"): - ifacer["ip-address"] = interface_RO.get("ip_address") - ifacer["mac-address"] = interface_RO.get("mac_address") - break - else: - raise LcmException("ns_update_vnfr: Not found member_vnf_index={} vdur={} interface={} " - "from VIM info" - .format(vnf_index, vdur["vdu-id-ref"], ifacer["name"])) - vnfr_update["vdur.{}".format(vdu_index)] = vdur - break - else: - raise LcmException("ns_update_vnfr: Not found member_vnf_index={} vdur={} count_index={} from " - "VIM info".format(vnf_index, vdur["vdu-id-ref"], vdur["count-index"])) - - for vld_index, vld in enumerate(get_iterable(db_vnfr, "vld")): - for net_RO in get_iterable(nsr_desc_RO, "nets"): - if vld["id"] != net_RO.get("vnf_net_osm_id"): - continue - vld["vim-id"] = net_RO.get("vim_net_id") - vld["name"] = net_RO.get("vim_name") - vld["status"] = net_RO.get("status") - vld["status-detailed"] = net_RO.get("error_msg") - vnfr_update["vld.{}".format(vld_index)] = vld - break - else: - raise LcmException("ns_update_vnfr: Not found member_vnf_index={} vld={} from VIM info".format( - vnf_index, vld["id"])) - - self.update_db_2("vnfrs", db_vnfr["_id"], vnfr_update) - break - - else: - raise LcmException("ns_update_vnfr: Not found member_vnf_index={} from VIM info".format(vnf_index)) - def _get_ns_config_info(self, nsr_id): """ Generates a mapping between vnf,vdu elements and the N2VC id @@ -598,13 +635,27 @@ class NsLcm(LcmBase): if not vca["vdu_id"]: mapping[vca["member-vnf-index"]] = vca["application"] else: - mapping["{}.{}.{}".format(vca["member-vnf-index"], vca["vdu_id"], vca["vdu_count_index"])] =\ - vca["application"] + mapping[ + "{}.{}.{}".format( + vca["member-vnf-index"], vca["vdu_id"], vca["vdu_count_index"] + ) + ] = vca["application"] return ns_config_info - async def _instantiate_ng_ro(self, logging_text, nsr_id, nsd, db_nsr, db_nslcmop, db_vnfrs, db_vnfds, - n2vc_key_list, stage, start_deploy, timeout_ns_deploy): - + async def _instantiate_ng_ro( + self, + logging_text, + nsr_id, + nsd, + db_nsr, + db_nslcmop, + db_vnfrs, + db_vnfds, + n2vc_key_list, + stage, + start_deploy, + timeout_ns_deploy, + ): db_vims = {} def get_vim_account(vim_account_id): @@ -616,27 +667,131 @@ class NsLcm(LcmBase): return db_vim # modify target_vld info with instantiation parameters - def parse_vld_instantiation_params(target_vim, target_vld, vld_params, target_sdn): + def parse_vld_instantiation_params( + target_vim, target_vld, vld_params, target_sdn + ): if vld_params.get("ip-profile"): - target_vld["vim_info"][target_vim]["ip_profile"] = vld_params["ip-profile"] + target_vld["vim_info"][target_vim]["ip_profile"] = vld_to_ro_ip_profile( + vld_params["ip-profile"] + ) if vld_params.get("provider-network"): - target_vld["vim_info"][target_vim]["provider_network"] = vld_params["provider-network"] + target_vld["vim_info"][target_vim]["provider_network"] = vld_params[ + "provider-network" + ] if "sdn-ports" in vld_params["provider-network"] and target_sdn: - target_vld["vim_info"][target_sdn]["sdn-ports"] = vld_params["provider-network"]["sdn-ports"] - if vld_params.get("wimAccountId"): - target_wim = "wim:{}".format(vld_params["wimAccountId"]) - target_vld["vim_info"][target_wim] = {} + target_vld["vim_info"][target_sdn]["sdn-ports"] = vld_params[ + "provider-network" + ]["sdn-ports"] + + # check if WIM is needed; if needed, choose a feasible WIM able to connect VIMs + # if wim_account_id is specified in vld_params, validate if it is feasible. + wim_account_id, db_wim = select_feasible_wim_account( + db_nsr, db_vnfrs, target_vld, vld_params, self.logger + ) + + if wim_account_id: + # WIM is needed and a feasible one was found, populate WIM target and SDN ports + self.logger.info("WIM selected: {:s}".format(str(wim_account_id))) + # update vld_params with correct WIM account Id + vld_params["wimAccountId"] = wim_account_id + + target_wim = "wim:{}".format(wim_account_id) + target_wim_attrs = get_target_wim_attrs(nsr_id, target_vld, vld_params) + sdn_ports = get_sdn_ports(vld_params, db_wim) + if len(sdn_ports) > 0: + target_vld["vim_info"][target_wim] = target_wim_attrs + target_vld["vim_info"][target_wim]["sdn-ports"] = sdn_ports + + self.logger.debug( + "Target VLD with WIM data: {:s}".format(str(target_vld)) + ) + for param in ("vim-network-name", "vim-network-id"): if vld_params.get(param): if isinstance(vld_params[param], dict): for vim, vim_net in vld_params[param].items(): other_target_vim = "vim:" + vim - populate_dict(target_vld["vim_info"], (other_target_vim, param.replace("-", "_")), vim_net) + populate_dict( + target_vld["vim_info"], + (other_target_vim, param.replace("-", "_")), + vim_net, + ) else: # isinstance str - target_vld["vim_info"][target_vim][param.replace("-", "_")] = vld_params[param] + target_vld["vim_info"][target_vim][ + param.replace("-", "_") + ] = vld_params[param] if vld_params.get("common_id"): target_vld["common_id"] = vld_params.get("common_id") + # modify target["ns"]["vld"] with instantiation parameters to override vnf vim-account + def update_ns_vld_target(target, ns_params): + for vnf_params in ns_params.get("vnf", ()): + if vnf_params.get("vimAccountId"): + target_vnf = next( + ( + vnfr + for vnfr in db_vnfrs.values() + if vnf_params["member-vnf-index"] + == vnfr["member-vnf-index-ref"] + ), + None, + ) + vdur = next((vdur for vdur in target_vnf.get("vdur", ())), None) + if not vdur: + continue + for a_index, a_vld in enumerate(target["ns"]["vld"]): + target_vld = find_in_list( + get_iterable(vdur, "interfaces"), + lambda iface: iface.get("ns-vld-id") == a_vld["name"], + ) + + vld_params = find_in_list( + get_iterable(ns_params, "vld"), + lambda v_vld: v_vld["name"] in (a_vld["name"], a_vld["id"]), + ) + if target_vld: + if vnf_params.get("vimAccountId") not in a_vld.get( + "vim_info", {} + ): + target_vim_network_list = [ + v for _, v in a_vld.get("vim_info").items() + ] + target_vim_network_name = next( + ( + item.get("vim_network_name", "") + for item in target_vim_network_list + ), + "", + ) + + target["ns"]["vld"][a_index].get("vim_info").update( + { + "vim:{}".format(vnf_params["vimAccountId"]): { + "vim_network_name": target_vim_network_name, + } + } + ) + + if vld_params: + for param in ("vim-network-name", "vim-network-id"): + if vld_params.get(param) and isinstance( + vld_params[param], dict + ): + for vim, vim_net in vld_params[ + param + ].items(): + other_target_vim = "vim:" + vim + populate_dict( + target["ns"]["vld"][a_index].get( + "vim_info" + ), + ( + other_target_vim, + param.replace("-", "_"), + ), + vim_net, + ) + nslcmop_id = db_nslcmop["_id"] target = { "name": db_nsr["name"], @@ -651,11 +806,28 @@ class NsLcm(LcmBase): image["vim_info"] = {} for flavor in target["flavor"]: flavor["vim_info"] = {} + if db_nsr.get("shared-volumes"): + target["shared-volumes"] = deepcopy(db_nsr["shared-volumes"]) + for shared_volumes in target["shared-volumes"]: + shared_volumes["vim_info"] = {} + if db_nsr.get("affinity-or-anti-affinity-group"): + target["affinity-or-anti-affinity-group"] = deepcopy( + db_nsr["affinity-or-anti-affinity-group"] + ) + for affinity_or_anti_affinity_group in target[ + "affinity-or-anti-affinity-group" + ]: + affinity_or_anti_affinity_group["vim_info"] = {} if db_nslcmop.get("lcmOperationType") != "instantiate": # get parameters of instantiation: - db_nslcmop_instantiate = self.db.get_list("nslcmops", {"nsInstanceId": db_nslcmop["nsInstanceId"], - "lcmOperationType": "instantiate"})[-1] + db_nslcmop_instantiate = self.db.get_list( + "nslcmops", + { + "nsInstanceId": db_nslcmop["nsInstanceId"], + "lcmOperationType": "instantiate", + }, + )[-1] ns_params = db_nslcmop_instantiate.get("operationParams") else: ns_params = db_nslcmop.get("operationParams") @@ -673,74 +845,91 @@ class NsLcm(LcmBase): "vim_info": { target_vim: { "vim_network_name": vld.get("vim-network-name"), - "vim_account_id": ns_params["vimAccountId"] + "vim_account_id": ns_params["vimAccountId"], } - } + }, } # check if this network needs SDN assist if vld.get("pci-interfaces"): db_vim = get_vim_account(ns_params["vimAccountId"]) - sdnc_id = db_vim["config"].get("sdn-controller") - if sdnc_id: - sdn_vld = "nsrs:{}:vld.{}".format(nsr_id, vld["id"]) - target_sdn = "sdn:{}".format(sdnc_id) - target_vld["vim_info"][target_sdn] = { - "sdn": True, "target_vim": target_vim, "vlds": [sdn_vld], "type": vld.get("type")} + if vim_config := db_vim.get("config"): + if sdnc_id := vim_config.get("sdn-controller"): + sdn_vld = "nsrs:{}:vld.{}".format(nsr_id, vld["id"]) + target_sdn = "sdn:{}".format(sdnc_id) + target_vld["vim_info"][target_sdn] = { + "sdn": True, + "target_vim": target_vim, + "vlds": [sdn_vld], + "type": vld.get("type"), + } nsd_vnf_profiles = get_vnf_profiles(nsd) for nsd_vnf_profile in nsd_vnf_profiles: for cp in nsd_vnf_profile["virtual-link-connectivity"]: if cp["virtual-link-profile-id"] == vld["id"]: - cp2target["member_vnf:{}.{}".format( - cp["constituent-cpd-id"][0]["constituent-base-element-id"], - cp["constituent-cpd-id"][0]["constituent-cpd-id"] - )] = "nsrs:{}:vld.{}".format(nsr_id, vld_index) + cp2target[ + "member_vnf:{}.{}".format( + cp["constituent-cpd-id"][0][ + "constituent-base-element-id" + ], + cp["constituent-cpd-id"][0]["constituent-cpd-id"], + ) + ] = "nsrs:{}:vld.{}".format(nsr_id, vld_index) # check at nsd descriptor, if there is an ip-profile vld_params = {} nsd_vlp = find_in_list( get_virtual_link_profiles(nsd), - lambda a_link_profile: a_link_profile["virtual-link-desc-id"] == vld["id"]) - if nsd_vlp and nsd_vlp.get("virtual-link-protocol-data") and \ - nsd_vlp["virtual-link-protocol-data"].get("l3-protocol-data"): - ip_profile_source_data = nsd_vlp["virtual-link-protocol-data"]["l3-protocol-data"] - ip_profile_dest_data = {} - if "ip-version" in ip_profile_source_data: - ip_profile_dest_data["ip-version"] = ip_profile_source_data["ip-version"] - if "cidr" in ip_profile_source_data: - ip_profile_dest_data["subnet-address"] = ip_profile_source_data["cidr"] - if "gateway-ip" in ip_profile_source_data: - ip_profile_dest_data["gateway-address"] = ip_profile_source_data["gateway-ip"] - if "dhcp-enabled" in ip_profile_source_data: - ip_profile_dest_data["dhcp-params"] = { - "enabled": ip_profile_source_data["dhcp-enabled"] - } - vld_params["ip-profile"] = ip_profile_dest_data + lambda a_link_profile: a_link_profile["virtual-link-desc-id"] + == vld["id"], + ) + if ( + nsd_vlp + and nsd_vlp.get("virtual-link-protocol-data") + and nsd_vlp["virtual-link-protocol-data"].get("l3-protocol-data") + ): + vld_params["ip-profile"] = nsd_vlp["virtual-link-protocol-data"][ + "l3-protocol-data" + ] # update vld_params with instantiation params - vld_instantiation_params = find_in_list(get_iterable(ns_params, "vld"), - lambda a_vld: a_vld["name"] in (vld["name"], vld["id"])) + vld_instantiation_params = find_in_list( + get_iterable(ns_params, "vld"), + lambda a_vld: a_vld["name"] in (vld["name"], vld["id"]), + ) if vld_instantiation_params: vld_params.update(vld_instantiation_params) parse_vld_instantiation_params(target_vim, target_vld, vld_params, None) target["ns"]["vld"].append(target_vld) + # Update the target ns_vld if vnf vim_account is overriden by instantiation params + update_ns_vld_target(target, ns_params) for vnfr in db_vnfrs.values(): - vnfd = find_in_list(db_vnfds, lambda db_vnf: db_vnf["id"] == vnfr["vnfd-ref"]) - vnf_params = find_in_list(get_iterable(ns_params, "vnf"), - lambda a_vnf: a_vnf["member-vnf-index"] == vnfr["member-vnf-index-ref"]) + vnfd = find_in_list( + db_vnfds, lambda db_vnf: db_vnf["id"] == vnfr["vnfd-ref"] + ) + vnf_params = find_in_list( + get_iterable(ns_params, "vnf"), + lambda a_vnf: a_vnf["member-vnf-index"] == vnfr["member-vnf-index-ref"], + ) target_vnf = deepcopy(vnfr) target_vim = "vim:{}".format(vnfr["vim-account-id"]) for vld in target_vnf.get("vld", ()): # check if connected to a ns.vld, to fill target' - vnf_cp = find_in_list(vnfd.get("int-virtual-link-desc", ()), - lambda cpd: cpd.get("id") == vld["id"]) + vnf_cp = find_in_list( + vnfd.get("int-virtual-link-desc", ()), + lambda cpd: cpd.get("id") == vld["id"], + ) if vnf_cp: - ns_cp = "member_vnf:{}.{}".format(vnfr["member-vnf-index-ref"], vnf_cp["id"]) + ns_cp = "member_vnf:{}.{}".format( + vnfr["member-vnf-index-ref"], vnf_cp["id"] + ) if cp2target.get(ns_cp): vld["target"] = cp2target[ns_cp] - vld["vim_info"] = {target_vim: {"vim_network_name": vld.get("vim-network-name")}} + vld["vim_info"] = { + target_vim: {"vim_network_name": vld.get("vim-network-name")} + } # check if this network needs SDN assist target_sdn = None if vld.get("pci-interfaces"): @@ -750,34 +939,32 @@ class NsLcm(LcmBase): sdn_vld = "vnfrs:{}:vld.{}".format(target_vnf["_id"], vld["id"]) target_sdn = "sdn:{}".format(sdnc_id) vld["vim_info"][target_sdn] = { - "sdn": True, "target_vim": target_vim, "vlds": [sdn_vld], "type": vld.get("type")} + "sdn": True, + "target_vim": target_vim, + "vlds": [sdn_vld], + "type": vld.get("type"), + } # check at vnfd descriptor, if there is an ip-profile vld_params = {} vnfd_vlp = find_in_list( get_virtual_link_profiles(vnfd), - lambda a_link_profile: a_link_profile["id"] == vld["id"] - ) - if vnfd_vlp and vnfd_vlp.get("virtual-link-protocol-data") and \ - vnfd_vlp["virtual-link-protocol-data"].get("l3-protocol-data"): - ip_profile_source_data = vnfd_vlp["virtual-link-protocol-data"]["l3-protocol-data"] - ip_profile_dest_data = {} - if "ip-version" in ip_profile_source_data: - ip_profile_dest_data["ip-version"] = ip_profile_source_data["ip-version"] - if "cidr" in ip_profile_source_data: - ip_profile_dest_data["subnet-address"] = ip_profile_source_data["cidr"] - if "gateway-ip" in ip_profile_source_data: - ip_profile_dest_data["gateway-address"] = ip_profile_source_data["gateway-ip"] - if "dhcp-enabled" in ip_profile_source_data: - ip_profile_dest_data["dhcp-params"] = { - "enabled": ip_profile_source_data["dhcp-enabled"] - } - - vld_params["ip-profile"] = ip_profile_dest_data + lambda a_link_profile: a_link_profile["id"] == vld["id"], + ) + if ( + vnfd_vlp + and vnfd_vlp.get("virtual-link-protocol-data") + and vnfd_vlp["virtual-link-protocol-data"].get("l3-protocol-data") + ): + vld_params["ip-profile"] = vnfd_vlp["virtual-link-protocol-data"][ + "l3-protocol-data" + ] # update vld_params with instantiation params if vnf_params: - vld_instantiation_params = find_in_list(get_iterable(vnf_params, "internal-vld"), - lambda i_vld: i_vld["name"] == vld["id"]) + vld_instantiation_params = find_in_list( + get_iterable(vnf_params, "internal-vld"), + lambda i_vld: i_vld["name"] == vld["id"], + ) if vld_instantiation_params: vld_params.update(vld_instantiation_params) parse_vld_instantiation_params(target_vim, vld, vld_params, target_sdn) @@ -793,17 +980,28 @@ class NsLcm(LcmBase): if ssh_keys_all: vdu_configuration = get_configuration(vnfd, vdur["vdu-id-ref"]) vnf_configuration = get_configuration(vnfd, vnfd["id"]) - if vdu_configuration and vdu_configuration.get("config-access") and \ - vdu_configuration.get("config-access").get("ssh-access"): + if ( + vdu_configuration + and vdu_configuration.get("config-access") + and vdu_configuration.get("config-access").get("ssh-access") + ): vdur["ssh-keys"] = ssh_keys_all - vdur["ssh-access-required"] = vdu_configuration["config-access"]["ssh-access"]["required"] - elif vnf_configuration and vnf_configuration.get("config-access") and \ - vnf_configuration.get("config-access").get("ssh-access") and \ - any(iface.get("mgmt-vnf") for iface in vdur["interfaces"]): + vdur["ssh-access-required"] = vdu_configuration[ + "config-access" + ]["ssh-access"]["required"] + elif ( + vnf_configuration + and vnf_configuration.get("config-access") + and vnf_configuration.get("config-access").get("ssh-access") + and any(iface.get("mgmt-vnf") for iface in vdur["interfaces"]) + ): vdur["ssh-keys"] = ssh_keys_all - vdur["ssh-access-required"] = vnf_configuration["config-access"]["ssh-access"]["required"] - elif ssh_keys_instantiation and \ - find_in_list(vdur["interfaces"], lambda iface: iface.get("mgmt-vnf")): + vdur["ssh-access-required"] = vnf_configuration[ + "config-access" + ]["ssh-access"]["required"] + elif ssh_keys_instantiation and find_in_list( + vdur["interfaces"], lambda iface: iface.get("mgmt-vnf") + ): vdur["ssh-keys"] = ssh_keys_instantiation self.logger.debug("NS > vdur > {}".format(vdur)) @@ -811,21 +1009,42 @@ class NsLcm(LcmBase): vdud = get_vdu(vnfd, vdur["vdu-id-ref"]) # cloud-init if vdud.get("cloud-init-file"): - vdur["cloud-init"] = "{}:file:{}".format(vnfd["_id"], vdud.get("cloud-init-file")) + vdur["cloud-init"] = "{}:file:{}".format( + vnfd["_id"], vdud.get("cloud-init-file") + ) # read file and put content at target.cloul_init_content. Avoid ng_ro to use shared package system if vdur["cloud-init"] not in target["cloud_init_content"]: base_folder = vnfd["_admin"]["storage"] - cloud_init_file = "{}/{}/cloud_init/{}".format(base_folder["folder"], base_folder["pkg-dir"], - vdud.get("cloud-init-file")) + if base_folder["pkg-dir"]: + cloud_init_file = "{}/{}/cloud_init/{}".format( + base_folder["folder"], + base_folder["pkg-dir"], + vdud.get("cloud-init-file"), + ) + else: + cloud_init_file = "{}/Scripts/cloud_init/{}".format( + base_folder["folder"], + vdud.get("cloud-init-file"), + ) with self.fs.file_open(cloud_init_file, "r") as ci_file: - target["cloud_init_content"][vdur["cloud-init"]] = ci_file.read() + target["cloud_init_content"][ + vdur["cloud-init"] + ] = ci_file.read() elif vdud.get("cloud-init"): - vdur["cloud-init"] = "{}:vdu:{}".format(vnfd["_id"], get_vdu_index(vnfd, vdur["vdu-id-ref"])) + vdur["cloud-init"] = "{}:vdu:{}".format( + vnfd["_id"], get_vdu_index(vnfd, vdur["vdu-id-ref"]) + ) # put content at target.cloul_init_content. Avoid ng_ro read vnfd descriptor - target["cloud_init_content"][vdur["cloud-init"]] = vdud["cloud-init"] + target["cloud_init_content"][vdur["cloud-init"]] = vdud[ + "cloud-init" + ] vdur["additionalParams"] = vdur.get("additionalParams") or {} - deploy_params_vdu = self._format_additional_params(vdur.get("additionalParams") or {}) - deploy_params_vdu["OSM"] = get_osm_params(vnfr, vdur["vdu-id-ref"], vdur["count-index"]) + deploy_params_vdu = self._format_additional_params( + vdur.get("additionalParams") or {} + ) + deploy_params_vdu["OSM"] = get_osm_params( + vnfr, vdur["vdu-id-ref"], vdur["count-index"] + ) vdur["additionalParams"] = deploy_params_vdu # flavor @@ -844,7 +1063,9 @@ class NsLcm(LcmBase): ns_alt_image = target["image"][int(alt_image_id)] if vim_type == ns_alt_image.get("vim-type"): # must use alternative image - self.logger.debug("use alternative image id: {}".format(alt_image_id)) + self.logger.debug( + "use alternative image id: {}".format(alt_image_id) + ) ns_image_id = alt_image_id vdur["ns-image-id"] = ns_image_id break @@ -852,37 +1073,84 @@ class NsLcm(LcmBase): if target_vim not in ns_image["vim_info"]: ns_image["vim_info"][target_vim] = {} + # Affinity groups + if vdur.get("affinity-or-anti-affinity-group-id"): + for ags_id in vdur["affinity-or-anti-affinity-group-id"]: + ns_ags = target["affinity-or-anti-affinity-group"][int(ags_id)] + if target_vim not in ns_ags["vim_info"]: + ns_ags["vim_info"][target_vim] = {} + + # shared-volumes + if vdur.get("shared-volumes-id"): + for sv_id in vdur["shared-volumes-id"]: + ns_sv = find_in_list( + target["shared-volumes"], lambda sv: sv_id in sv["id"] + ) + if ns_sv: + ns_sv["vim_info"][target_vim] = {} + vdur["vim_info"] = {target_vim: {}} # instantiation parameters - # if vnf_params: - # vdu_instantiation_params = next((v for v in get_iterable(vnf_params, "vdu") if v["id"] == - # vdud["id"]), None) + if vnf_params: + vdu_instantiation_params = find_in_list( + get_iterable(vnf_params, "vdu"), + lambda i_vdu: i_vdu["id"] == vdud["id"], + ) + if vdu_instantiation_params: + # Parse the vdu_volumes from the instantiation params + vdu_volumes = get_volumes_from_instantiation_params( + vdu_instantiation_params, vdud + ) + vdur["additionalParams"]["OSM"]["vdu_volumes"] = vdu_volumes + vdur["additionalParams"]["OSM"][ + "vim_flavor_id" + ] = vdu_instantiation_params.get("vim-flavor-id") vdur_list.append(vdur) target_vnf["vdur"] = vdur_list target["vnf"].append(target_vnf) + self.logger.debug("Send to RO > nsr_id={} target={}".format(nsr_id, target)) desc = await self.RO.deploy(nsr_id, target) self.logger.debug("RO return > {}".format(desc)) action_id = desc["action_id"] - await self._wait_ng_ro(nsr_id, action_id, nslcmop_id, start_deploy, timeout_ns_deploy, stage) + await self._wait_ng_ro( + nsr_id, + action_id, + nslcmop_id, + start_deploy, + timeout_ns_deploy, + stage, + operation="instantiation", + ) # Updating NSR db_nsr_update = { "_admin.deployed.RO.operational-status": "running", - "detailed-status": " ".join(stage) + "detailed-status": " ".join(stage), } # db_nsr["_admin.deployed.RO.detailed-status"] = "Deployed at VIM" self.update_db_2("nsrs", nsr_id, db_nsr_update) self._write_op_status(nslcmop_id, stage) - self.logger.debug(logging_text + "ns deployed at RO. RO_id={}".format(action_id)) + self.logger.debug( + logging_text + "ns deployed at RO. RO_id={}".format(action_id) + ) return - async def _wait_ng_ro(self, nsr_id, action_id, nslcmop_id=None, start_time=None, timeout=600, stage=None): + async def _wait_ng_ro( + self, + nsr_id, + action_id, + nslcmop_id=None, + start_time=None, + timeout=600, + stage=None, + operation=None, + ): detailed_status_old = None db_nsr_update = {} start_time = start_time or time() while time() <= start_time + timeout: - desc_status = await self.RO.status(nsr_id, action_id) + desc_status = await self.op_status_map[operation](nsr_id, action_id) self.logger.debug("Wait NG RO > {}".format(desc_status)) if desc_status["status"] == "FAILED": raise NgRoException(desc_status["details"]) @@ -894,17 +1162,21 @@ class NsLcm(LcmBase): stage[2] = "Deployed at VIM" break else: - assert False, "ROclient.check_ns_status returns unknown {}".format(desc_status["status"]) + assert False, "ROclient.check_ns_status returns unknown {}".format( + desc_status["status"] + ) if stage and nslcmop_id and stage[2] != detailed_status_old: detailed_status_old = stage[2] db_nsr_update["detailed-status"] = " ".join(stage) self.update_db_2("nsrs", nsr_id, db_nsr_update) self._write_op_status(nslcmop_id, stage) - await asyncio.sleep(15, loop=self.loop) + await asyncio.sleep(15) else: # timeout_ns_deploy raise NgRoException("Timeout waiting ns to deploy") - async def _terminate_ng_ro(self, logging_text, nsr_deployed, nsr_id, nslcmop_id, stage): + async def _terminate_ng_ro( + self, logging_text, nsr_deployed, nsr_id, nslcmop_id, stage + ): db_nsr_update = {} failed_detail = [] action_id = None @@ -915,34 +1187,54 @@ class NsLcm(LcmBase): "vnf": [], "image": [], "flavor": [], - "action_id": nslcmop_id + "action_id": nslcmop_id, } desc = await self.RO.deploy(nsr_id, target) action_id = desc["action_id"] - db_nsr_update["_admin.deployed.RO.nsr_delete_action_id"] = action_id db_nsr_update["_admin.deployed.RO.nsr_status"] = "DELETING" - self.logger.debug(logging_text + "ns terminate action at RO. action_id={}".format(action_id)) + self.logger.debug( + logging_text + + "ns terminate action at RO. action_id={}".format(action_id) + ) # wait until done delete_timeout = 20 * 60 # 20 minutes - await self._wait_ng_ro(nsr_id, action_id, nslcmop_id, start_deploy, delete_timeout, stage) - - db_nsr_update["_admin.deployed.RO.nsr_delete_action_id"] = None + await self._wait_ng_ro( + nsr_id, + action_id, + nslcmop_id, + start_deploy, + delete_timeout, + stage, + operation="termination", + ) db_nsr_update["_admin.deployed.RO.nsr_status"] = "DELETED" # delete all nsr await self.RO.delete(nsr_id) - except Exception as e: - if isinstance(e, NgRoException) and e.http_code == 404: # not found + except NgRoException as e: + if e.http_code == 404: # not found db_nsr_update["_admin.deployed.RO.nsr_id"] = None db_nsr_update["_admin.deployed.RO.nsr_status"] = "DELETED" - db_nsr_update["_admin.deployed.RO.nsr_delete_action_id"] = None - self.logger.debug(logging_text + "RO_action_id={} already deleted".format(action_id)) - elif isinstance(e, NgRoException) and e.http_code == 409: # conflict + self.logger.debug( + logging_text + "RO_action_id={} already deleted".format(action_id) + ) + elif e.http_code == 409: # conflict failed_detail.append("delete conflict: {}".format(e)) - self.logger.debug(logging_text + "RO_action_id={} delete conflict: {}".format(action_id, e)) + self.logger.debug( + logging_text + + "RO_action_id={} delete conflict: {}".format(action_id, e) + ) else: failed_detail.append("delete error: {}".format(e)) - self.logger.error(logging_text + "RO_action_id={} delete error: {}".format(action_id, e)) + self.logger.error( + logging_text + + "RO_action_id={} delete error: {}".format(action_id, e) + ) + except Exception as e: + failed_detail.append("delete error: {}".format(e)) + self.logger.error( + logging_text + "RO_action_id={} delete error: {}".format(action_id, e) + ) if failed_detail: stage[2] = "Error deleting from VIM" @@ -956,8 +1248,18 @@ class NsLcm(LcmBase): raise LcmException("; ".join(failed_detail)) return - async def instantiate_RO(self, logging_text, nsr_id, nsd, db_nsr, db_nslcmop, db_vnfrs, db_vnfds, - n2vc_key_list, stage): + async def instantiate_RO( + self, + logging_text, + nsr_id, + nsd, + db_nsr, + db_nslcmop, + db_vnfrs, + db_vnfds, + n2vc_key_list, + stage, + ): """ Instantiate at RO :param logging_text: preffix text to use at logging @@ -977,7 +1279,7 @@ class NsLcm(LcmBase): if ns_params and ns_params.get("timeout_ns_deploy"): timeout_ns_deploy = ns_params["timeout_ns_deploy"] else: - timeout_ns_deploy = self.timeout.get("ns_deploy", self.timeout_ns_deploy) + timeout_ns_deploy = self.timeout.ns_deploy # Check for and optionally request placement optimization. Database will be updated if placement activated stage[2] = "Waiting for Placement." @@ -989,14 +1291,34 @@ class NsLcm(LcmBase): else: ns_params["vimAccountId"] == vnfr["vim-account-id"] - return await self._instantiate_ng_ro(logging_text, nsr_id, nsd, db_nsr, db_nslcmop, db_vnfrs, - db_vnfds, n2vc_key_list, stage, start_deploy, timeout_ns_deploy) + return await self._instantiate_ng_ro( + logging_text, + nsr_id, + nsd, + db_nsr, + db_nslcmop, + db_vnfrs, + db_vnfds, + n2vc_key_list, + stage, + start_deploy, + timeout_ns_deploy, + ) except Exception as e: stage[2] = "ERROR deploying at VIM" self.set_vnfr_at_error(db_vnfrs, str(e)) - self.logger.error("Error deploying at VIM {}".format(e), - exc_info=not isinstance(e, (ROclient.ROClientException, LcmException, DbException, - NgRoException))) + self.logger.error( + "Error deploying at VIM {}".format(e), + exc_info=not isinstance( + e, + ( + ROclient.ROClientException, + LcmException, + DbException, + NgRoException, + ), + ), + ) raise async def wait_kdu_up(self, logging_text, nsr_id, vnfr_id, kdu_name): @@ -1006,7 +1328,7 @@ class NsLcm(LcmBase): :param nsr_id: :param vnfr_id: :param kdu_name: - :return: IP address + :return: IP address, K8s services """ # self.logger.debug(logging_text + "Starting wait_kdu_up") @@ -1014,20 +1336,33 @@ class NsLcm(LcmBase): while nb_tries < 360: db_vnfr = self.db.get_one("vnfrs", {"_id": vnfr_id}) - kdur = next((x for x in get_iterable(db_vnfr, "kdur") if x.get("kdu-name") == kdu_name), None) + kdur = next( + ( + x + for x in get_iterable(db_vnfr, "kdur") + if x.get("kdu-name") == kdu_name + ), + None, + ) if not kdur: - raise LcmException("Not found vnfr_id={}, kdu_name={}".format(vnfr_id, kdu_name)) + raise LcmException( + "Not found vnfr_id={}, kdu_name={}".format(vnfr_id, kdu_name) + ) if kdur.get("status"): if kdur["status"] in ("READY", "ENABLED"): - return kdur.get("ip-address") + return kdur.get("ip-address"), kdur.get("services") else: - raise LcmException("target KDU={} is in error state".format(kdu_name)) + raise LcmException( + "target KDU={} is in error state".format(kdu_name) + ) - await asyncio.sleep(10, loop=self.loop) + await asyncio.sleep(10) nb_tries += 1 raise LcmException("Timeout waiting KDU={} instantiated".format(kdu_name)) - async def wait_vm_up_insert_key_ro(self, logging_text, nsr_id, vnfr_id, vdu_id, vdu_index, pub_key=None, user=None): + async def wait_vm_up_insert_key_ro( + self, logging_text, nsr_id, vnfr_id, vdu_id, vdu_index, pub_key=None, user=None + ): """ Wait for ip addres at RO, and optionally, insert public key in virtual machine :param logging_text: prefix use for logging @@ -1041,19 +1376,18 @@ class NsLcm(LcmBase): """ self.logger.debug(logging_text + "Starting wait_vm_up_insert_key_ro") - ro_nsr_id = None ip_address = None - nb_tries = 0 target_vdu_id = None ro_retries = 0 while True: - ro_retries += 1 if ro_retries >= 360: # 1 hour - raise LcmException("Not found _admin.deployed.RO.nsr_id for nsr_id: {}".format(nsr_id)) + raise LcmException( + "Not found _admin.deployed.RO.nsr_id for nsr_id: {}".format(nsr_id) + ) - await asyncio.sleep(10, loop=self.loop) + await asyncio.sleep(10) # get ip address if not target_vdu_id: @@ -1061,33 +1395,62 @@ class NsLcm(LcmBase): if not vdu_id: # for the VNF case if db_vnfr.get("status") == "ERROR": - raise LcmException("Cannot inject ssh-key because target VNF is in error state") + raise LcmException( + "Cannot inject ssh-key because target VNF is in error state" + ) ip_address = db_vnfr.get("ip-address") if not ip_address: continue - vdur = next((x for x in get_iterable(db_vnfr, "vdur") if x.get("ip-address") == ip_address), None) + vdur = next( + ( + x + for x in get_iterable(db_vnfr, "vdur") + if x.get("ip-address") == ip_address + ), + None, + ) else: # VDU case - vdur = next((x for x in get_iterable(db_vnfr, "vdur") - if x.get("vdu-id-ref") == vdu_id and x.get("count-index") == vdu_index), None) + vdur = next( + ( + x + for x in get_iterable(db_vnfr, "vdur") + if x.get("vdu-id-ref") == vdu_id + and x.get("count-index") == vdu_index + ), + None, + ) - if not vdur and len(db_vnfr.get("vdur", ())) == 1: # If only one, this should be the target vdu + if ( + not vdur and len(db_vnfr.get("vdur", ())) == 1 + ): # If only one, this should be the target vdu vdur = db_vnfr["vdur"][0] if not vdur: - raise LcmException("Not found vnfr_id={}, vdu_id={}, vdu_index={}".format(vnfr_id, vdu_id, - vdu_index)) + raise LcmException( + "Not found vnfr_id={}, vdu_id={}, vdu_index={}".format( + vnfr_id, vdu_id, vdu_index + ) + ) # New generation RO stores information at "vim_info" ng_ro_status = None target_vim = None if vdur.get("vim_info"): - target_vim = next(t for t in vdur["vim_info"]) # there should be only one key + target_vim = next( + t for t in vdur["vim_info"] + ) # there should be only one key ng_ro_status = vdur["vim_info"][target_vim].get("vim_status") - if vdur.get("pdu-type") or vdur.get("status") == "ACTIVE" or ng_ro_status == "ACTIVE": + if ( + vdur.get("pdu-type") + or vdur.get("status") == "ACTIVE" + or ng_ro_status == "ACTIVE" + ): ip_address = vdur.get("ip-address") if not ip_address: continue target_vdu_id = vdur["vdu-id-ref"] elif vdur.get("status") == "ERROR" or ng_ro_status == "ERROR": - raise LcmException("Cannot inject ssh-key because target VM is in error state") + raise LcmException( + "Cannot inject ssh-key because target VM is in error state" + ) if not target_vdu_id: continue @@ -1100,46 +1463,24 @@ class NsLcm(LcmBase): self.logger.error(logging_text + "Cannot inject ssh-ky to a PDU") return ip_address try: - ro_vm_id = "{}-{}".format(db_vnfr["member-vnf-index-ref"], target_vdu_id) # TODO add vdu_index - if self.ng_ro: - target = {"action": {"action": "inject_ssh_key", "key": pub_key, "user": user}, - "vnf": [{"_id": vnfr_id, "vdur": [{"id": vdur["id"]}]}], - } - desc = await self.RO.deploy(nsr_id, target) - action_id = desc["action_id"] - await self._wait_ng_ro(nsr_id, action_id, timeout=600) - break - else: - # wait until NS is deployed at RO - if not ro_nsr_id: - db_nsrs = self.db.get_one("nsrs", {"_id": nsr_id}) - ro_nsr_id = deep_get(db_nsrs, ("_admin", "deployed", "RO", "nsr_id")) - if not ro_nsr_id: - continue - result_dict = await self.RO.create_action( - item="ns", - item_id_name=ro_nsr_id, - descriptor={"add_public_key": pub_key, "vms": [ro_vm_id], "user": user} - ) - # result_dict contains the format {VM-id: {vim_result: 200, description: text}} - if not result_dict or not isinstance(result_dict, dict): - raise LcmException("Unknown response from RO when injecting key") - for result in result_dict.values(): - if result.get("vim_result") == 200: - break - else: - raise ROclient.ROClientException("error injecting key: {}".format( - result.get("description"))) - break + target = { + "action": { + "action": "inject_ssh_key", + "key": pub_key, + "user": user, + }, + "vnf": [{"_id": vnfr_id, "vdur": [{"id": vdur["id"]}]}], + } + desc = await self.RO.deploy(nsr_id, target) + action_id = desc["action_id"] + await self._wait_ng_ro( + nsr_id, action_id, timeout=600, operation="instantiation" + ) + break except NgRoException as e: - raise LcmException("Reaching max tries injecting key. Error: {}".format(e)) - except ROclient.ROClientException as e: - if not nb_tries: - self.logger.debug(logging_text + "error injecting key: {}. Retrying until {} seconds". - format(e, 20*10)) - nb_tries += 1 - if nb_tries >= 20: - raise LcmException("Reaching max tries injecting key. Error: {}".format(e)) + raise LcmException( + "Reaching max tries injecting key. Error: {}".format(e) + ) else: break @@ -1162,13 +1503,17 @@ class NsLcm(LcmBase): if index == vca_index: # myself continue - if not my_vca.get("member-vnf-index") or \ - (vca_deployed.get("member-vnf-index") == my_vca.get("member-vnf-index")): + if not my_vca.get("member-vnf-index") or ( + vca_deployed.get("member-vnf-index") + == my_vca.get("member-vnf-index") + ): internal_status = configuration_status_list[index].get("status") - if internal_status == 'READY': + if internal_status == "READY": continue - elif internal_status == 'BROKEN': - raise LcmException("Configuration aborted because dependent charm/s has failed") + elif internal_status == "BROKEN": + raise LcmException( + "Configuration aborted because dependent charm/s has failed" + ) else: break else: @@ -1179,23 +1524,48 @@ class NsLcm(LcmBase): raise LcmException("Configuration aborted because dependent charm/s timeout") - async def instantiate_N2VC(self, logging_text, vca_index, nsi_id, db_nsr, db_vnfr, vdu_id, kdu_name, vdu_index, - config_descriptor, deploy_params, base_folder, nslcmop_id, stage, vca_type, vca_name, - ee_config_descriptor): + def get_vca_id(self, db_vnfr: dict, db_nsr: dict): + vca_id = None + if db_vnfr: + vca_id = deep_get(db_vnfr, ("vca-id",)) + elif db_nsr: + vim_account_id = deep_get(db_nsr, ("instantiate_params", "vimAccountId")) + vca_id = VimAccountDB.get_vim_account_with_id(vim_account_id).get("vca") + return vca_id + + async def instantiate_N2VC( + self, + logging_text, + vca_index, + nsi_id, + db_nsr, + db_vnfr, + vdu_id, + kdu_name, + vdu_index, + kdu_index, + config_descriptor, + deploy_params, + base_folder, + nslcmop_id, + stage, + vca_type, + vca_name, + ee_config_descriptor, + ): nsr_id = db_nsr["_id"] db_update_entry = "_admin.deployed.VCA.{}.".format(vca_index) vca_deployed_list = db_nsr["_admin"]["deployed"]["VCA"] vca_deployed = db_nsr["_admin"]["deployed"]["VCA"][vca_index] osm_config = {"osm": {"ns_id": db_nsr["_id"]}} db_dict = { - 'collection': 'nsrs', - 'filter': {'_id': nsr_id}, - 'path': db_update_entry + "collection": "nsrs", + "filter": {"_id": nsr_id}, + "path": db_update_entry, } step = "" try: - - element_type = 'NS' + element_type = "NS" element_under_configuration = nsr_id vnfr_id = None @@ -1203,108 +1573,141 @@ class NsLcm(LcmBase): vnfr_id = db_vnfr["_id"] osm_config["osm"]["vnf_id"] = vnfr_id - namespace = "{nsi}.{ns}".format( - nsi=nsi_id if nsi_id else "", - ns=nsr_id) + namespace = "{nsi}.{ns}".format(nsi=nsi_id if nsi_id else "", ns=nsr_id) + + if vca_type == "native_charm": + index_number = 0 + else: + index_number = vdu_index or 0 if vnfr_id: - element_type = 'VNF' + element_type = "VNF" element_under_configuration = vnfr_id - namespace += ".{}-{}".format(vnfr_id, vdu_index or 0) + namespace += ".{}-{}".format(vnfr_id, index_number) if vdu_id: - namespace += ".{}-{}".format(vdu_id, vdu_index or 0) - element_type = 'VDU' - element_under_configuration = "{}-{}".format(vdu_id, vdu_index or 0) + namespace += ".{}-{}".format(vdu_id, index_number) + element_type = "VDU" + element_under_configuration = "{}-{}".format(vdu_id, index_number) osm_config["osm"]["vdu_id"] = vdu_id elif kdu_name: - namespace += ".{}.{}".format(kdu_name, vdu_index or 0) - element_type = 'KDU' + namespace += ".{}".format(kdu_name) + element_type = "KDU" element_under_configuration = kdu_name osm_config["osm"]["kdu_name"] = kdu_name # Get artifact path - artifact_path = "{}/{}/{}/{}".format( - base_folder["folder"], - base_folder["pkg-dir"], - "charms" if vca_type in ("native_charm", "lxc_proxy_charm", "k8s_proxy_charm") else "helm-charts", - vca_name - ) + if base_folder["pkg-dir"]: + artifact_path = "{}/{}/{}/{}".format( + base_folder["folder"], + base_folder["pkg-dir"], + "charms" + if vca_type + in ("native_charm", "lxc_proxy_charm", "k8s_proxy_charm") + else "helm-charts", + vca_name, + ) + else: + artifact_path = "{}/Scripts/{}/{}/".format( + base_folder["folder"], + "charms" + if vca_type + in ("native_charm", "lxc_proxy_charm", "k8s_proxy_charm") + else "helm-charts", + vca_name, + ) self.logger.debug("Artifact path > {}".format(artifact_path)) # get initial_config_primitive_list that applies to this element - initial_config_primitive_list = config_descriptor.get('initial-config-primitive') + initial_config_primitive_list = config_descriptor.get( + "initial-config-primitive" + ) - self.logger.debug("Initial config primitive list > {}".format(initial_config_primitive_list)) + self.logger.debug( + "Initial config primitive list > {}".format( + initial_config_primitive_list + ) + ) # add config if not present for NS charm ee_descriptor_id = ee_config_descriptor.get("id") self.logger.debug("EE Descriptor > {}".format(ee_descriptor_id)) - initial_config_primitive_list = get_ee_sorted_initial_config_primitive_list(initial_config_primitive_list, - vca_deployed, ee_descriptor_id) + initial_config_primitive_list = get_ee_sorted_initial_config_primitive_list( + initial_config_primitive_list, vca_deployed, ee_descriptor_id + ) - self.logger.debug("Initial config primitive list #2 > {}".format(initial_config_primitive_list)) + self.logger.debug( + "Initial config primitive list #2 > {}".format( + initial_config_primitive_list + ) + ) # n2vc_redesign STEP 3.1 # find old ee_id if exists ee_id = vca_deployed.get("ee_id") - vim_account_id = ( - deep_get(db_vnfr, ("vim-account-id",)) or - deep_get(deploy_params, ("OSM", "vim_account_id")) - ) - vca_cloud, vca_cloud_credential = self.get_vca_cloud_and_credentials(vim_account_id) - vca_k8s_cloud, vca_k8s_cloud_credential = self.get_vca_k8s_cloud_and_credentials(vim_account_id) + vca_id = self.get_vca_id(db_vnfr, db_nsr) # create or register execution environment in VCA if vca_type in ("lxc_proxy_charm", "k8s_proxy_charm", "helm", "helm-v3"): - self._write_configuration_status( nsr_id=nsr_id, vca_index=vca_index, - status='CREATING', + status="CREATING", element_under_configuration=element_under_configuration, - element_type=element_type + element_type=element_type, ) step = "create execution environment" - self.logger.debug(logging_text + step) + self.logger.debug(logging_text + step) ee_id = None credentials = None if vca_type == "k8s_proxy_charm": ee_id = await self.vca_map[vca_type].install_k8s_proxy_charm( - charm_name=artifact_path[artifact_path.rfind("/") + 1:], + charm_name=artifact_path[artifact_path.rfind("/") + 1 :], namespace=namespace, artifact_path=artifact_path, db_dict=db_dict, - cloud_name=vca_k8s_cloud, - credential_name=vca_k8s_cloud_credential, + vca_id=vca_id, ) - elif vca_type == "helm" or vca_type == "helm-v3": - ee_id, credentials = await self.vca_map[vca_type].create_execution_environment( - namespace=namespace, + elif vca_type == "helm" or vca_type == "helm-v3": + ee_id, credentials = await self.vca_map[ + vca_type + ].create_execution_environment( + namespace=nsr_id, reuse_ee_id=ee_id, db_dict=db_dict, config=osm_config, artifact_path=artifact_path, - vca_type=vca_type + chart_model=vca_name, + vca_type=vca_type, ) - else: - ee_id, credentials = await self.vca_map[vca_type].create_execution_environment( + else: + ee_id, credentials = await self.vca_map[ + vca_type + ].create_execution_environment( namespace=namespace, reuse_ee_id=ee_id, db_dict=db_dict, - cloud_name=vca_cloud, - credential_name=vca_cloud_credential, + vca_id=vca_id, ) elif vca_type == "native_charm": step = "Waiting to VM being up and getting IP address" self.logger.debug(logging_text + step) - rw_mgmt_ip = await self.wait_vm_up_insert_key_ro(logging_text, nsr_id, vnfr_id, vdu_id, vdu_index, - user=None, pub_key=None) + rw_mgmt_ip = await self.wait_vm_up_insert_key_ro( + logging_text, + nsr_id, + vnfr_id, + vdu_id, + vdu_index, + user=None, + pub_key=None, + ) credentials = {"hostname": rw_mgmt_ip} # get username - username = deep_get(config_descriptor, ("config-access", "ssh-access", "default-user")) + username = deep_get( + config_descriptor, ("config-access", "ssh-access", "default-user") + ) # TODO remove this when changes on IM regarding config-access:ssh-access:default-user were # merged. Meanwhile let's get username from initial-config-primitive if not username and initial_config_primitive_list: @@ -1314,17 +1717,19 @@ class NsLcm(LcmBase): username = param["value"] break if not username: - raise LcmException("Cannot determine the username neither with 'initial-config-primitive' nor with " - "'config-access.ssh-access.default-user'") + raise LcmException( + "Cannot determine the username neither with 'initial-config-primitive' nor with " + "'config-access.ssh-access.default-user'" + ) credentials["username"] = username # n2vc_redesign STEP 3.2 self._write_configuration_status( nsr_id=nsr_id, vca_index=vca_index, - status='REGISTERING', + status="REGISTERING", element_under_configuration=element_under_configuration, - element_type=element_type + element_type=element_type, ) step = "register execution environment {}".format(credentials) @@ -1333,13 +1738,12 @@ class NsLcm(LcmBase): credentials=credentials, namespace=namespace, db_dict=db_dict, - cloud_name=vca_cloud, - credential_name=vca_cloud_credential, + vca_id=vca_id, ) # for compatibility with MON/POL modules, the need model and application name at database # TODO ask MON/POL if needed to not assuming anymore the format "model_name.application_name" - ee_id_parts = ee_id.split('.') + ee_id_parts = ee_id.split(".") db_nsr_update = {db_update_entry + "ee_id": ee_id} if len(ee_id_parts) >= 2: model_name = ee_id_parts[0] @@ -1353,22 +1757,23 @@ class NsLcm(LcmBase): self._write_configuration_status( nsr_id=nsr_id, vca_index=vca_index, - status='INSTALLING SW', + status="INSTALLING SW", element_under_configuration=element_under_configuration, element_type=element_type, - other_update=db_nsr_update + other_update=db_nsr_update, ) # TODO check if already done self.logger.debug(logging_text + step) config = None if vca_type == "native_charm": - config_primitive = next((p for p in initial_config_primitive_list if p["name"] == "config"), None) + config_primitive = next( + (p for p in initial_config_primitive_list if p["name"] == "config"), + None, + ) if config_primitive: config = self._map_primitive_params( - config_primitive, - {}, - deploy_params + config_primitive, {}, deploy_params ) num_units = 1 if vca_type == "lxc_proxy_charm": @@ -1388,14 +1793,25 @@ class NsLcm(LcmBase): db_dict=db_dict, config=config, num_units=num_units, + vca_id=vca_id, + vca_type=vca_type, ) # write in db flag of configuration_sw already installed - self.update_db_2("nsrs", nsr_id, {db_update_entry + "config_sw_installed": True}) + self.update_db_2( + "nsrs", nsr_id, {db_update_entry + "config_sw_installed": True} + ) # add relations for this VCA (wait for other peers related with this VCA) - await self._add_vca_relations(logging_text=logging_text, nsr_id=nsr_id, - vca_index=vca_index, vca_type=vca_type) + is_relation_added = await self._add_vca_relations( + logging_text=logging_text, + nsr_id=nsr_id, + vca_type=vca_type, + vca_index=vca_index, + ) + + if not is_relation_added: + raise LcmException("Relations could not be added to VCA.") # if SSH access is required, then get execution environment SSH public # if native charm we have waited already to VM be UP @@ -1403,37 +1819,86 @@ class NsLcm(LcmBase): pub_key = None user = None # self.logger.debug("get ssh key block") - if deep_get(config_descriptor, ("config-access", "ssh-access", "required")): + if deep_get( + config_descriptor, ("config-access", "ssh-access", "required") + ): # self.logger.debug("ssh key needed") # Needed to inject a ssh key - user = deep_get(config_descriptor, ("config-access", "ssh-access", "default-user")) + user = deep_get( + config_descriptor, + ("config-access", "ssh-access", "default-user"), + ) step = "Install configuration Software, getting public ssh key" - pub_key = await self.vca_map[vca_type].get_ee_ssh_public__key(ee_id=ee_id, db_dict=db_dict) + pub_key = await self.vca_map[vca_type].get_ee_ssh_public__key( + ee_id=ee_id, db_dict=db_dict, vca_id=vca_id + ) - step = "Insert public key into VM user={} ssh_key={}".format(user, pub_key) + step = "Insert public key into VM user={} ssh_key={}".format( + user, pub_key + ) else: # self.logger.debug("no need to get ssh key") step = "Waiting to VM being up and getting IP address" self.logger.debug(logging_text + step) + # default rw_mgmt_ip to None, avoiding the non definition of the variable + rw_mgmt_ip = None + # n2vc_redesign STEP 5.1 # wait for RO (ip-address) Insert pub_key into VM if vnfr_id: if kdu_name: - rw_mgmt_ip = await self.wait_kdu_up(logging_text, nsr_id, vnfr_id, kdu_name) - else: - rw_mgmt_ip = await self.wait_vm_up_insert_key_ro(logging_text, nsr_id, vnfr_id, vdu_id, - vdu_index, user=user, pub_key=pub_key) - else: - rw_mgmt_ip = None # This is for a NS configuration + rw_mgmt_ip, services = await self.wait_kdu_up( + logging_text, nsr_id, vnfr_id, kdu_name + ) + vnfd = self.db.get_one( + "vnfds_revisions", + {"_id": f'{db_vnfr["vnfd-id"]}:{db_vnfr["revision"]}'}, + ) + kdu = get_kdu(vnfd, kdu_name) + kdu_services = [ + service["name"] for service in get_kdu_services(kdu) + ] + exposed_services = [] + for service in services: + if any(s in service["name"] for s in kdu_services): + exposed_services.append(service) + await self.vca_map[vca_type].exec_primitive( + ee_id=ee_id, + primitive_name="config", + params_dict={ + "osm-config": json.dumps( + OsmConfigBuilder( + k8s={"services": exposed_services} + ).build() + ) + }, + vca_id=vca_id, + ) + + # This verification is needed in order to avoid trying to add a public key + # to a VM, when the VNF is a KNF (in the edge case where the user creates a VCA + # for a KNF and not for its KDUs, the previous verification gives False, and the code + # jumps to this block, meaning that there is the need to verify if the VNF is actually a VNF + # or it is a KNF) + elif db_vnfr.get("vdur"): + rw_mgmt_ip = await self.wait_vm_up_insert_key_ro( + logging_text, + nsr_id, + vnfr_id, + vdu_id, + vdu_index, + user=user, + pub_key=pub_key, + ) - self.logger.debug(logging_text + ' VM_ip_address={}'.format(rw_mgmt_ip)) + self.logger.debug(logging_text + " VM_ip_address={}".format(rw_mgmt_ip)) # store rw_mgmt_ip in deploy params for later replacement deploy_params["rw_mgmt_ip"] = rw_mgmt_ip # n2vc_redesign STEP 6 Execute initial config primitive - step = 'execute initial config primitive' + step = "execute initial config primitive" # wait for dependent primitives execution (NS -> VNF -> VDU) if initial_config_primitive_list: @@ -1443,84 +1908,118 @@ class NsLcm(LcmBase): my_vca = vca_deployed_list[vca_index] if my_vca.get("vdu_id") or my_vca.get("kdu_name"): # VDU or KDU - stage[0] = 'Stage 3/5: running Day-1 primitives for VDU.' + stage[0] = "Stage 3/5: running Day-1 primitives for VDU." elif my_vca.get("member-vnf-index"): # VNF - stage[0] = 'Stage 4/5: running Day-1 primitives for VNF.' + stage[0] = "Stage 4/5: running Day-1 primitives for VNF." else: # NS - stage[0] = 'Stage 5/5: running Day-1 primitives for NS.' + stage[0] = "Stage 5/5: running Day-1 primitives for NS." self._write_configuration_status( - nsr_id=nsr_id, - vca_index=vca_index, - status='EXECUTING PRIMITIVE' + nsr_id=nsr_id, vca_index=vca_index, status="EXECUTING PRIMITIVE" ) - self._write_op_status( - op_id=nslcmop_id, - stage=stage - ) + self._write_op_status(op_id=nslcmop_id, stage=stage) check_if_terminated_needed = True for initial_config_primitive in initial_config_primitive_list: # adding information on the vca_deployed if it is a NS execution environment if not vca_deployed["member-vnf-index"]: - deploy_params["ns_config_info"] = json.dumps(self._get_ns_config_info(nsr_id)) + deploy_params["ns_config_info"] = json.dumps( + self._get_ns_config_info(nsr_id) + ) # TODO check if already done - primitive_params_ = self._map_primitive_params(initial_config_primitive, {}, deploy_params) + primitive_params_ = self._map_primitive_params( + initial_config_primitive, {}, deploy_params + ) - step = "execute primitive '{}' params '{}'".format(initial_config_primitive["name"], primitive_params_) + step = "execute primitive '{}' params '{}'".format( + initial_config_primitive["name"], primitive_params_ + ) self.logger.debug(logging_text + step) await self.vca_map[vca_type].exec_primitive( ee_id=ee_id, primitive_name=initial_config_primitive["name"], params_dict=primitive_params_, - db_dict=db_dict + db_dict=db_dict, + vca_id=vca_id, + vca_type=vca_type, ) # Once some primitive has been exec, check and write at db if it needs to exec terminated primitives if check_if_terminated_needed: - if config_descriptor.get('terminate-config-primitive'): - self.update_db_2("nsrs", nsr_id, {db_update_entry + "needed_terminate": True}) + if config_descriptor.get("terminate-config-primitive"): + self.update_db_2( + "nsrs", nsr_id, {db_update_entry + "needed_terminate": True} + ) check_if_terminated_needed = False # TODO register in database that primitive is done # STEP 7 Configure metrics if vca_type == "helm" or vca_type == "helm-v3": - prometheus_jobs = await self.add_prometheus_metrics( + # TODO: review for those cases where the helm chart is a reference and + # is not part of the NF package + prometheus_jobs = await self.extract_prometheus_scrape_jobs( ee_id=ee_id, artifact_path=artifact_path, ee_config_descriptor=ee_config_descriptor, vnfr_id=vnfr_id, nsr_id=nsr_id, target_ip=rw_mgmt_ip, + element_type=element_type, + vnf_member_index=db_vnfr.get("member-vnf-index-ref", ""), + vdu_id=vdu_id, + vdu_index=vdu_index, + kdu_name=kdu_name, + kdu_index=kdu_index, ) if prometheus_jobs: - self.update_db_2("nsrs", nsr_id, {db_update_entry + "prometheus_jobs": prometheus_jobs}) + self.update_db_2( + "nsrs", + nsr_id, + {db_update_entry + "prometheus_jobs": prometheus_jobs}, + ) + + for job in prometheus_jobs: + self.db.set_one( + "prometheus_jobs", + {"job_name": job["job_name"]}, + job, + upsert=True, + fail_on_empty=False, + ) step = "instantiated at VCA" self.logger.debug(logging_text + step) self._write_configuration_status( - nsr_id=nsr_id, - vca_index=vca_index, - status='READY' + nsr_id=nsr_id, vca_index=vca_index, status="READY" ) except Exception as e: # TODO not use Exception but N2VC exception # self.update_db_2("nsrs", nsr_id, {db_update_entry + "instantiation": "FAILED"}) - if not isinstance(e, (DbException, N2VCException, LcmException, asyncio.CancelledError)): - self.logger.error("Exception while {} : {}".format(step, e), exc_info=True) + if not isinstance( + e, (DbException, N2VCException, LcmException, asyncio.CancelledError) + ): + self.logger.error( + "Exception while {} : {}".format(step, e), exc_info=True + ) self._write_configuration_status( - nsr_id=nsr_id, - vca_index=vca_index, - status='BROKEN' + nsr_id=nsr_id, vca_index=vca_index, status="BROKEN" ) - raise LcmException("{} {}".format(step, e)) from e - - def _write_ns_status(self, nsr_id: str, ns_state: str, current_operation: str, current_operation_id: str, - error_description: str = None, error_detail: str = None, other_update: dict = None): + raise LcmException("{}. {}".format(step, e)) from e + + def _write_ns_status( + self, + nsr_id: str, + ns_state: str, + current_operation: str, + current_operation_id: str, + error_description: str = None, + error_detail: str = None, + other_update: dict = None, + ): """ Update db_nsr fields. :param nsr_id: @@ -1534,9 +2033,13 @@ class NsLcm(LcmBase): """ try: db_dict = other_update or {} - db_dict["_admin.nslcmop"] = current_operation_id # for backward compatibility + db_dict[ + "_admin.nslcmop" + ] = current_operation_id # for backward compatibility db_dict["_admin.current-operation"] = current_operation_id - db_dict["_admin.operation-type"] = current_operation if current_operation != "IDLE" else None + db_dict["_admin.operation-type"] = ( + current_operation if current_operation != "IDLE" else None + ) db_dict["currentOperation"] = current_operation db_dict["currentOperationID"] = current_operation_id db_dict["errorDescription"] = error_description @@ -1546,62 +2049,86 @@ class NsLcm(LcmBase): db_dict["nsState"] = ns_state self.update_db_2("nsrs", nsr_id, db_dict) except DbException as e: - self.logger.warn('Error writing NS status, ns={}: {}'.format(nsr_id, e)) - - def _write_op_status(self, op_id: str, stage: list = None, error_message: str = None, queuePosition: int = 0, - operation_state: str = None, other_update: dict = None): + self.logger.warn("Error writing NS status, ns={}: {}".format(nsr_id, e)) + + def _write_op_status( + self, + op_id: str, + stage: list = None, + error_message: str = None, + queuePosition: int = 0, + operation_state: str = None, + other_update: dict = None, + ): try: db_dict = other_update or {} - db_dict['queuePosition'] = queuePosition + db_dict["queuePosition"] = queuePosition if isinstance(stage, list): - db_dict['stage'] = stage[0] - db_dict['detailed-status'] = " ".join(stage) + db_dict["stage"] = stage[0] + db_dict["detailed-status"] = " ".join(stage) elif stage is not None: - db_dict['stage'] = str(stage) + db_dict["stage"] = str(stage) if error_message is not None: - db_dict['errorMessage'] = error_message + db_dict["errorMessage"] = error_message if operation_state is not None: - db_dict['operationState'] = operation_state + db_dict["operationState"] = operation_state db_dict["statusEnteredTime"] = time() self.update_db_2("nslcmops", op_id, db_dict) except DbException as e: - self.logger.warn('Error writing OPERATION status for op_id: {} -> {}'.format(op_id, e)) + self.logger.warn( + "Error writing OPERATION status for op_id: {} -> {}".format(op_id, e) + ) def _write_all_config_status(self, db_nsr: dict, status: str): try: nsr_id = db_nsr["_id"] # configurationStatus - config_status = db_nsr.get('configurationStatus') + config_status = db_nsr.get("configurationStatus") if config_status: - db_nsr_update = {"configurationStatus.{}.status".format(index): status for index, v in - enumerate(config_status) if v} + db_nsr_update = { + "configurationStatus.{}.status".format(index): status + for index, v in enumerate(config_status) + if v + } # update status self.update_db_2("nsrs", nsr_id, db_nsr_update) except DbException as e: - self.logger.warn('Error writing all configuration status, ns={}: {}'.format(nsr_id, e)) - - def _write_configuration_status(self, nsr_id: str, vca_index: int, status: str = None, - element_under_configuration: str = None, element_type: str = None, - other_update: dict = None): + self.logger.warn( + "Error writing all configuration status, ns={}: {}".format(nsr_id, e) + ) + def _write_configuration_status( + self, + nsr_id: str, + vca_index: int, + status: str = None, + element_under_configuration: str = None, + element_type: str = None, + other_update: dict = None, + ): # self.logger.debug('_write_configuration_status(): vca_index={}, status={}' # .format(vca_index, status)) try: - db_path = 'configurationStatus.{}.'.format(vca_index) + db_path = "configurationStatus.{}.".format(vca_index) db_dict = other_update or {} if status: - db_dict[db_path + 'status'] = status + db_dict[db_path + "status"] = status if element_under_configuration: - db_dict[db_path + 'elementUnderConfiguration'] = element_under_configuration + db_dict[ + db_path + "elementUnderConfiguration" + ] = element_under_configuration if element_type: - db_dict[db_path + 'elementType'] = element_type + db_dict[db_path + "elementType"] = element_type self.update_db_2("nsrs", nsr_id, db_dict) except DbException as e: - self.logger.warn('Error writing configuration status={}, ns={}, vca_index={}: {}' - .format(status, nsr_id, vca_index, e)) + self.logger.warn( + "Error writing configuration status={}, ns={}, vca_index={}: {}".format( + status, nsr_id, vca_index, e + ) + ) async def _do_placement(self, logging_text, db_nslcmop, db_vnfrs): """ @@ -1615,11 +2142,13 @@ class NsLcm(LcmBase): computed 'vim-account-id' """ modified = False - nslcmop_id = db_nslcmop['_id'] - placement_engine = deep_get(db_nslcmop, ('operationParams', 'placement-engine')) + nslcmop_id = db_nslcmop["_id"] + placement_engine = deep_get(db_nslcmop, ("operationParams", "placement-engine")) if placement_engine == "PLA": - self.logger.debug(logging_text + "Invoke and wait for placement optimization") - await self.msg.aiowrite("pla", "get_placement", {'nslcmopId': nslcmop_id}, loop=self.loop) + self.logger.debug( + logging_text + "Invoke and wait for placement optimization" + ) + await self.msg.aiowrite("pla", "get_placement", {"nslcmopId": nslcmop_id}) db_poll_interval = 5 wait = db_poll_interval * 10 pla_result = None @@ -1627,27 +2156,328 @@ class NsLcm(LcmBase): await asyncio.sleep(db_poll_interval) wait -= db_poll_interval db_nslcmop = self.db.get_one("nslcmops", {"_id": nslcmop_id}) - pla_result = deep_get(db_nslcmop, ('_admin', 'pla')) + pla_result = deep_get(db_nslcmop, ("_admin", "pla")) if not pla_result: - raise LcmException("Placement timeout for nslcmopId={}".format(nslcmop_id)) + raise LcmException( + "Placement timeout for nslcmopId={}".format(nslcmop_id) + ) - for pla_vnf in pla_result['vnf']: - vnfr = db_vnfrs.get(pla_vnf['member-vnf-index']) - if not pla_vnf.get('vimAccountId') or not vnfr: + for pla_vnf in pla_result["vnf"]: + vnfr = db_vnfrs.get(pla_vnf["member-vnf-index"]) + if not pla_vnf.get("vimAccountId") or not vnfr: continue modified = True - self.db.set_one("vnfrs", {"_id": vnfr["_id"]}, {"vim-account-id": pla_vnf['vimAccountId']}) + self.db.set_one( + "vnfrs", + {"_id": vnfr["_id"]}, + {"vim-account-id": pla_vnf["vimAccountId"]}, + ) # Modifies db_vnfrs - vnfr["vim-account-id"] = pla_vnf['vimAccountId'] + vnfr["vim-account-id"] = pla_vnf["vimAccountId"] return modified + def _gather_vnfr_healing_alerts(self, vnfr, vnfd): + alerts = [] + nsr_id = vnfr["nsr-id-ref"] + df = vnfd.get("df", [{}])[0] + # Checking for auto-healing configuration + if "healing-aspect" in df: + healing_aspects = df["healing-aspect"] + for healing in healing_aspects: + for healing_policy in healing.get("healing-policy", ()): + vdu_id = healing_policy["vdu-id"] + vdur = next( + (vdur for vdur in vnfr["vdur"] if vdu_id == vdur["vdu-id-ref"]), + {}, + ) + if not vdur: + continue + metric_name = "vm_status" + vdu_name = vdur.get("name") + vnf_member_index = vnfr["member-vnf-index-ref"] + uuid = str(uuid4()) + name = f"healing_{uuid}" + action = healing_policy + # action_on_recovery = healing.get("action-on-recovery") + # cooldown_time = healing.get("cooldown-time") + # day1 = healing.get("day1") + alert = { + "uuid": uuid, + "name": name, + "metric": metric_name, + "tags": { + "ns_id": nsr_id, + "vnf_member_index": vnf_member_index, + "vdu_name": vdu_name, + }, + "alarm_status": "ok", + "action_type": "healing", + "action": action, + } + alerts.append(alert) + return alerts + + def _gather_vnfr_scaling_alerts(self, vnfr, vnfd): + alerts = [] + nsr_id = vnfr["nsr-id-ref"] + df = vnfd.get("df", [{}])[0] + # Checking for auto-scaling configuration + if "scaling-aspect" in df: + scaling_aspects = df["scaling-aspect"] + all_vnfd_monitoring_params = {} + for ivld in vnfd.get("int-virtual-link-desc", ()): + for mp in ivld.get("monitoring-parameters", ()): + all_vnfd_monitoring_params[mp.get("id")] = mp + for vdu in vnfd.get("vdu", ()): + for mp in vdu.get("monitoring-parameter", ()): + all_vnfd_monitoring_params[mp.get("id")] = mp + for df in vnfd.get("df", ()): + for mp in df.get("monitoring-parameter", ()): + all_vnfd_monitoring_params[mp.get("id")] = mp + for scaling_aspect in scaling_aspects: + scaling_group_name = scaling_aspect.get("name", "") + # Get monitored VDUs + all_monitored_vdus = set() + for delta in scaling_aspect.get("aspect-delta-details", {}).get( + "deltas", () + ): + for vdu_delta in delta.get("vdu-delta", ()): + all_monitored_vdus.add(vdu_delta.get("id")) + monitored_vdurs = list( + filter( + lambda vdur: vdur["vdu-id-ref"] in all_monitored_vdus, + vnfr["vdur"], + ) + ) + if not monitored_vdurs: + self.logger.error( + "Scaling criteria is referring to a vnf-monitoring-param that does not contain a reference to a vdu or vnf metric" + ) + continue + for scaling_policy in scaling_aspect.get("scaling-policy", ()): + if scaling_policy["scaling-type"] != "automatic": + continue + threshold_time = scaling_policy.get("threshold-time", "1") + cooldown_time = scaling_policy.get("cooldown-time", "0") + for scaling_criteria in scaling_policy["scaling-criteria"]: + monitoring_param_ref = scaling_criteria.get( + "vnf-monitoring-param-ref" + ) + vnf_monitoring_param = all_vnfd_monitoring_params[ + monitoring_param_ref + ] + for vdur in monitored_vdurs: + vdu_id = vdur["vdu-id-ref"] + metric_name = vnf_monitoring_param.get("performance-metric") + metric_name = f"osm_{metric_name}" + vnf_member_index = vnfr["member-vnf-index-ref"] + scalein_threshold = scaling_criteria.get( + "scale-in-threshold" + ) + scaleout_threshold = scaling_criteria.get( + "scale-out-threshold" + ) + # Looking for min/max-number-of-instances + instances_min_number = 1 + instances_max_number = 1 + vdu_profile = df["vdu-profile"] + if vdu_profile: + profile = next( + item for item in vdu_profile if item["id"] == vdu_id + ) + instances_min_number = profile.get( + "min-number-of-instances", 1 + ) + instances_max_number = profile.get( + "max-number-of-instances", 1 + ) + + if scalein_threshold: + uuid = str(uuid4()) + name = f"scalein_{uuid}" + operation = scaling_criteria[ + "scale-in-relational-operation" + ] + rel_operator = self.rel_operation_types.get( + operation, "<=" + ) + metric_selector = f'{metric_name}{{ns_id="{nsr_id}", vnf_member_index="{vnf_member_index}", vdu_id="{vdu_id}"}}' + expression = f"(count ({metric_selector}) > {instances_min_number}) and (avg({metric_selector}) {rel_operator} {scalein_threshold})" + labels = { + "ns_id": nsr_id, + "vnf_member_index": vnf_member_index, + "vdu_id": vdu_id, + } + prom_cfg = { + "alert": name, + "expr": expression, + "for": str(threshold_time) + "m", + "labels": labels, + } + action = scaling_policy + action = { + "scaling-group": scaling_group_name, + "cooldown-time": cooldown_time, + } + alert = { + "uuid": uuid, + "name": name, + "metric": metric_name, + "tags": { + "ns_id": nsr_id, + "vnf_member_index": vnf_member_index, + "vdu_id": vdu_id, + }, + "alarm_status": "ok", + "action_type": "scale_in", + "action": action, + "prometheus_config": prom_cfg, + } + alerts.append(alert) + + if scaleout_threshold: + uuid = str(uuid4()) + name = f"scaleout_{uuid}" + operation = scaling_criteria[ + "scale-out-relational-operation" + ] + rel_operator = self.rel_operation_types.get( + operation, "<=" + ) + metric_selector = f'{metric_name}{{ns_id="{nsr_id}", vnf_member_index="{vnf_member_index}", vdu_id="{vdu_id}"}}' + expression = f"(count ({metric_selector}) < {instances_max_number}) and (avg({metric_selector}) {rel_operator} {scaleout_threshold})" + labels = { + "ns_id": nsr_id, + "vnf_member_index": vnf_member_index, + "vdu_id": vdu_id, + } + prom_cfg = { + "alert": name, + "expr": expression, + "for": str(threshold_time) + "m", + "labels": labels, + } + action = scaling_policy + action = { + "scaling-group": scaling_group_name, + "cooldown-time": cooldown_time, + } + alert = { + "uuid": uuid, + "name": name, + "metric": metric_name, + "tags": { + "ns_id": nsr_id, + "vnf_member_index": vnf_member_index, + "vdu_id": vdu_id, + }, + "alarm_status": "ok", + "action_type": "scale_out", + "action": action, + "prometheus_config": prom_cfg, + } + alerts.append(alert) + return alerts + + def _gather_vnfr_alarm_alerts(self, vnfr, vnfd): + alerts = [] + nsr_id = vnfr["nsr-id-ref"] + vnf_member_index = vnfr["member-vnf-index-ref"] + + # Checking for VNF alarm configuration + for vdur in vnfr["vdur"]: + vdu_id = vdur["vdu-id-ref"] + vdu = next(filter(lambda vdu: vdu["id"] == vdu_id, vnfd["vdu"])) + if "alarm" in vdu: + # Get VDU monitoring params, since alerts are based on them + vdu_monitoring_params = {} + for mp in vdu.get("monitoring-parameter", []): + vdu_monitoring_params[mp.get("id")] = mp + if not vdu_monitoring_params: + self.logger.error( + "VDU alarm refers to a VDU monitoring param, but there are no VDU monitoring params in the VDU" + ) + continue + # Get alarms in the VDU + alarm_descriptors = vdu["alarm"] + # Create VDU alarms for each alarm in the VDU + for alarm_descriptor in alarm_descriptors: + # Check that the VDU alarm refers to a proper monitoring param + alarm_monitoring_param = alarm_descriptor.get( + "vnf-monitoring-param-ref", "" + ) + vdu_specific_monitoring_param = vdu_monitoring_params.get( + alarm_monitoring_param, {} + ) + if not vdu_specific_monitoring_param: + self.logger.error( + "VDU alarm refers to a VDU monitoring param not present in the VDU" + ) + continue + metric_name = vdu_specific_monitoring_param.get( + "performance-metric" + ) + if not metric_name: + self.logger.error( + "VDU alarm refers to a VDU monitoring param that has no associated performance-metric" + ) + continue + # Set params of the alarm to be created in Prometheus + metric_name = f"osm_{metric_name}" + metric_threshold = alarm_descriptor.get("value") + uuid = str(uuid4()) + alert_name = f"vdu_alarm_{uuid}" + operation = alarm_descriptor["operation"] + rel_operator = self.rel_operation_types.get(operation, "<=") + metric_selector = f'{metric_name}{{ns_id="{nsr_id}", vnf_member_index="{vnf_member_index}", vdu_id="{vdu_id}"}}' + expression = f"{metric_selector} {rel_operator} {metric_threshold}" + labels = { + "ns_id": nsr_id, + "vnf_member_index": vnf_member_index, + "vdu_id": vdu_id, + "vdu_name": "{{ $labels.vdu_name }}", + } + prom_cfg = { + "alert": alert_name, + "expr": expression, + "for": "1m", # default value. Ideally, this should be related to an IM param, but there is not such param + "labels": labels, + } + alarm_action = dict() + for action_type in ["ok", "insufficient-data", "alarm"]: + if ( + "actions" in alarm_descriptor + and action_type in alarm_descriptor["actions"] + ): + alarm_action[action_type] = alarm_descriptor["actions"][ + action_type + ] + alert = { + "uuid": uuid, + "name": alert_name, + "metric": metric_name, + "tags": { + "ns_id": nsr_id, + "vnf_member_index": vnf_member_index, + "vdu_id": vdu_id, + }, + "alarm_status": "ok", + "action_type": "vdu_alarm", + "action": alarm_action, + "prometheus_config": prom_cfg, + } + alerts.append(alert) + return alerts + def update_nsrs_with_pla_result(self, params): try: - nslcmop_id = deep_get(params, ('placement', 'nslcmopId')) - self.update_db_2("nslcmops", nslcmop_id, {"_admin.pla": params.get('placement')}) + nslcmop_id = deep_get(params, ("placement", "nslcmopId")) + self.update_db_2( + "nslcmops", nslcmop_id, {"_admin.pla": params.get("placement")} + ) except Exception as e: - self.logger.warn('Update failed for nslcmop_id={}:{}'.format(nslcmop_id, e)) + self.logger.warn("Update failed for nslcmop_id={}:{}".format(nslcmop_id, e)) async def instantiate(self, nsr_id, nslcmop_id): """ @@ -1658,9 +2488,11 @@ class NsLcm(LcmBase): """ # Try to lock HA task here - task_is_locked_by_me = self.lcm_tasks.lock_HA('ns', 'nslcmops', nslcmop_id) + task_is_locked_by_me = self.lcm_tasks.lock_HA("ns", "nslcmops", nslcmop_id) if not task_is_locked_by_me: - self.logger.debug('instantiate() task is not locked by me, ns={}'.format(nsr_id)) + self.logger.debug( + "instantiate() task is not locked by me, ns={}".format(nsr_id) + ) return logging_text = "Task ns={} instantiate={} ".format(nsr_id, nslcmop_id) @@ -1679,20 +2511,23 @@ class NsLcm(LcmBase): # update operation on nslcmops db_nslcmop_update = {} + timeout_ns_deploy = self.timeout.ns_deploy + nslcmop_operation_state = None - db_vnfrs = {} # vnf's info indexed by member-index + db_vnfrs = {} # vnf's info indexed by member-index # n2vc_info = {} tasks_dict_info = {} # from task to info text exc = None error_list = [] - stage = ['Stage 1/5: preparation of the environment.', "Waiting for previous operations to terminate.", ""] + stage = [ + "Stage 1/5: preparation of the environment.", + "Waiting for previous operations to terminate.", + "", + ] # ^ stage, step, VIM progress try: # wait for any previous tasks in process - await self.lcm_tasks.waitfor_related_HA('ns', 'nslcmops', nslcmop_id) - - stage[1] = "Sync filesystem from database." - self.fs.sync() # TODO, make use of partial sync, only for the needed packages + await self.lcm_tasks.waitfor_related_HA("ns", "nslcmops", nslcmop_id) # STEP 0: Reading database (nslcmops, nsrs, nsds, vnfrs, vnfds) stage[1] = "Reading from database." @@ -1704,28 +2539,29 @@ class NsLcm(LcmBase): ns_state="BUILDING", current_operation="INSTANTIATING", current_operation_id=nslcmop_id, - other_update=db_nsr_update - ) - self._write_op_status( - op_id=nslcmop_id, - stage=stage, - queuePosition=0 + other_update=db_nsr_update, ) + self._write_op_status(op_id=nslcmop_id, stage=stage, queuePosition=0) # read from db: operation stage[1] = "Getting nslcmop={} from db.".format(nslcmop_id) db_nslcmop = self.db.get_one("nslcmops", {"_id": nslcmop_id}) + if db_nslcmop["operationParams"].get("additionalParamsForVnf"): + db_nslcmop["operationParams"]["additionalParamsForVnf"] = json.loads( + db_nslcmop["operationParams"]["additionalParamsForVnf"] + ) ns_params = db_nslcmop.get("operationParams") if ns_params and ns_params.get("timeout_ns_deploy"): timeout_ns_deploy = ns_params["timeout_ns_deploy"] - else: - timeout_ns_deploy = self.timeout.get("ns_deploy", self.timeout_ns_deploy) # read from db: ns stage[1] = "Getting nsr={} from db.".format(nsr_id) + self.logger.debug(logging_text + stage[1]) db_nsr = self.db.get_one("nsrs", {"_id": nsr_id}) stage[1] = "Getting nsd={} from db.".format(db_nsr["nsd-id"]) + self.logger.debug(logging_text + stage[1]) nsd = self.db.get_one("nsds", {"_id": db_nsr["nsd-id"]}) + self.fs.sync(db_nsr["nsd-id"]) db_nsr["nsd"] = nsd # nsr_name = db_nsr["name"] # TODO short-name?? @@ -1735,18 +2571,31 @@ class NsLcm(LcmBase): db_vnfrs_list = self.db.get_list("vnfrs", {"nsr-id-ref": nsr_id}) # read from db: vnfd's for every vnf - db_vnfds = [] # every vnfd data + db_vnfds = [] # every vnfd data # for each vnf in ns, read vnfd for vnfr in db_vnfrs_list: + if vnfr.get("kdur"): + kdur_list = [] + for kdur in vnfr["kdur"]: + if kdur.get("additionalParams"): + kdur["additionalParams"] = json.loads( + kdur["additionalParams"] + ) + kdur_list.append(kdur) + vnfr["kdur"] = kdur_list + db_vnfrs[vnfr["member-vnf-index-ref"]] = vnfr vnfd_id = vnfr["vnfd-id"] vnfd_ref = vnfr["vnfd-ref"] + self.fs.sync(vnfd_id) # if we haven't this vnfd, read it from db if vnfd_id not in db_vnfds: # read from db - stage[1] = "Getting vnfd={} id='{}' from db.".format(vnfd_id, vnfd_ref) + stage[1] = "Getting vnfd={} id='{}' from db.".format( + vnfd_id, vnfd_ref + ) self.logger.debug(logging_text + stage[1]) vnfd = self.db.get_one("vnfds", {"_id": vnfd_id}) @@ -1770,21 +2619,22 @@ class NsLcm(LcmBase): db_nsr_update["_admin.deployed.VCA"] = vca_deployed_list populate_dict(db_nsr, ("_admin", "deployed", "VCA"), vca_deployed_list) - if not isinstance(deep_get(db_nsr, ("_admin", "deployed", "RO", "vnfd")), list): + if not isinstance( + deep_get(db_nsr, ("_admin", "deployed", "RO", "vnfd")), list + ): populate_dict(db_nsr, ("_admin", "deployed", "RO", "vnfd"), []) db_nsr_update["_admin.deployed.RO.vnfd"] = [] # set state to INSTANTIATED. When instantiated NBI will not delete directly db_nsr_update["_admin.nsState"] = "INSTANTIATED" self.update_db_2("nsrs", nsr_id, db_nsr_update) - self.db.set_list("vnfrs", {"nsr-id-ref": nsr_id}, {"_admin.nsState": "INSTANTIATED"}) + self.db.set_list( + "vnfrs", {"nsr-id-ref": nsr_id}, {"_admin.nsState": "INSTANTIATED"} + ) # n2vc_redesign STEP 2 Deploy Network Scenario - stage[0] = 'Stage 2/5: deployment of KDUs, VMs and execution environments.' - self._write_op_status( - op_id=nslcmop_id, - stage=stage - ) + stage[0] = "Stage 2/5: deployment of KDUs, VMs and execution environments." + self._write_op_status(op_id=nslcmop_id, stage=stage) stage[1] = "Deploying KDUs." # self.logger.debug(logging_text + "Before deploy_kdus") @@ -1803,8 +2653,8 @@ class NsLcm(LcmBase): # feature 1429. Add n2vc public key to needed VMs n2vc_key = self.n2vc.get_public_key() n2vc_key_list = [n2vc_key] - if self.vca_config.get("public_key"): - n2vc_key_list.append(self.vca_config["public_key"]) + if self.vca_config.public_key: + n2vc_key_list.append(self.vca_config.public_key) stage[1] = "Deploying NS at VIM." task_ro = asyncio.ensure_future( @@ -1817,7 +2667,7 @@ class NsLcm(LcmBase): db_vnfrs=db_vnfrs, db_vnfds=db_vnfds, n2vc_key_list=n2vc_key_list, - stage=stage + stage=stage, ) ) self.lcm_tasks.register("ns", nsr_id, nslcmop_id, "instantiate_RO", task_ro) @@ -1827,6 +2677,20 @@ class NsLcm(LcmBase): stage[1] = "Deploying Execution Environments." self.logger.debug(logging_text + stage[1]) + # create namespace and certificate if any helm based EE is present in the NS + if check_helm_ee_in_ns(db_vnfds): + await self.vca_map["helm-v3"].setup_ns_namespace( + name=nsr_id, + ) + # create TLS certificates + await self.vca_map["helm-v3"].create_tls_certificate( + secret_name=self.EE_TLS_NAME, + dns_prefix="*", + nsr_id=nsr_id, + usage="server auth", + namespace=nsr_id, + ) + nsi_id = None # TODO put nsi_id when this nsr belongs to a NSI for vnf_profile in get_vnf_profiles(nsd): vnfd_id = vnf_profile["vnfd-id"] @@ -1838,16 +2702,20 @@ class NsLcm(LcmBase): vdu_index = 0 vdu_name = None kdu_name = None + kdu_index = None # Get additional parameters deploy_params = {"OSM": get_osm_params(db_vnfr)} if db_vnfr.get("additionalParamsForVnf"): - deploy_params.update(parse_yaml_strings(db_vnfr["additionalParamsForVnf"].copy())) + deploy_params.update( + parse_yaml_strings(db_vnfr["additionalParamsForVnf"].copy()) + ) descriptor_config = get_configuration(vnfd, vnfd["id"]) if descriptor_config: self._deploy_n2vc( - logging_text=logging_text + "member_vnf_index={} ".format(member_vnf_index), + logging_text=logging_text + + "member_vnf_index={} ".format(member_vnf_index), db_nsr=db_nsr, db_vnfr=db_vnfr, nslcmop_id=nslcmop_id, @@ -1858,37 +2726,47 @@ class NsLcm(LcmBase): kdu_name=kdu_name, member_vnf_index=member_vnf_index, vdu_index=vdu_index, + kdu_index=kdu_index, vdu_name=vdu_name, deploy_params=deploy_params, descriptor_config=descriptor_config, base_folder=base_folder, task_instantiation_info=tasks_dict_info, - stage=stage + stage=stage, ) # Deploy charms for each VDU that supports one. for vdud in get_vdu_list(vnfd): vdu_id = vdud["id"] descriptor_config = get_configuration(vnfd, vdu_id) - vdur = find_in_list(db_vnfr["vdur"], lambda vdu: vdu["vdu-id-ref"] == vdu_id) + vdur = find_in_list( + db_vnfr["vdur"], lambda vdu: vdu["vdu-id-ref"] == vdu_id + ) if vdur.get("additionalParams"): deploy_params_vdu = parse_yaml_strings(vdur["additionalParams"]) else: deploy_params_vdu = deploy_params - deploy_params_vdu["OSM"] = get_osm_params(db_vnfr, vdu_id, vdu_count_index=0) - vdud_count = get_vdu_profile(vnfd, vdu_id).get("max-number-of-instances", 1) + deploy_params_vdu["OSM"] = get_osm_params( + db_vnfr, vdu_id, vdu_count_index=0 + ) + vdud_count = get_number_of_instances(vnfd, vdu_id) self.logger.debug("VDUD > {}".format(vdud)) - self.logger.debug("Descriptor config > {}".format(descriptor_config)) + self.logger.debug( + "Descriptor config > {}".format(descriptor_config) + ) if descriptor_config: vdu_name = None kdu_name = None + kdu_index = None for vdu_index in range(vdud_count): # TODO vnfr_params["rw_mgmt_ip"] = vdur["ip-address"] self._deploy_n2vc( - logging_text=logging_text + "member_vnf_index={}, vdu_id={}, vdu_index={} ".format( - member_vnf_index, vdu_id, vdu_index), + logging_text=logging_text + + "member_vnf_index={}, vdu_id={}, vdu_index={} ".format( + member_vnf_index, vdu_id, vdu_index + ), db_nsr=db_nsr, db_vnfr=db_vnfr, nslcmop_id=nslcmop_id, @@ -1897,6 +2775,7 @@ class NsLcm(LcmBase): vnfd_id=vnfd_id, vdu_id=vdu_id, kdu_name=kdu_name, + kdu_index=kdu_index, member_vnf_index=member_vnf_index, vdu_index=vdu_index, vdu_name=vdu_name, @@ -1904,7 +2783,7 @@ class NsLcm(LcmBase): descriptor_config=descriptor_config, base_folder=base_folder, task_instantiation_info=tasks_dict_info, - stage=stage + stage=stage, ) for kdud in get_kdu_list(vnfd): kdu_name = kdud["name"] @@ -1913,10 +2792,16 @@ class NsLcm(LcmBase): vdu_id = None vdu_index = 0 vdu_name = None - kdur = next(x for x in db_vnfr["kdur"] if x["kdu-name"] == kdu_name) + kdu_index, kdur = next( + x + for x in enumerate(db_vnfr["kdur"]) + if x[1]["kdu-name"] == kdu_name + ) deploy_params_kdu = {"OSM": get_osm_params(db_vnfr)} if kdur.get("additionalParams"): - deploy_params_kdu = parse_yaml_strings(kdur["additionalParams"]) + deploy_params_kdu.update( + parse_yaml_strings(kdur["additionalParams"].copy()) + ) self._deploy_n2vc( logging_text=logging_text, @@ -1930,12 +2815,72 @@ class NsLcm(LcmBase): kdu_name=kdu_name, member_vnf_index=member_vnf_index, vdu_index=vdu_index, + kdu_index=kdu_index, vdu_name=vdu_name, deploy_params=deploy_params_kdu, descriptor_config=descriptor_config, base_folder=base_folder, task_instantiation_info=tasks_dict_info, - stage=stage + stage=stage, + ) + + # Check if each vnf has exporter for metric collection if so update prometheus job records + if "exporters-endpoints" in vnfd.get("df")[0]: + exporter_config = vnfd.get("df")[0].get("exporters-endpoints") + self.logger.debug("exporter config :{}".format(exporter_config)) + artifact_path = "{}/{}/{}".format( + base_folder["folder"], + base_folder["pkg-dir"], + "exporter-endpoint", + ) + ee_id = None + ee_config_descriptor = exporter_config + vnfr_id = db_vnfr["id"] + rw_mgmt_ip = await self.wait_vm_up_insert_key_ro( + logging_text, + nsr_id, + vnfr_id, + vdu_id=None, + vdu_index=None, + user=None, + pub_key=None, + ) + self.logger.debug("rw_mgmt_ip:{}".format(rw_mgmt_ip)) + self.logger.debug("Artifact_path:{}".format(artifact_path)) + db_vnfr = self.db.get_one("vnfrs", {"_id": vnfr_id}) + vdu_id_for_prom = None + vdu_index_for_prom = None + for x in get_iterable(db_vnfr, "vdur"): + vdu_id_for_prom = x.get("vdu-id-ref") + vdu_index_for_prom = x.get("count-index") + prometheus_jobs = await self.extract_prometheus_scrape_jobs( + ee_id=ee_id, + artifact_path=artifact_path, + ee_config_descriptor=ee_config_descriptor, + vnfr_id=vnfr_id, + nsr_id=nsr_id, + target_ip=rw_mgmt_ip, + element_type="VDU", + vdu_id=vdu_id_for_prom, + vdu_index=vdu_index_for_prom, + ) + + self.logger.debug("Prometheus job:{}".format(prometheus_jobs)) + if prometheus_jobs: + db_nsr_update["_admin.deployed.prometheus_jobs"] = prometheus_jobs + self.update_db_2( + "nsrs", + nsr_id, + db_nsr_update, + ) + + for job in prometheus_jobs: + self.db.set_one( + "prometheus_jobs", + {"job_name": job["job_name"]}, + job, + upsert=True, + fail_on_empty=False, ) # Check if this NS has a charm configuration @@ -1946,13 +2891,16 @@ class NsLcm(LcmBase): member_vnf_index = None vdu_id = None kdu_name = None + kdu_index = None vdu_index = 0 vdu_name = None # Get additional parameters deploy_params = {"OSM": {"vim_account_id": ns_params["vimAccountId"]}} if db_nsr.get("additionalParamsForNs"): - deploy_params.update(parse_yaml_strings(db_nsr["additionalParamsForNs"].copy())) + deploy_params.update( + parse_yaml_strings(db_nsr["additionalParamsForNs"].copy()) + ) base_folder = nsd["_admin"]["storage"] self._deploy_n2vc( logging_text=logging_text, @@ -1966,25 +2914,38 @@ class NsLcm(LcmBase): kdu_name=kdu_name, member_vnf_index=member_vnf_index, vdu_index=vdu_index, + kdu_index=kdu_index, vdu_name=vdu_name, deploy_params=deploy_params, descriptor_config=descriptor_config, base_folder=base_folder, task_instantiation_info=tasks_dict_info, - stage=stage + stage=stage, ) # rest of staff will be done at finally - except (ROclient.ROClientException, DbException, LcmException, N2VCException) as e: - self.logger.error(logging_text + "Exit Exception while '{}': {}".format(stage[1], e)) + except ( + ROclient.ROClientException, + DbException, + LcmException, + N2VCException, + ) as e: + self.logger.error( + logging_text + "Exit Exception while '{}': {}".format(stage[1], e) + ) exc = e except asyncio.CancelledError: - self.logger.error(logging_text + "Cancelled Exception while '{}'".format(stage[1])) + self.logger.error( + logging_text + "Cancelled Exception while '{}'".format(stage[1]) + ) exc = "Operation was cancelled" except Exception as e: exc = traceback.format_exc() - self.logger.critical(logging_text + "Exit Exception while '{}': {}".format(stage[1], e), exc_info=True) + self.logger.critical( + logging_text + "Exit Exception while '{}': {}".format(stage[1], e), + exc_info=True, + ) finally: if exc: error_list.append(str(exc)) @@ -1993,8 +2954,14 @@ class NsLcm(LcmBase): if tasks_dict_info: stage[1] = "Waiting for instantiate pending tasks." self.logger.debug(logging_text + stage[1]) - error_list += await self._wait_for_tasks(logging_text, tasks_dict_info, timeout_ns_deploy, - stage, nslcmop_id, nsr_id=nsr_id) + error_list += await self._wait_for_tasks( + logging_text, + tasks_dict_info, + timeout_ns_deploy, + stage, + nslcmop_id, + nsr_id=nsr_id, + ) stage[1] = stage[2] = "" except asyncio.CancelledError: error_list.append("Cancelled") @@ -2019,10 +2986,16 @@ class NsLcm(LcmBase): if error_list: error_detail = ". ".join(error_list) self.logger.error(logging_text + error_detail) - error_description_nslcmop = '{} Detail: {}'.format(stage[0], error_detail) - error_description_nsr = 'Operation: INSTANTIATING.{}, {}'.format(nslcmop_id, stage[0]) + error_description_nslcmop = "{} Detail: {}".format( + stage[0], error_detail + ) + error_description_nsr = "Operation: INSTANTIATING.{}, {}".format( + nslcmop_id, stage[0] + ) - db_nsr_update["detailed-status"] = error_description_nsr + " Detail: " + error_detail + db_nsr_update["detailed-status"] = ( + error_description_nsr + " Detail: " + error_detail + ) db_nslcmop_update["detailed-status"] = error_detail nslcmop_operation_state = "FAILED" ns_state = "BROKEN" @@ -2033,7 +3006,27 @@ class NsLcm(LcmBase): db_nsr_update["detailed-status"] = "Done" db_nslcmop_update["detailed-status"] = "Done" nslcmop_operation_state = "COMPLETED" - + # Gather auto-healing and auto-scaling alerts for each vnfr + healing_alerts = [] + scaling_alerts = [] + for vnfr in self.db.get_list("vnfrs", {"nsr-id-ref": nsr_id}): + vnfd = next( + (sub for sub in db_vnfds if sub["_id"] == vnfr["vnfd-id"]), None + ) + healing_alerts = self._gather_vnfr_healing_alerts(vnfr, vnfd) + for alert in healing_alerts: + self.logger.info(f"Storing healing alert in MongoDB: {alert}") + self.db.create("alerts", alert) + + scaling_alerts = self._gather_vnfr_scaling_alerts(vnfr, vnfd) + for alert in scaling_alerts: + self.logger.info(f"Storing scaling alert in MongoDB: {alert}") + self.db.create("alerts", alert) + + alarm_alerts = self._gather_vnfr_alarm_alerts(vnfr, vnfd) + for alert in alarm_alerts: + self.logger.info(f"Storing VNF alarm alert in MongoDB: {alert}") + self.db.create("alerts", alert) if db_nsr: self._write_ns_status( nsr_id=nsr_id, @@ -2042,7 +3035,7 @@ class NsLcm(LcmBase): current_operation_id=None, error_description=error_description_nsr, error_detail=error_detail, - other_update=db_nsr_update + other_update=db_nsr_update, ) self._write_op_status( op_id=nslcmop_id, @@ -2054,64 +3047,362 @@ class NsLcm(LcmBase): if nslcmop_operation_state: try: - await self.msg.aiowrite("ns", "instantiated", {"nsr_id": nsr_id, "nslcmop_id": nslcmop_id, - "operationState": nslcmop_operation_state}, - loop=self.loop) + await self.msg.aiowrite( + "ns", + "instantiated", + { + "nsr_id": nsr_id, + "nslcmop_id": nslcmop_id, + "operationState": nslcmop_operation_state, + "startTime": db_nslcmop["startTime"], + "links": db_nslcmop["links"], + "operationParams": { + "nsInstanceId": nsr_id, + "nsdId": db_nsr["nsd-id"], + }, + }, + ) except Exception as e: - self.logger.error(logging_text + "kafka_write notification Exception {}".format(e)) + self.logger.error( + logging_text + "kafka_write notification Exception {}".format(e) + ) self.logger.debug(logging_text + "Exit") self.lcm_tasks.remove("ns", nsr_id, nslcmop_id, "ns_instantiate") - async def _add_vca_relations(self, logging_text, nsr_id, vca_index: int, - timeout: int = 3600, vca_type: str = None) -> bool: + def _get_vnfd(self, vnfd_id: str, projects_read: str, cached_vnfds: Dict[str, Any]): + if vnfd_id not in cached_vnfds: + cached_vnfds[vnfd_id] = self.db.get_one( + "vnfds", {"id": vnfd_id, "_admin.projects_read": projects_read} + ) + return cached_vnfds[vnfd_id] + + def _get_vnfr(self, nsr_id: str, vnf_profile_id: str, cached_vnfrs: Dict[str, Any]): + if vnf_profile_id not in cached_vnfrs: + cached_vnfrs[vnf_profile_id] = self.db.get_one( + "vnfrs", + { + "member-vnf-index-ref": vnf_profile_id, + "nsr-id-ref": nsr_id, + }, + ) + return cached_vnfrs[vnf_profile_id] + + def _is_deployed_vca_in_relation( + self, vca: DeployedVCA, relation: Relation + ) -> bool: + found = False + for endpoint in (relation.provider, relation.requirer): + if endpoint["kdu-resource-profile-id"]: + continue + found = ( + vca.vnf_profile_id == endpoint.vnf_profile_id + and vca.vdu_profile_id == endpoint.vdu_profile_id + and vca.execution_environment_ref == endpoint.execution_environment_ref + ) + if found: + break + return found + def _update_ee_relation_data_with_implicit_data( + self, nsr_id, nsd, ee_relation_data, cached_vnfds, vnf_profile_id: str = None + ): + ee_relation_data = safe_get_ee_relation( + nsr_id, ee_relation_data, vnf_profile_id=vnf_profile_id + ) + ee_relation_level = EELevel.get_level(ee_relation_data) + if (ee_relation_level in (EELevel.VNF, EELevel.VDU)) and not ee_relation_data[ + "execution-environment-ref" + ]: + vnf_profile = get_vnf_profile(nsd, ee_relation_data["vnf-profile-id"]) + vnfd_id = vnf_profile["vnfd-id"] + project = nsd["_admin"]["projects_read"][0] + db_vnfd = self._get_vnfd(vnfd_id, project, cached_vnfds) + entity_id = ( + vnfd_id + if ee_relation_level == EELevel.VNF + else ee_relation_data["vdu-profile-id"] + ) + ee = get_juju_ee_ref(db_vnfd, entity_id) + if not ee: + raise Exception( + f"not execution environments found for ee_relation {ee_relation_data}" + ) + ee_relation_data["execution-environment-ref"] = ee["id"] + return ee_relation_data + + def _get_ns_relations( + self, + nsr_id: str, + nsd: Dict[str, Any], + vca: DeployedVCA, + cached_vnfds: Dict[str, Any], + ) -> List[Relation]: + relations = [] + db_ns_relations = get_ns_configuration_relation_list(nsd) + for r in db_ns_relations: + provider_dict = None + requirer_dict = None + if all(key in r for key in ("provider", "requirer")): + provider_dict = r["provider"] + requirer_dict = r["requirer"] + elif "entities" in r: + provider_id = r["entities"][0]["id"] + provider_dict = { + "nsr-id": nsr_id, + "endpoint": r["entities"][0]["endpoint"], + } + if provider_id != nsd["id"]: + provider_dict["vnf-profile-id"] = provider_id + requirer_id = r["entities"][1]["id"] + requirer_dict = { + "nsr-id": nsr_id, + "endpoint": r["entities"][1]["endpoint"], + } + if requirer_id != nsd["id"]: + requirer_dict["vnf-profile-id"] = requirer_id + else: + raise Exception( + "provider/requirer or entities must be included in the relation." + ) + relation_provider = self._update_ee_relation_data_with_implicit_data( + nsr_id, nsd, provider_dict, cached_vnfds + ) + relation_requirer = self._update_ee_relation_data_with_implicit_data( + nsr_id, nsd, requirer_dict, cached_vnfds + ) + provider = EERelation(relation_provider) + requirer = EERelation(relation_requirer) + relation = Relation(r["name"], provider, requirer) + vca_in_relation = self._is_deployed_vca_in_relation(vca, relation) + if vca_in_relation: + relations.append(relation) + return relations + + def _get_vnf_relations( + self, + nsr_id: str, + nsd: Dict[str, Any], + vca: DeployedVCA, + cached_vnfds: Dict[str, Any], + ) -> List[Relation]: + relations = [] + if vca.target_element == "ns": + self.logger.debug("VCA is a NS charm, not a VNF.") + return relations + vnf_profile = get_vnf_profile(nsd, vca.vnf_profile_id) + vnf_profile_id = vnf_profile["id"] + vnfd_id = vnf_profile["vnfd-id"] + project = nsd["_admin"]["projects_read"][0] + db_vnfd = self._get_vnfd(vnfd_id, project, cached_vnfds) + db_vnf_relations = get_relation_list(db_vnfd, vnfd_id) + for r in db_vnf_relations: + provider_dict = None + requirer_dict = None + if all(key in r for key in ("provider", "requirer")): + provider_dict = r["provider"] + requirer_dict = r["requirer"] + elif "entities" in r: + provider_id = r["entities"][0]["id"] + provider_dict = { + "nsr-id": nsr_id, + "vnf-profile-id": vnf_profile_id, + "endpoint": r["entities"][0]["endpoint"], + } + if provider_id != vnfd_id: + provider_dict["vdu-profile-id"] = provider_id + requirer_id = r["entities"][1]["id"] + requirer_dict = { + "nsr-id": nsr_id, + "vnf-profile-id": vnf_profile_id, + "endpoint": r["entities"][1]["endpoint"], + } + if requirer_id != vnfd_id: + requirer_dict["vdu-profile-id"] = requirer_id + else: + raise Exception( + "provider/requirer or entities must be included in the relation." + ) + relation_provider = self._update_ee_relation_data_with_implicit_data( + nsr_id, nsd, provider_dict, cached_vnfds, vnf_profile_id=vnf_profile_id + ) + relation_requirer = self._update_ee_relation_data_with_implicit_data( + nsr_id, nsd, requirer_dict, cached_vnfds, vnf_profile_id=vnf_profile_id + ) + provider = EERelation(relation_provider) + requirer = EERelation(relation_requirer) + relation = Relation(r["name"], provider, requirer) + vca_in_relation = self._is_deployed_vca_in_relation(vca, relation) + if vca_in_relation: + relations.append(relation) + return relations + + def _get_kdu_resource_data( + self, + ee_relation: EERelation, + db_nsr: Dict[str, Any], + cached_vnfds: Dict[str, Any], + ) -> DeployedK8sResource: + nsd = get_nsd(db_nsr) + vnf_profiles = get_vnf_profiles(nsd) + vnfd_id = find_in_list( + vnf_profiles, + lambda vnf_profile: vnf_profile["id"] == ee_relation.vnf_profile_id, + )["vnfd-id"] + project = nsd["_admin"]["projects_read"][0] + db_vnfd = self._get_vnfd(vnfd_id, project, cached_vnfds) + kdu_resource_profile = get_kdu_resource_profile( + db_vnfd, ee_relation.kdu_resource_profile_id + ) + kdu_name = kdu_resource_profile["kdu-name"] + deployed_kdu, _ = get_deployed_kdu( + db_nsr.get("_admin", ()).get("deployed", ()), + kdu_name, + ee_relation.vnf_profile_id, + ) + deployed_kdu.update({"resource-name": kdu_resource_profile["resource-name"]}) + return deployed_kdu + + def _get_deployed_component( + self, + ee_relation: EERelation, + db_nsr: Dict[str, Any], + cached_vnfds: Dict[str, Any], + ) -> DeployedComponent: + nsr_id = db_nsr["_id"] + deployed_component = None + ee_level = EELevel.get_level(ee_relation) + if ee_level == EELevel.NS: + vca = get_deployed_vca(db_nsr, {"vdu_id": None, "member-vnf-index": None}) + if vca: + deployed_component = DeployedVCA(nsr_id, vca) + elif ee_level == EELevel.VNF: + vca = get_deployed_vca( + db_nsr, + { + "vdu_id": None, + "member-vnf-index": ee_relation.vnf_profile_id, + "ee_descriptor_id": ee_relation.execution_environment_ref, + }, + ) + if vca: + deployed_component = DeployedVCA(nsr_id, vca) + elif ee_level == EELevel.VDU: + vca = get_deployed_vca( + db_nsr, + { + "vdu_id": ee_relation.vdu_profile_id, + "member-vnf-index": ee_relation.vnf_profile_id, + "ee_descriptor_id": ee_relation.execution_environment_ref, + }, + ) + if vca: + deployed_component = DeployedVCA(nsr_id, vca) + elif ee_level == EELevel.KDU: + kdu_resource_data = self._get_kdu_resource_data( + ee_relation, db_nsr, cached_vnfds + ) + if kdu_resource_data: + deployed_component = DeployedK8sResource(kdu_resource_data) + return deployed_component + + async def _add_relation( + self, + relation: Relation, + vca_type: str, + db_nsr: Dict[str, Any], + cached_vnfds: Dict[str, Any], + cached_vnfrs: Dict[str, Any], + ) -> bool: + deployed_provider = self._get_deployed_component( + relation.provider, db_nsr, cached_vnfds + ) + deployed_requirer = self._get_deployed_component( + relation.requirer, db_nsr, cached_vnfds + ) + if ( + deployed_provider + and deployed_requirer + and deployed_provider.config_sw_installed + and deployed_requirer.config_sw_installed + ): + provider_db_vnfr = ( + self._get_vnfr( + relation.provider.nsr_id, + relation.provider.vnf_profile_id, + cached_vnfrs, + ) + if relation.provider.vnf_profile_id + else None + ) + requirer_db_vnfr = ( + self._get_vnfr( + relation.requirer.nsr_id, + relation.requirer.vnf_profile_id, + cached_vnfrs, + ) + if relation.requirer.vnf_profile_id + else None + ) + provider_vca_id = self.get_vca_id(provider_db_vnfr, db_nsr) + requirer_vca_id = self.get_vca_id(requirer_db_vnfr, db_nsr) + provider_relation_endpoint = RelationEndpoint( + deployed_provider.ee_id, + provider_vca_id, + relation.provider.endpoint, + ) + requirer_relation_endpoint = RelationEndpoint( + deployed_requirer.ee_id, + requirer_vca_id, + relation.requirer.endpoint, + ) + try: + await self.vca_map[vca_type].add_relation( + provider=provider_relation_endpoint, + requirer=requirer_relation_endpoint, + ) + except N2VCException as exception: + self.logger.error(exception) + raise LcmException(exception) + return True + return False + + async def _add_vca_relations( + self, + logging_text, + nsr_id, + vca_type: str, + vca_index: int, + timeout: int = 3600, + ) -> bool: # steps: # 1. find all relations for this VCA # 2. wait for other peers related # 3. add relations try: - vca_type = vca_type or "lxc_proxy_charm" - # STEP 1: find all relations for this VCA # read nsr record db_nsr = self.db.get_one("nsrs", {"_id": nsr_id}) - nsd = self.db.get_one("nsds", {"_id": db_nsr["nsd-id"]}) + nsd = get_nsd(db_nsr) # this VCA data - my_vca = deep_get(db_nsr, ('_admin', 'deployed', 'VCA'))[vca_index] - - # read all ns-configuration relations - ns_relations = list() - db_ns_relations = deep_get(nsd, ('ns-configuration', 'relation')) - if db_ns_relations: - for r in db_ns_relations: - # check if this VCA is in the relation - if my_vca.get('member-vnf-index') in\ - (r.get('entities')[0].get('id'), r.get('entities')[1].get('id')): - ns_relations.append(r) - - # read all vnf-configuration relations - vnf_relations = list() - db_vnfd_list = db_nsr.get('vnfd-id') - if db_vnfd_list: - for vnfd in db_vnfd_list: - db_vnfd = self.db.get_one("vnfds", {"_id": vnfd}) - db_vnf_relations = get_configuration(db_vnfd, db_vnfd["id"]).get("relation", []) - if db_vnf_relations: - for r in db_vnf_relations: - # check if this VCA is in the relation - if my_vca.get('vdu_id') in (r.get('entities')[0].get('id'), r.get('entities')[1].get('id')): - vnf_relations.append(r) + deployed_vca_dict = get_deployed_vca_list(db_nsr)[vca_index] + my_vca = DeployedVCA(nsr_id, deployed_vca_dict) + + cached_vnfds = {} + cached_vnfrs = {} + relations = [] + relations.extend(self._get_ns_relations(nsr_id, nsd, my_vca, cached_vnfds)) + relations.extend(self._get_vnf_relations(nsr_id, nsd, my_vca, cached_vnfds)) # if no relations, terminate - if not ns_relations and not vnf_relations: - self.logger.debug(logging_text + ' No relations') + if not relations: + self.logger.debug(logging_text + " No relations") return True - self.logger.debug(logging_text + ' adding relations\n {}\n {}'.format(ns_relations, vnf_relations)) + self.logger.debug(logging_text + " adding relations {}".format(relations)) # add all relations start = time() @@ -2119,132 +3410,106 @@ class NsLcm(LcmBase): # check timeout now = time() if now - start >= timeout: - self.logger.error(logging_text + ' : timeout adding relations') + self.logger.error(logging_text + " : timeout adding relations") return False - # reload nsr from database (we need to update record: _admin.deloyed.VCA) + # reload nsr from database (we need to update record: _admin.deployed.VCA) db_nsr = self.db.get_one("nsrs", {"_id": nsr_id}) - # for each defined NS relation, find the VCA's related - for r in ns_relations.copy(): - from_vca_ee_id = None - to_vca_ee_id = None - from_vca_endpoint = None - to_vca_endpoint = None - vca_list = deep_get(db_nsr, ('_admin', 'deployed', 'VCA')) - for vca in vca_list: - if vca.get('member-vnf-index') == r.get('entities')[0].get('id') \ - and vca.get('config_sw_installed'): - from_vca_ee_id = vca.get('ee_id') - from_vca_endpoint = r.get('entities')[0].get('endpoint') - if vca.get('member-vnf-index') == r.get('entities')[1].get('id') \ - and vca.get('config_sw_installed'): - to_vca_ee_id = vca.get('ee_id') - to_vca_endpoint = r.get('entities')[1].get('endpoint') - if from_vca_ee_id and to_vca_ee_id: - # add relation - await self.vca_map[vca_type].add_relation( - ee_id_1=from_vca_ee_id, - ee_id_2=to_vca_ee_id, - endpoint_1=from_vca_endpoint, - endpoint_2=to_vca_endpoint) - # remove entry from relations list - ns_relations.remove(r) - else: - # check failed peers - try: - vca_status_list = db_nsr.get('configurationStatus') - if vca_status_list: - for i in range(len(vca_list)): - vca = vca_list[i] - vca_status = vca_status_list[i] - if vca.get('member-vnf-index') == r.get('entities')[0].get('id'): - if vca_status.get('status') == 'BROKEN': - # peer broken: remove relation from list - ns_relations.remove(r) - if vca.get('member-vnf-index') == r.get('entities')[1].get('id'): - if vca_status.get('status') == 'BROKEN': - # peer broken: remove relation from list - ns_relations.remove(r) - except Exception: - # ignore - pass - - # for each defined VNF relation, find the VCA's related - for r in vnf_relations.copy(): - from_vca_ee_id = None - to_vca_ee_id = None - from_vca_endpoint = None - to_vca_endpoint = None - vca_list = deep_get(db_nsr, ('_admin', 'deployed', 'VCA')) - for vca in vca_list: - key_to_check = "vdu_id" - if vca.get("vdu_id") is None: - key_to_check = "vnfd_id" - if vca.get(key_to_check) == r.get('entities')[0].get('id') and vca.get('config_sw_installed'): - from_vca_ee_id = vca.get('ee_id') - from_vca_endpoint = r.get('entities')[0].get('endpoint') - if vca.get(key_to_check) == r.get('entities')[1].get('id') and vca.get('config_sw_installed'): - to_vca_ee_id = vca.get('ee_id') - to_vca_endpoint = r.get('entities')[1].get('endpoint') - if from_vca_ee_id and to_vca_ee_id: - # add relation - await self.vca_map[vca_type].add_relation( - ee_id_1=from_vca_ee_id, - ee_id_2=to_vca_ee_id, - endpoint_1=from_vca_endpoint, - endpoint_2=to_vca_endpoint) - # remove entry from relations list - vnf_relations.remove(r) - else: - # check failed peers - try: - vca_status_list = db_nsr.get('configurationStatus') - if vca_status_list: - for i in range(len(vca_list)): - vca = vca_list[i] - vca_status = vca_status_list[i] - if vca.get('vdu_id') == r.get('entities')[0].get('id'): - if vca_status.get('status') == 'BROKEN': - # peer broken: remove relation from list - vnf_relations.remove(r) - if vca.get('vdu_id') == r.get('entities')[1].get('id'): - if vca_status.get('status') == 'BROKEN': - # peer broken: remove relation from list - vnf_relations.remove(r) - except Exception: - # ignore - pass - - # wait for next try - await asyncio.sleep(5.0) + # for each relation, find the VCA's related + for relation in relations.copy(): + added = await self._add_relation( + relation, + vca_type, + db_nsr, + cached_vnfds, + cached_vnfrs, + ) + if added: + relations.remove(relation) - if not ns_relations and not vnf_relations: - self.logger.debug('Relations added') + if not relations: + self.logger.debug("Relations added") break + await asyncio.sleep(5.0) return True except Exception as e: - self.logger.warn(logging_text + ' ERROR adding relations: {}'.format(e)) + self.logger.warn(logging_text + " ERROR adding relations: {}".format(e)) return False - async def _install_kdu(self, nsr_id: str, nsr_db_path: str, vnfr_data: dict, kdu_index: int, kdud: dict, - vnfd: dict, k8s_instance_info: dict, k8params: dict = None, timeout: int = 600): - + async def _install_kdu( + self, + nsr_id: str, + nsr_db_path: str, + vnfr_data: dict, + kdu_index: int, + kdud: dict, + vnfd: dict, + k8s_instance_info: dict, + k8params: dict = None, + timeout: int = 600, + vca_id: str = None, + ): try: k8sclustertype = k8s_instance_info["k8scluster-type"] # Instantiate kdu - db_dict_install = {"collection": "nsrs", - "filter": {"_id": nsr_id}, - "path": nsr_db_path} + db_dict_install = { + "collection": "nsrs", + "filter": {"_id": nsr_id}, + "path": nsr_db_path, + } - kdu_instance = self.k8scluster_map[k8sclustertype].generate_kdu_instance_name( - db_dict=db_dict_install, - kdu_model=k8s_instance_info["kdu-model"], - kdu_name=k8s_instance_info["kdu-name"], + if k8s_instance_info.get("kdu-deployment-name"): + kdu_instance = k8s_instance_info.get("kdu-deployment-name") + else: + kdu_instance = self.k8scluster_map[ + k8sclustertype + ].generate_kdu_instance_name( + db_dict=db_dict_install, + kdu_model=k8s_instance_info["kdu-model"], + kdu_name=k8s_instance_info["kdu-name"], + ) + + # Update the nsrs table with the kdu-instance value + self.update_db_2( + item="nsrs", + _id=nsr_id, + _desc={nsr_db_path + ".kdu-instance": kdu_instance}, ) - self.update_db_2("nsrs", nsr_id, {nsr_db_path + ".kdu-instance": kdu_instance}) + + # Update the nsrs table with the actual namespace being used, if the k8scluster-type is `juju` or + # `juju-bundle`. This verification is needed because there is not a standard/homogeneous namespace + # between the Helm Charts and Juju Bundles-based KNFs. If we found a way of having an homogeneous + # namespace, this first verification could be removed, and the next step would be done for any kind + # of KNF. + # TODO -> find a way to have an homogeneous namespace between the Helm Charts and Juju Bundles-based + # KNFs (Bug 2027: https://osm.etsi.org/bugzilla/show_bug.cgi?id=2027) + if k8sclustertype in ("juju", "juju-bundle"): + # First, verify if the current namespace is present in the `_admin.projects_read` (if not, it means + # that the user passed a namespace which he wants its KDU to be deployed in) + if ( + self.db.count( + table="nsrs", + q_filter={ + "_id": nsr_id, + "_admin.projects_write": k8s_instance_info["namespace"], + "_admin.projects_read": k8s_instance_info["namespace"], + }, + ) + > 0 + ): + self.logger.debug( + f"Updating namespace/model for Juju Bundle from {k8s_instance_info['namespace']} to {kdu_instance}" + ) + self.update_db_2( + item="nsrs", + _id=nsr_id, + _desc={f"{nsr_db_path}.namespace": kdu_instance}, + ) + k8s_instance_info["namespace"] = kdu_instance + await self.k8scluster_map[k8sclustertype].install( cluster_uuid=k8s_instance_info["k8scluster-uuid"], kdu_model=k8s_instance_info["kdu-model"], @@ -2255,20 +3520,31 @@ class NsLcm(LcmBase): kdu_name=k8s_instance_info["kdu-name"], namespace=k8s_instance_info["namespace"], kdu_instance=kdu_instance, + vca_id=vca_id, ) - self.update_db_2("nsrs", nsr_id, {nsr_db_path + ".kdu-instance": kdu_instance}) # Obtain services to obtain management service ip services = await self.k8scluster_map[k8sclustertype].get_services( cluster_uuid=k8s_instance_info["k8scluster-uuid"], kdu_instance=kdu_instance, - namespace=k8s_instance_info["namespace"]) + namespace=k8s_instance_info["namespace"], + ) # Obtain management service info (if exists) vnfr_update_dict = {} + kdu_config = get_configuration(vnfd, kdud["name"]) + if kdu_config: + target_ee_list = kdu_config.get("execution-environment-list", []) + else: + target_ee_list = [] + if services: vnfr_update_dict["kdur.{}.services".format(kdu_index)] = services - mgmt_services = [service for service in kdud.get("service", []) if service.get("mgmt-service")] + mgmt_services = [ + service + for service in kdud.get("service", []) + if service.get("mgmt-service") + ] for mgmt_service in mgmt_services: for service in services: if service["name"].startswith(mgmt_service["name"]): @@ -2277,55 +3553,107 @@ class NsLcm(LcmBase): if isinstance(ip, list) and len(ip) == 1: ip = ip[0] - vnfr_update_dict["kdur.{}.ip-address".format(kdu_index)] = ip + vnfr_update_dict[ + "kdur.{}.ip-address".format(kdu_index) + ] = ip # Check if must update also mgmt ip at the vnf - service_external_cp = mgmt_service.get("external-connection-point-ref") + service_external_cp = mgmt_service.get( + "external-connection-point-ref" + ) if service_external_cp: - if deep_get(vnfd, ("mgmt-interface", "cp")) == service_external_cp: + if ( + deep_get(vnfd, ("mgmt-interface", "cp")) + == service_external_cp + ): vnfr_update_dict["ip-address"] = ip + if find_in_list( + target_ee_list, + lambda ee: ee.get( + "external-connection-point-ref", "" + ) + == service_external_cp, + ): + vnfr_update_dict[ + "kdur.{}.ip-address".format(kdu_index) + ] = ip break else: - self.logger.warn("Mgmt service name: {} not found".format(mgmt_service["name"])) + self.logger.warn( + "Mgmt service name: {} not found".format( + mgmt_service["name"] + ) + ) vnfr_update_dict["kdur.{}.status".format(kdu_index)] = "READY" self.update_db_2("vnfrs", vnfr_data.get("_id"), vnfr_update_dict) kdu_config = get_configuration(vnfd, k8s_instance_info["kdu-name"]) - if kdu_config and kdu_config.get("initial-config-primitive") and \ - get_juju_ee_ref(vnfd, k8s_instance_info["kdu-name"]) is None: - initial_config_primitive_list = kdu_config.get("initial-config-primitive") + if ( + kdu_config + and kdu_config.get("initial-config-primitive") + and get_juju_ee_ref(vnfd, k8s_instance_info["kdu-name"]) is None + ): + initial_config_primitive_list = kdu_config.get( + "initial-config-primitive" + ) initial_config_primitive_list.sort(key=lambda val: int(val["seq"])) for initial_config_primitive in initial_config_primitive_list: - primitive_params_ = self._map_primitive_params(initial_config_primitive, {}, {}) + primitive_params_ = self._map_primitive_params( + initial_config_primitive, {}, {} + ) await asyncio.wait_for( self.k8scluster_map[k8sclustertype].exec_primitive( cluster_uuid=k8s_instance_info["k8scluster-uuid"], kdu_instance=kdu_instance, primitive_name=initial_config_primitive["name"], - params=primitive_params_, db_dict=db_dict_install), - timeout=timeout) + params=primitive_params_, + db_dict=db_dict_install, + vca_id=vca_id, + ), + timeout=timeout, + ) except Exception as e: # Prepare update db with error and raise exception try: - self.update_db_2("nsrs", nsr_id, {nsr_db_path + ".detailed-status": str(e)}) - self.update_db_2("vnfrs", vnfr_data.get("_id"), {"kdur.{}.status".format(kdu_index): "ERROR"}) - except Exception: + self.update_db_2( + "nsrs", nsr_id, {nsr_db_path + ".detailed-status": str(e)} + ) + self.update_db_2( + "vnfrs", + vnfr_data.get("_id"), + {"kdur.{}.status".format(kdu_index): "ERROR"}, + ) + except Exception as error: # ignore to keep original exception - pass + self.logger.warning( + f"An exception occurred while updating DB: {str(error)}" + ) # reraise original error raise return kdu_instance - async def deploy_kdus(self, logging_text, nsr_id, nslcmop_id, db_vnfrs, db_vnfds, task_instantiation_info): + async def deploy_kdus( + self, + logging_text, + nsr_id, + nslcmop_id, + db_vnfrs, + db_vnfds, + task_instantiation_info, + ): # Launch kdus if present in the descriptor - k8scluster_id_2_uuic = {"helm-chart-v3": {}, "helm-chart": {}, "juju-bundle": {}} + k8scluster_id_2_uuic = { + "helm-chart-v3": {}, + "helm-chart": {}, + "juju-bundle": {}, + } async def _get_cluster_id(cluster_id, cluster_type): nonlocal k8scluster_id_2_uuic @@ -2333,13 +3661,19 @@ class NsLcm(LcmBase): return k8scluster_id_2_uuic[cluster_type][cluster_id] # check if K8scluster is creating and wait look if previous tasks in process - task_name, task_dependency = self.lcm_tasks.lookfor_related("k8scluster", cluster_id) + task_name, task_dependency = self.lcm_tasks.lookfor_related( + "k8scluster", cluster_id + ) if task_dependency: - text = "Waiting for related tasks '{}' on k8scluster {} to be completed".format(task_name, cluster_id) + text = "Waiting for related tasks '{}' on k8scluster {} to be completed".format( + task_name, cluster_id + ) self.logger.debug(logging_text + text) await asyncio.wait(task_dependency, timeout=3600) - db_k8scluster = self.db.get_one("k8sclusters", {"_id": cluster_id}, fail_on_empty=False) + db_k8scluster = self.db.get_one( + "k8sclusters", {"_id": cluster_id}, fail_on_empty=False + ) if not db_k8scluster: raise LcmException("K8s cluster {} cannot be found".format(cluster_id)) @@ -2348,22 +3682,40 @@ class NsLcm(LcmBase): if cluster_type == "helm-chart-v3": try: # backward compatibility for existing clusters that have not been initialized for helm v3 - k8s_credentials = yaml.safe_dump(db_k8scluster.get("credentials")) - k8s_id, uninstall_sw = await self.k8sclusterhelm3.init_env(k8s_credentials, - reuse_cluster_uuid=cluster_id) + k8s_credentials = yaml.safe_dump( + db_k8scluster.get("credentials") + ) + k8s_id, uninstall_sw = await self.k8sclusterhelm3.init_env( + k8s_credentials, reuse_cluster_uuid=cluster_id + ) db_k8scluster_update = {} db_k8scluster_update["_admin.helm-chart-v3.error_msg"] = None db_k8scluster_update["_admin.helm-chart-v3.id"] = k8s_id - db_k8scluster_update["_admin.helm-chart-v3.created"] = uninstall_sw - db_k8scluster_update["_admin.helm-chart-v3.operationalState"] = "ENABLED" - self.update_db_2("k8sclusters", cluster_id, db_k8scluster_update) + db_k8scluster_update[ + "_admin.helm-chart-v3.created" + ] = uninstall_sw + db_k8scluster_update[ + "_admin.helm-chart-v3.operationalState" + ] = "ENABLED" + self.update_db_2( + "k8sclusters", cluster_id, db_k8scluster_update + ) except Exception as e: - self.logger.error(logging_text + "error initializing helm-v3 cluster: {}".format(str(e))) - raise LcmException("K8s cluster '{}' has not been initialized for '{}'".format(cluster_id, - cluster_type)) + self.logger.error( + logging_text + + "error initializing helm-v3 cluster: {}".format(str(e)) + ) + raise LcmException( + "K8s cluster '{}' has not been initialized for '{}'".format( + cluster_id, cluster_type + ) + ) else: - raise LcmException("K8s cluster '{}' has not been initialized for '{}'". - format(cluster_id, cluster_type)) + raise LcmException( + "K8s cluster '{}' has not been initialized for '{}'".format( + cluster_id, cluster_type + ) + ) k8scluster_id_2_uuic[cluster_type][cluster_id] = k8s_id return k8s_id @@ -2378,86 +3730,170 @@ class NsLcm(LcmBase): updated_v3_cluster_list = [] for vnfr_data in db_vnfrs.values(): + vca_id = self.get_vca_id(vnfr_data, {}) for kdu_index, kdur in enumerate(get_iterable(vnfr_data, "kdur")): # Step 0: Prepare and set parameters desc_params = parse_yaml_strings(kdur.get("additionalParams")) - vnfd_id = vnfr_data.get('vnfd-id') - vnfd_with_id = find_in_list(db_vnfds, lambda vnfd: vnfd["_id"] == vnfd_id) - kdud = next(kdud for kdud in vnfd_with_id["kdu"] if kdud["name"] == kdur["kdu-name"]) + vnfd_id = vnfr_data.get("vnfd-id") + vnfd_with_id = find_in_list( + db_vnfds, lambda vnfd: vnfd["_id"] == vnfd_id + ) + kdud = next( + kdud + for kdud in vnfd_with_id["kdu"] + if kdud["name"] == kdur["kdu-name"] + ) namespace = kdur.get("k8s-namespace") + kdu_deployment_name = kdur.get("kdu-deployment-name") if kdur.get("helm-chart"): kdumodel = kdur["helm-chart"] # Default version: helm3, if helm-version is v2 assign v2 k8sclustertype = "helm-chart-v3" self.logger.debug("kdur: {}".format(kdur)) - if kdur.get("helm-version") and kdur.get("helm-version") == "v2": + if ( + kdur.get("helm-version") + and kdur.get("helm-version") == "v2" + ): k8sclustertype = "helm-chart" elif kdur.get("juju-bundle"): kdumodel = kdur["juju-bundle"] k8sclustertype = "juju-bundle" else: - raise LcmException("kdu type for kdu='{}.{}' is neither helm-chart nor " - "juju-bundle. Maybe an old NBI version is running". - format(vnfr_data["member-vnf-index-ref"], kdur["kdu-name"])) + raise LcmException( + "kdu type for kdu='{}.{}' is neither helm-chart nor " + "juju-bundle. Maybe an old NBI version is running".format( + vnfr_data["member-vnf-index-ref"], kdur["kdu-name"] + ) + ) # check if kdumodel is a file and exists try: - vnfd_with_id = find_in_list(db_vnfds, lambda vnfd: vnfd["_id"] == vnfd_id) - storage = deep_get(vnfd_with_id, ('_admin', 'storage')) - if storage and storage.get('pkg-dir'): # may be not present if vnfd has not artifacts + vnfd_with_id = find_in_list( + db_vnfds, lambda vnfd: vnfd["_id"] == vnfd_id + ) + storage = deep_get(vnfd_with_id, ("_admin", "storage")) + if storage: # may be not present if vnfd has not artifacts # path format: /vnfdid/pkkdir/helm-charts|juju-bundles/kdumodel - filename = '{}/{}/{}s/{}'.format(storage["folder"], storage["pkg-dir"], k8sclustertype, - kdumodel) - if self.fs.file_exists(filename, mode='file') or self.fs.file_exists(filename, mode='dir'): + if storage["pkg-dir"]: + filename = "{}/{}/{}s/{}".format( + storage["folder"], + storage["pkg-dir"], + k8sclustertype, + kdumodel, + ) + else: + filename = "{}/Scripts/{}s/{}".format( + storage["folder"], + k8sclustertype, + kdumodel, + ) + if self.fs.file_exists( + filename, mode="file" + ) or self.fs.file_exists(filename, mode="dir"): kdumodel = self.fs.path + filename except (asyncio.TimeoutError, asyncio.CancelledError): raise - except Exception: # it is not a file - pass + except Exception as e: # it is not a file + self.logger.warning(f"An exception occurred: {str(e)}") k8s_cluster_id = kdur["k8s-cluster"]["id"] - step = "Synchronize repos for k8s cluster '{}'".format(k8s_cluster_id) + step = "Synchronize repos for k8s cluster '{}'".format( + k8s_cluster_id + ) cluster_uuid = await _get_cluster_id(k8s_cluster_id, k8sclustertype) # Synchronize repos - if (k8sclustertype == "helm-chart" and cluster_uuid not in updated_cluster_list)\ - or (k8sclustertype == "helm-chart-v3" and cluster_uuid not in updated_v3_cluster_list): + if ( + k8sclustertype == "helm-chart" + and cluster_uuid not in updated_cluster_list + ) or ( + k8sclustertype == "helm-chart-v3" + and cluster_uuid not in updated_v3_cluster_list + ): del_repo_list, added_repo_dict = await asyncio.ensure_future( - self.k8scluster_map[k8sclustertype].synchronize_repos(cluster_uuid=cluster_uuid)) + self.k8scluster_map[k8sclustertype].synchronize_repos( + cluster_uuid=cluster_uuid + ) + ) if del_repo_list or added_repo_dict: if k8sclustertype == "helm-chart": - unset = {'_admin.helm_charts_added.' + item: None for item in del_repo_list} - updated = {'_admin.helm_charts_added.' + - item: name for item, name in added_repo_dict.items()} + unset = { + "_admin.helm_charts_added." + item: None + for item in del_repo_list + } + updated = { + "_admin.helm_charts_added." + item: name + for item, name in added_repo_dict.items() + } updated_cluster_list.append(cluster_uuid) elif k8sclustertype == "helm-chart-v3": - unset = {'_admin.helm_charts_v3_added.' + item: None for item in del_repo_list} - updated = {'_admin.helm_charts_v3_added.' + - item: name for item, name in added_repo_dict.items()} + unset = { + "_admin.helm_charts_v3_added." + item: None + for item in del_repo_list + } + updated = { + "_admin.helm_charts_v3_added." + item: name + for item, name in added_repo_dict.items() + } updated_v3_cluster_list.append(cluster_uuid) - self.logger.debug(logging_text + "repos synchronized on k8s cluster " - "'{}' to_delete: {}, to_add: {}". - format(k8s_cluster_id, del_repo_list, added_repo_dict)) - self.db.set_one("k8sclusters", {"_id": k8s_cluster_id}, updated, unset=unset) + self.logger.debug( + logging_text + "repos synchronized on k8s cluster " + "'{}' to_delete: {}, to_add: {}".format( + k8s_cluster_id, del_repo_list, added_repo_dict + ) + ) + self.db.set_one( + "k8sclusters", + {"_id": k8s_cluster_id}, + updated, + unset=unset, + ) # Instantiate kdu - step = "Instantiating KDU {}.{} in k8s cluster {}".format(vnfr_data["member-vnf-index-ref"], - kdur["kdu-name"], k8s_cluster_id) - k8s_instance_info = {"kdu-instance": None, - "k8scluster-uuid": cluster_uuid, - "k8scluster-type": k8sclustertype, - "member-vnf-index": vnfr_data["member-vnf-index-ref"], - "kdu-name": kdur["kdu-name"], - "kdu-model": kdumodel, - "namespace": namespace} + step = "Instantiating KDU {}.{} in k8s cluster {}".format( + vnfr_data["member-vnf-index-ref"], + kdur["kdu-name"], + k8s_cluster_id, + ) + k8s_instance_info = { + "kdu-instance": None, + "k8scluster-uuid": cluster_uuid, + "k8scluster-type": k8sclustertype, + "member-vnf-index": vnfr_data["member-vnf-index-ref"], + "kdu-name": kdur["kdu-name"], + "kdu-model": kdumodel, + "namespace": namespace, + "kdu-deployment-name": kdu_deployment_name, + } db_path = "_admin.deployed.K8s.{}".format(index) db_nsr_update[db_path] = k8s_instance_info self.update_db_2("nsrs", nsr_id, db_nsr_update) - vnfd_with_id = find_in_list(db_vnfds, lambda vnf: vnf["_id"] == vnfd_id) + vnfd_with_id = find_in_list( + db_vnfds, lambda vnf: vnf["_id"] == vnfd_id + ) task = asyncio.ensure_future( - self._install_kdu(nsr_id, db_path, vnfr_data, kdu_index, kdud, vnfd_with_id, - k8s_instance_info, k8params=desc_params, timeout=600)) - self.lcm_tasks.register("ns", nsr_id, nslcmop_id, "instantiate_KDU-{}".format(index), task) - task_instantiation_info[task] = "Deploying KDU {}".format(kdur["kdu-name"]) + self._install_kdu( + nsr_id, + db_path, + vnfr_data, + kdu_index, + kdud, + vnfd_with_id, + k8s_instance_info, + k8params=desc_params, + timeout=1800, + vca_id=vca_id, + ) + ) + self.lcm_tasks.register( + "ns", + nsr_id, + nslcmop_id, + "instantiate_KDU-{}".format(index), + task, + ) + task_instantiation_info[task] = "Deploying KDU {}".format( + kdur["kdu-name"] + ) index += 1 @@ -2474,54 +3910,100 @@ class NsLcm(LcmBase): if db_nsr_update: self.update_db_2("nsrs", nsr_id, db_nsr_update) - def _deploy_n2vc(self, logging_text, db_nsr, db_vnfr, nslcmop_id, nsr_id, nsi_id, vnfd_id, vdu_id, - kdu_name, member_vnf_index, vdu_index, vdu_name, deploy_params, descriptor_config, - base_folder, task_instantiation_info, stage): + def _deploy_n2vc( + self, + logging_text, + db_nsr, + db_vnfr, + nslcmop_id, + nsr_id, + nsi_id, + vnfd_id, + vdu_id, + kdu_name, + member_vnf_index, + vdu_index, + kdu_index, + vdu_name, + deploy_params, + descriptor_config, + base_folder, + task_instantiation_info, + stage, + ): # launch instantiate_N2VC in a asyncio task and register task object # Look where information of this charm is at database ._admin.deployed.VCA # if not found, create one entry and update database # fill db_nsr._admin.deployed.VCA. - self.logger.debug(logging_text + "_deploy_n2vc vnfd_id={}, vdu_id={}".format(vnfd_id, vdu_id)) + self.logger.debug( + logging_text + "_deploy_n2vc vnfd_id={}, vdu_id={}".format(vnfd_id, vdu_id) + ) + + charm_name = "" + get_charm_name = False if "execution-environment-list" in descriptor_config: ee_list = descriptor_config.get("execution-environment-list", []) + elif "juju" in descriptor_config: + ee_list = [descriptor_config] # ns charms + if "execution-environment-list" not in descriptor_config: + # charm name is only required for ns charms + get_charm_name = True else: # other types as script are not supported ee_list = [] for ee_item in ee_list: - self.logger.debug(logging_text + "_deploy_n2vc ee_item juju={}, helm={}".format(ee_item.get('juju'), - ee_item.get("helm-chart"))) + self.logger.debug( + logging_text + + "_deploy_n2vc ee_item juju={}, helm={}".format( + ee_item.get("juju"), ee_item.get("helm-chart") + ) + ) ee_descriptor_id = ee_item.get("id") if ee_item.get("juju"): - vca_name = ee_item['juju'].get('charm') - vca_type = "lxc_proxy_charm" if ee_item['juju'].get('charm') is not None else "native_charm" - if ee_item['juju'].get('cloud') == "k8s": + vca_name = ee_item["juju"].get("charm") + if get_charm_name: + charm_name = self.find_charm_name(db_nsr, str(vca_name)) + vca_type = ( + "lxc_proxy_charm" + if ee_item["juju"].get("charm") is not None + else "native_charm" + ) + if ee_item["juju"].get("cloud") == "k8s": vca_type = "k8s_proxy_charm" - elif ee_item['juju'].get('proxy') is False: + elif ee_item["juju"].get("proxy") is False: vca_type = "native_charm" elif ee_item.get("helm-chart"): - vca_name = ee_item['helm-chart'] + vca_name = ee_item["helm-chart"] if ee_item.get("helm-version") and ee_item.get("helm-version") == "v2": vca_type = "helm" else: vca_type = "helm-v3" else: - self.logger.debug(logging_text + "skipping non juju neither charm configuration") + self.logger.debug( + logging_text + "skipping non juju neither charm configuration" + ) continue vca_index = -1 - for vca_index, vca_deployed in enumerate(db_nsr["_admin"]["deployed"]["VCA"]): + for vca_index, vca_deployed in enumerate( + db_nsr["_admin"]["deployed"]["VCA"] + ): if not vca_deployed: continue - if vca_deployed.get("member-vnf-index") == member_vnf_index and \ - vca_deployed.get("vdu_id") == vdu_id and \ - vca_deployed.get("kdu_name") == kdu_name and \ - vca_deployed.get("vdu_count_index", 0) == vdu_index and \ - vca_deployed.get("ee_descriptor_id") == ee_descriptor_id: + if ( + vca_deployed.get("member-vnf-index") == member_vnf_index + and vca_deployed.get("vdu_id") == vdu_id + and vca_deployed.get("kdu_name") == kdu_name + and vca_deployed.get("vdu_count_index", 0) == vdu_index + and vca_deployed.get("ee_descriptor_id") == ee_descriptor_id + ): break else: # not found, create one. - target = "ns" if not member_vnf_index else "vnf/{}".format(member_vnf_index) + target = ( + "ns" if not member_vnf_index else "vnf/{}".format(member_vnf_index) + ) if vdu_id: target += "/vdu/{}/{}".format(vdu_id, vdu_index or 0) elif kdu_name: @@ -2535,18 +4017,19 @@ class NsLcm(LcmBase): "vdu_count_index": vdu_index, "operational-status": "init", # TODO revise "detailed-status": "", # TODO revise - "step": "initial-deploy", # TODO revise + "step": "initial-deploy", # TODO revise "vnfd_id": vnfd_id, "vdu_name": vdu_name, "type": vca_type, - "ee_descriptor_id": ee_descriptor_id + "ee_descriptor_id": ee_descriptor_id, + "charm_name": charm_name, } vca_index += 1 # create VCA and configurationStatus in db db_dict = { "_admin.deployed.VCA.{}".format(vca_index): vca_deployed, - "configurationStatus.{}".format(vca_index): dict() + "configurationStatus.{}".format(vca_index): dict(), } self.update_db_2("nsrs", nsr_id, db_dict) @@ -2567,6 +4050,7 @@ class NsLcm(LcmBase): vdu_id=vdu_id, kdu_name=kdu_name, vdu_index=vdu_index, + kdu_index=kdu_index, deploy_params=deploy_params, config_descriptor=descriptor_config, base_folder=base_folder, @@ -2574,46 +4058,21 @@ class NsLcm(LcmBase): stage=stage, vca_type=vca_type, vca_name=vca_name, - ee_config_descriptor=ee_item + ee_config_descriptor=ee_item, ) ) - self.lcm_tasks.register("ns", nsr_id, nslcmop_id, "instantiate_N2VC-{}".format(vca_index), task_n2vc) - task_instantiation_info[task_n2vc] = self.task_name_deploy_vca + " {}.{}".format( - member_vnf_index or "", vdu_id or "") - - @staticmethod - def _create_nslcmop(nsr_id, operation, params): - """ - Creates a ns-lcm-opp content to be stored at database. - :param nsr_id: internal id of the instance - :param operation: instantiate, terminate, scale, action, ... - :param params: user parameters for the operation - :return: dictionary following SOL005 format - """ - # Raise exception if invalid arguments - if not (nsr_id and operation and params): - raise LcmException( - "Parameters 'nsr_id', 'operation' and 'params' needed to create primitive not provided") - now = time() - _id = str(uuid4()) - nslcmop = { - "id": _id, - "_id": _id, - # COMPLETED,PARTIALLY_COMPLETED,FAILED_TEMP,FAILED,ROLLING_BACK,ROLLED_BACK - "operationState": "PROCESSING", - "statusEnteredTime": now, - "nsInstanceId": nsr_id, - "lcmOperationType": operation, - "startTime": now, - "isAutomaticInvocation": False, - "operationParams": params, - "isCancelPending": False, - "links": { - "self": "/osm/nslcm/v1/ns_lcm_op_occs/" + _id, - "nsInstance": "/osm/nslcm/v1/ns_instances/" + nsr_id, - } - } - return nslcmop + self.lcm_tasks.register( + "ns", + nsr_id, + nslcmop_id, + "instantiate_N2VC-{}".format(vca_index), + task_n2vc, + ) + task_instantiation_info[ + task_n2vc + ] = self.task_name_deploy_vca + " {}.{}".format( + member_vnf_index or "", vdu_id or "" + ) def _format_additional_params(self, params): params = params or {} @@ -2623,7 +4082,7 @@ class NsLcm(LcmBase): return params def _get_terminate_primitive_params(self, seq, vnf_index): - primitive = seq.get('name') + primitive = seq.get("name") primitive_params = {} params = { "member_vnf_index": vnf_index, @@ -2636,8 +4095,8 @@ class NsLcm(LcmBase): # sub-operations def _retry_or_skip_suboperation(self, db_nslcmop, op_index): - op = deep_get(db_nslcmop, ('_admin', 'operations'), [])[op_index] - if op.get('operationState') == 'COMPLETED': + op = deep_get(db_nslcmop, ("_admin", "operations"), [])[op_index] + if op.get("operationState") == "COMPLETED": # b. Skip sub-operation # _ns_execute_primitive() or RO.create_action() will NOT be executed return self.SUBOPERATION_STATUS_SKIP @@ -2645,10 +4104,11 @@ class NsLcm(LcmBase): # c. retry executing sub-operation # The sub-operation exists, and operationState != 'COMPLETED' # Update operationState = 'PROCESSING' to indicate a retry. - operationState = 'PROCESSING' - detailed_status = 'In progress' + operationState = "PROCESSING" + detailed_status = "In progress" self._update_suboperation_status( - db_nslcmop, op_index, operationState, detailed_status) + db_nslcmop, op_index, operationState, detailed_status + ) # Return the sub-operation index # _ns_execute_primitive() or RO.create_action() will be called from scale() # with arguments extracted from the sub-operation @@ -2658,22 +4118,25 @@ class NsLcm(LcmBase): # Returns the index of the matching sub-operation, or SUBOPERATION_STATUS_NOT_FOUND if no match def _find_suboperation(self, db_nslcmop, match): if db_nslcmop and match: - op_list = db_nslcmop.get('_admin', {}).get('operations', []) + op_list = db_nslcmop.get("_admin", {}).get("operations", []) for i, op in enumerate(op_list): if all(op.get(k) == match[k] for k in match): return i return self.SUBOPERATION_STATUS_NOT_FOUND # Update status for a sub-operation given its index - def _update_suboperation_status(self, db_nslcmop, op_index, operationState, detailed_status): + def _update_suboperation_status( + self, db_nslcmop, op_index, operationState, detailed_status + ): # Update DB for HA tasks - q_filter = {'_id': db_nslcmop['_id']} - update_dict = {'_admin.operations.{}.operationState'.format(op_index): operationState, - '_admin.operations.{}.detailed-status'.format(op_index): detailed_status} - self.db.set_one("nslcmops", - q_filter=q_filter, - update_dict=update_dict, - fail_on_empty=False) + q_filter = {"_id": db_nslcmop["_id"]} + update_dict = { + "_admin.operations.{}.operationState".format(op_index): operationState, + "_admin.operations.{}.detailed-status".format(op_index): detailed_status, + } + self.db.set_one( + "nslcmops", q_filter=q_filter, update_dict=update_dict, fail_on_empty=False + ) # Add sub-operation, return the index of the added sub-operation # Optionally, set operationState, detailed-status, and operationType @@ -2682,40 +4145,54 @@ class NsLcm(LcmBase): # 'detailed-status' : status message # 'operationType': may be any type, in the case of scaling: 'PRE-SCALE' | 'POST-SCALE' # Status and operation type are currently only used for 'scale', but NOT for 'terminate' sub-operations. - def _add_suboperation(self, db_nslcmop, vnf_index, vdu_id, vdu_count_index, vdu_name, primitive, - mapped_primitive_params, operationState=None, detailed_status=None, operationType=None, - RO_nsr_id=None, RO_scaling_info=None): + def _add_suboperation( + self, + db_nslcmop, + vnf_index, + vdu_id, + vdu_count_index, + vdu_name, + primitive, + mapped_primitive_params, + operationState=None, + detailed_status=None, + operationType=None, + RO_nsr_id=None, + RO_scaling_info=None, + ): if not db_nslcmop: return self.SUBOPERATION_STATUS_NOT_FOUND # Get the "_admin.operations" list, if it exists - db_nslcmop_admin = db_nslcmop.get('_admin', {}) - op_list = db_nslcmop_admin.get('operations') + db_nslcmop_admin = db_nslcmop.get("_admin", {}) + op_list = db_nslcmop_admin.get("operations") # Create or append to the "_admin.operations" list - new_op = {'member_vnf_index': vnf_index, - 'vdu_id': vdu_id, - 'vdu_count_index': vdu_count_index, - 'primitive': primitive, - 'primitive_params': mapped_primitive_params} + new_op = { + "member_vnf_index": vnf_index, + "vdu_id": vdu_id, + "vdu_count_index": vdu_count_index, + "primitive": primitive, + "primitive_params": mapped_primitive_params, + } if operationState: - new_op['operationState'] = operationState + new_op["operationState"] = operationState if detailed_status: - new_op['detailed-status'] = detailed_status + new_op["detailed-status"] = detailed_status if operationType: - new_op['lcmOperationType'] = operationType + new_op["lcmOperationType"] = operationType if RO_nsr_id: - new_op['RO_nsr_id'] = RO_nsr_id + new_op["RO_nsr_id"] = RO_nsr_id if RO_scaling_info: - new_op['RO_scaling_info'] = RO_scaling_info + new_op["RO_scaling_info"] = RO_scaling_info if not op_list: # No existing operations, create key 'operations' with current operation as first list element - db_nslcmop_admin.update({'operations': [new_op]}) - op_list = db_nslcmop_admin.get('operations') + db_nslcmop_admin.update({"operations": [new_op]}) + op_list = db_nslcmop_admin.get("operations") else: # Existing operations, append operation to list op_list.append(new_op) - db_nslcmop_update = {'_admin.operations': op_list} - self.update_db_2("nslcmops", db_nslcmop['_id'], db_nslcmop_update) + db_nslcmop_update = {"_admin.operations": op_list} + self.update_db_2("nslcmops", db_nslcmop["_id"], db_nslcmop_update) op_index = len(op_list) - 1 return op_index @@ -2726,22 +4203,30 @@ class NsLcm(LcmBase): # a. New: First time execution, return SUBOPERATION_STATUS_NEW # b. Skip: Existing sub-operation exists, operationState == 'COMPLETED', return SUBOPERATION_STATUS_SKIP # c. retry: Existing sub-operation exists, operationState != 'COMPLETED', return op_index to re-execute - def _check_or_add_scale_suboperation(self, db_nslcmop, vnf_index, vnf_config_primitive, primitive_params, - operationType, RO_nsr_id=None, RO_scaling_info=None): + def _check_or_add_scale_suboperation( + self, + db_nslcmop, + vnf_index, + vnf_config_primitive, + primitive_params, + operationType, + RO_nsr_id=None, + RO_scaling_info=None, + ): # Find this sub-operation if RO_nsr_id and RO_scaling_info: - operationType = 'SCALE-RO' + operationType = "SCALE-RO" match = { - 'member_vnf_index': vnf_index, - 'RO_nsr_id': RO_nsr_id, - 'RO_scaling_info': RO_scaling_info, + "member_vnf_index": vnf_index, + "RO_nsr_id": RO_nsr_id, + "RO_scaling_info": RO_scaling_info, } else: match = { - 'member_vnf_index': vnf_index, - 'primitive': vnf_config_primitive, - 'primitive_params': primitive_params, - 'lcmOperationType': operationType + "member_vnf_index": vnf_index, + "primitive": vnf_config_primitive, + "primitive_params": primitive_params, + "lcmOperationType": operationType, } op_index = self._find_suboperation(db_nslcmop, match) if op_index == self.SUBOPERATION_STATUS_NOT_FOUND: @@ -2759,21 +4244,23 @@ class NsLcm(LcmBase): RO_nsr_id = None RO_scaling_info = None # Initial status for sub-operation - operationState = 'PROCESSING' - detailed_status = 'In progress' + operationState = "PROCESSING" + detailed_status = "In progress" # Add sub-operation for pre/post-scaling (zero or more operations) - self._add_suboperation(db_nslcmop, - vnf_index, - vdu_id, - vdu_count_index, - vdu_name, - vnf_config_primitive, - primitive_params, - operationState, - detailed_status, - operationType, - RO_nsr_id, - RO_scaling_info) + self._add_suboperation( + db_nslcmop, + vnf_index, + vdu_id, + vdu_count_index, + vdu_name, + vnf_config_primitive, + primitive_params, + operationState, + detailed_status, + operationType, + RO_nsr_id, + RO_scaling_info, + ) return self.SUBOPERATION_STATUS_NEW else: # Return either SUBOPERATION_STATUS_SKIP (operationState == 'COMPLETED'), @@ -2782,14 +4269,18 @@ class NsLcm(LcmBase): # Function to return execution_environment id - def _get_ee_id(self, vnf_index, vdu_id, vca_deployed_list): - # TODO vdu_index_count - for vca in vca_deployed_list: - if vca["member-vnf-index"] == vnf_index and vca["vdu_id"] == vdu_id: - return vca["ee_id"] - - async def destroy_N2VC(self, logging_text, db_nslcmop, vca_deployed, config_descriptor, - vca_index, destroy_ee=True, exec_primitives=True, scaling_in=False): + async def destroy_N2VC( + self, + logging_text, + db_nslcmop, + vca_deployed, + config_descriptor, + vca_index, + destroy_ee=True, + exec_primitives=True, + scaling_in=False, + vca_id: str = None, + ): """ Execute the terminate primitives and destroy the execution environment (if destroy_ee=False :param logging_text: @@ -2805,7 +4296,8 @@ class NsLcm(LcmBase): """ self.logger.debug( - logging_text + " vca_index: {}, vca_deployed: {}, config_descriptor: {}, destroy_ee: {}".format( + logging_text + + " vca_index: {}, vca_deployed: {}, config_descriptor: {}, destroy_ee: {}".format( vca_index, vca_deployed, config_descriptor, destroy_ee ) ) @@ -2815,7 +4307,9 @@ class NsLcm(LcmBase): # execute terminate_primitives if exec_primitives: terminate_primitives = get_ee_sorted_terminate_config_primitive_list( - config_descriptor.get("terminate-config-primitive"), vca_deployed.get("ee_descriptor_id")) + config_descriptor.get("terminate-config-primitive"), + vca_deployed.get("ee_descriptor_id"), + ) vdu_id = vca_deployed.get("vdu_id") vdu_count_index = vca_deployed.get("vdu_count_index") vdu_name = vca_deployed.get("vdu_name") @@ -2824,217 +4318,103 @@ class NsLcm(LcmBase): for seq in terminate_primitives: # For each sequence in list, get primitive and call _ns_execute_primitive() step = "Calling terminate action for vnf_member_index={} primitive={}".format( - vnf_index, seq.get("name")) + vnf_index, seq.get("name") + ) self.logger.debug(logging_text + step) # Create the primitive for each sequence, i.e. "primitive": "touch" - primitive = seq.get('name') - mapped_primitive_params = self._get_terminate_primitive_params(seq, vnf_index) + primitive = seq.get("name") + mapped_primitive_params = self._get_terminate_primitive_params( + seq, vnf_index + ) # Add sub-operation - self._add_suboperation(db_nslcmop, - vnf_index, - vdu_id, - vdu_count_index, - vdu_name, - primitive, - mapped_primitive_params) + self._add_suboperation( + db_nslcmop, + vnf_index, + vdu_id, + vdu_count_index, + vdu_name, + primitive, + mapped_primitive_params, + ) # Sub-operations: Call _ns_execute_primitive() instead of action() try: - result, result_detail = await self._ns_execute_primitive(vca_deployed["ee_id"], primitive, - mapped_primitive_params, - vca_type=vca_type) + result, result_detail = await self._ns_execute_primitive( + vca_deployed["ee_id"], + primitive, + mapped_primitive_params, + vca_type=vca_type, + vca_id=vca_id, + ) except LcmException: # this happens when VCA is not deployed. In this case it is not needed to terminate continue - result_ok = ['COMPLETED', 'PARTIALLY_COMPLETED'] + result_ok = ["COMPLETED", "PARTIALLY_COMPLETED"] if result not in result_ok: - raise LcmException("terminate_primitive {} for vnf_member_index={} fails with " - "error {}".format(seq.get("name"), vnf_index, result_detail)) + raise LcmException( + "terminate_primitive {} for vnf_member_index={} fails with " + "error {}".format(seq.get("name"), vnf_index, result_detail) + ) # set that this VCA do not need terminated - db_update_entry = "_admin.deployed.VCA.{}.needed_terminate".format(vca_index) - self.update_db_2("nsrs", db_nslcmop["nsInstanceId"], {db_update_entry: False}) + db_update_entry = "_admin.deployed.VCA.{}.needed_terminate".format( + vca_index + ) + self.update_db_2( + "nsrs", db_nslcmop["nsInstanceId"], {db_update_entry: False} + ) - if vca_deployed.get("prometheus_jobs") and self.prometheus: - await self.prometheus.update(remove_jobs=vca_deployed["prometheus_jobs"]) + # Delete Prometheus Jobs if any + # This uses NSR_ID, so it will destroy any jobs under this index + self.db.del_list("prometheus_jobs", {"nsr_id": db_nslcmop["nsInstanceId"]}) if destroy_ee: - await self.vca_map[vca_type].delete_execution_environment(vca_deployed["ee_id"], scaling_in=scaling_in) + await self.vca_map[vca_type].delete_execution_environment( + vca_deployed["ee_id"], + scaling_in=scaling_in, + vca_type=vca_type, + vca_id=vca_id, + ) - async def _delete_all_N2VC(self, db_nsr: dict): - self._write_all_config_status(db_nsr=db_nsr, status='TERMINATING') + async def _delete_all_N2VC(self, db_nsr: dict, vca_id: str = None): + self._write_all_config_status(db_nsr=db_nsr, status="TERMINATING") namespace = "." + db_nsr["_id"] try: - await self.n2vc.delete_namespace(namespace=namespace, total_timeout=self.timeout_charm_delete) + await self.n2vc.delete_namespace( + namespace=namespace, + total_timeout=self.timeout.charm_delete, + vca_id=vca_id, + ) except N2VCNotFound: # already deleted. Skip pass - self._write_all_config_status(db_nsr=db_nsr, status='DELETED') - - async def _terminate_RO(self, logging_text, nsr_deployed, nsr_id, nslcmop_id, stage): - """ - Terminates a deployment from RO - :param logging_text: - :param nsr_deployed: db_nsr._admin.deployed - :param nsr_id: - :param nslcmop_id: - :param stage: list of string with the content to write on db_nslcmop.detailed-status. - this method will update only the index 2, but it will write on database the concatenated content of the list - :return: - """ - db_nsr_update = {} - failed_detail = [] - ro_nsr_id = ro_delete_action = None - if nsr_deployed and nsr_deployed.get("RO"): - ro_nsr_id = nsr_deployed["RO"].get("nsr_id") - ro_delete_action = nsr_deployed["RO"].get("nsr_delete_action_id") - try: - if ro_nsr_id: - stage[2] = "Deleting ns from VIM." - db_nsr_update["detailed-status"] = " ".join(stage) - self._write_op_status(nslcmop_id, stage) - self.logger.debug(logging_text + stage[2]) - self.update_db_2("nsrs", nsr_id, db_nsr_update) - self._write_op_status(nslcmop_id, stage) - desc = await self.RO.delete("ns", ro_nsr_id) - ro_delete_action = desc["action_id"] - db_nsr_update["_admin.deployed.RO.nsr_delete_action_id"] = ro_delete_action - db_nsr_update["_admin.deployed.RO.nsr_id"] = None - db_nsr_update["_admin.deployed.RO.nsr_status"] = "DELETED" - if ro_delete_action: - # wait until NS is deleted from VIM - stage[2] = "Waiting ns deleted from VIM." - detailed_status_old = None - self.logger.debug(logging_text + stage[2] + " RO_id={} ro_delete_action={}".format(ro_nsr_id, - ro_delete_action)) - self.update_db_2("nsrs", nsr_id, db_nsr_update) - self._write_op_status(nslcmop_id, stage) - - delete_timeout = 20 * 60 # 20 minutes - while delete_timeout > 0: - desc = await self.RO.show( - "ns", - item_id_name=ro_nsr_id, - extra_item="action", - extra_item_id=ro_delete_action) - - # deploymentStatus - self._on_update_ro_db(nsrs_id=nsr_id, ro_descriptor=desc) - - ns_status, ns_status_info = self.RO.check_action_status(desc) - if ns_status == "ERROR": - raise ROclient.ROClientException(ns_status_info) - elif ns_status == "BUILD": - stage[2] = "Deleting from VIM {}".format(ns_status_info) - elif ns_status == "ACTIVE": - db_nsr_update["_admin.deployed.RO.nsr_delete_action_id"] = None - db_nsr_update["_admin.deployed.RO.nsr_status"] = "DELETED" - break - else: - assert False, "ROclient.check_action_status returns unknown {}".format(ns_status) - if stage[2] != detailed_status_old: - detailed_status_old = stage[2] - db_nsr_update["detailed-status"] = " ".join(stage) - self._write_op_status(nslcmop_id, stage) - self.update_db_2("nsrs", nsr_id, db_nsr_update) - await asyncio.sleep(5, loop=self.loop) - delete_timeout -= 5 - else: # delete_timeout <= 0: - raise ROclient.ROClientException("Timeout waiting ns deleted from VIM") - - except Exception as e: - self.update_db_2("nsrs", nsr_id, db_nsr_update) - if isinstance(e, ROclient.ROClientException) and e.http_code == 404: # not found - db_nsr_update["_admin.deployed.RO.nsr_id"] = None - db_nsr_update["_admin.deployed.RO.nsr_status"] = "DELETED" - db_nsr_update["_admin.deployed.RO.nsr_delete_action_id"] = None - self.logger.debug(logging_text + "RO_ns_id={} already deleted".format(ro_nsr_id)) - elif isinstance(e, ROclient.ROClientException) and e.http_code == 409: # conflict - failed_detail.append("delete conflict: {}".format(e)) - self.logger.debug(logging_text + "RO_ns_id={} delete conflict: {}".format(ro_nsr_id, e)) - else: - failed_detail.append("delete error: {}".format(e)) - self.logger.error(logging_text + "RO_ns_id={} delete error: {}".format(ro_nsr_id, e)) - - # Delete nsd - if not failed_detail and deep_get(nsr_deployed, ("RO", "nsd_id")): - ro_nsd_id = nsr_deployed["RO"]["nsd_id"] - try: - stage[2] = "Deleting nsd from RO." - db_nsr_update["detailed-status"] = " ".join(stage) - self.update_db_2("nsrs", nsr_id, db_nsr_update) - self._write_op_status(nslcmop_id, stage) - await self.RO.delete("nsd", ro_nsd_id) - self.logger.debug(logging_text + "ro_nsd_id={} deleted".format(ro_nsd_id)) - db_nsr_update["_admin.deployed.RO.nsd_id"] = None - except Exception as e: - if isinstance(e, ROclient.ROClientException) and e.http_code == 404: # not found - db_nsr_update["_admin.deployed.RO.nsd_id"] = None - self.logger.debug(logging_text + "ro_nsd_id={} already deleted".format(ro_nsd_id)) - elif isinstance(e, ROclient.ROClientException) and e.http_code == 409: # conflict - failed_detail.append("ro_nsd_id={} delete conflict: {}".format(ro_nsd_id, e)) - self.logger.debug(logging_text + failed_detail[-1]) - else: - failed_detail.append("ro_nsd_id={} delete error: {}".format(ro_nsd_id, e)) - self.logger.error(logging_text + failed_detail[-1]) - - if not failed_detail and deep_get(nsr_deployed, ("RO", "vnfd")): - for index, vnf_deployed in enumerate(nsr_deployed["RO"]["vnfd"]): - if not vnf_deployed or not vnf_deployed["id"]: - continue - try: - ro_vnfd_id = vnf_deployed["id"] - stage[2] = "Deleting member_vnf_index={} ro_vnfd_id={} from RO.".format( - vnf_deployed["member-vnf-index"], ro_vnfd_id) - db_nsr_update["detailed-status"] = " ".join(stage) - self.update_db_2("nsrs", nsr_id, db_nsr_update) - self._write_op_status(nslcmop_id, stage) - await self.RO.delete("vnfd", ro_vnfd_id) - self.logger.debug(logging_text + "ro_vnfd_id={} deleted".format(ro_vnfd_id)) - db_nsr_update["_admin.deployed.RO.vnfd.{}.id".format(index)] = None - except Exception as e: - if isinstance(e, ROclient.ROClientException) and e.http_code == 404: # not found - db_nsr_update["_admin.deployed.RO.vnfd.{}.id".format(index)] = None - self.logger.debug(logging_text + "ro_vnfd_id={} already deleted ".format(ro_vnfd_id)) - elif isinstance(e, ROclient.ROClientException) and e.http_code == 409: # conflict - failed_detail.append("ro_vnfd_id={} delete conflict: {}".format(ro_vnfd_id, e)) - self.logger.debug(logging_text + failed_detail[-1]) - else: - failed_detail.append("ro_vnfd_id={} delete error: {}".format(ro_vnfd_id, e)) - self.logger.error(logging_text + failed_detail[-1]) - - if failed_detail: - stage[2] = "Error deleting from VIM" - else: - stage[2] = "Deleted from VIM" - db_nsr_update["detailed-status"] = " ".join(stage) - self.update_db_2("nsrs", nsr_id, db_nsr_update) - self._write_op_status(nslcmop_id, stage) - - if failed_detail: - raise LcmException("; ".join(failed_detail)) + self._write_all_config_status(db_nsr=db_nsr, status="DELETED") async def terminate(self, nsr_id, nslcmop_id): # Try to lock HA task here - task_is_locked_by_me = self.lcm_tasks.lock_HA('ns', 'nslcmops', nslcmop_id) + task_is_locked_by_me = self.lcm_tasks.lock_HA("ns", "nslcmops", nslcmop_id) if not task_is_locked_by_me: return logging_text = "Task ns={} terminate={} ".format(nsr_id, nslcmop_id) self.logger.debug(logging_text + "Enter") - timeout_ns_terminate = self.timeout_ns_terminate + timeout_ns_terminate = self.timeout.ns_terminate db_nsr = None db_nslcmop = None operation_params = None exc = None - error_list = [] # annotates all failed error messages + error_list = [] # annotates all failed error messages db_nslcmop_update = {} autoremove = False # autoremove after terminated tasks_dict_info = {} db_nsr_update = {} - stage = ["Stage 1/3: Preparing task.", "Waiting for previous operations to terminate.", ""] + stage = [ + "Stage 1/3: Preparing task.", + "Waiting for previous operations to terminate.", + "", + ] # ^ contains [stage, step, VIM-status] try: # wait for any previous tasks in process - await self.lcm_tasks.waitfor_related_HA("ns", 'nslcmops', nslcmop_id) + await self.lcm_tasks.waitfor_related_HA("ns", "nslcmops", nslcmop_id) stage[1] = "Getting nslcmop={} from db.".format(nslcmop_id) db_nslcmop = self.db.get_one("nslcmops", {"_id": nslcmop_id}) @@ -3051,19 +4431,18 @@ class NsLcm(LcmBase): ns_state="TERMINATING", current_operation="TERMINATING", current_operation_id=nslcmop_id, - other_update=db_nsr_update - ) - self._write_op_status( - op_id=nslcmop_id, - queuePosition=0, - stage=stage + other_update=db_nsr_update, ) + self._write_op_status(op_id=nslcmop_id, queuePosition=0, stage=stage) nsr_deployed = deepcopy(db_nsr["_admin"].get("deployed")) or {} if db_nsr["_admin"]["nsState"] == "NOT_INSTANTIATED": return stage[1] = "Getting vnf descriptors from db." db_vnfrs_list = self.db.get_list("vnfrs", {"nsr-id-ref": nsr_id}) + db_vnfrs_dict = { + db_vnfr["member-vnf-index-ref"]: db_vnfr for db_vnfr in db_vnfrs_list + } db_vnfds_from_id = {} db_vnfds_from_member_index = {} # Loop over VNFRs @@ -3072,7 +4451,9 @@ class NsLcm(LcmBase): if vnfd_id not in db_vnfds_from_id: vnfd = self.db.get_one("vnfds", {"_id": vnfd_id}) db_vnfds_from_id[vnfd_id] = vnfd - db_vnfds_from_member_index[vnfr["member-vnf-index-ref"]] = db_vnfds_from_id[vnfd_id] + db_vnfds_from_member_index[ + vnfr["member-vnf-index-ref"] + ] = db_vnfds_from_id[vnfd_id] # Destroy individual execution environments when there are terminating primitives. # Rest of EE will be deleted at once @@ -3086,6 +4467,13 @@ class NsLcm(LcmBase): for vca_index, vca in enumerate(get_iterable(nsr_deployed, "VCA")): config_descriptor = None + vca_member_vnf_index = vca.get("member-vnf-index") + vca_id = self.get_vca_id( + db_vnfrs_dict.get(vca_member_vnf_index) + if vca_member_vnf_index + else None, + db_nsr, + ) if not vca or not vca.get("ee_id"): continue if not vca.get("member-vnf-index"): @@ -3101,27 +4489,46 @@ class NsLcm(LcmBase): db_vnfd = db_vnfds_from_member_index[vca["member-vnf-index"]] config_descriptor = get_configuration(db_vnfd, db_vnfd["id"]) vca_type = vca.get("type") - exec_terminate_primitives = (not operation_params.get("skip_terminate_primitives") and - vca.get("needed_terminate")) + exec_terminate_primitives = not operation_params.get( + "skip_terminate_primitives" + ) and vca.get("needed_terminate") # For helm we must destroy_ee. Also for native_charm, as juju_model cannot be deleted if there are # pending native charms - destroy_ee = True if vca_type in ("helm", "helm-v3", "native_charm") else False + destroy_ee = ( + True if vca_type in ("helm", "helm-v3", "native_charm") else False + ) # self.logger.debug(logging_text + "vca_index: {}, ee_id: {}, vca_type: {} destroy_ee: {}".format( # vca_index, vca.get("ee_id"), vca_type, destroy_ee)) task = asyncio.ensure_future( - self.destroy_N2VC(logging_text, db_nslcmop, vca, config_descriptor, vca_index, - destroy_ee, exec_terminate_primitives)) + self.destroy_N2VC( + logging_text, + db_nslcmop, + vca, + config_descriptor, + vca_index, + destroy_ee, + exec_terminate_primitives, + vca_id=vca_id, + ) + ) tasks_dict_info[task] = "Terminating VCA {}".format(vca.get("ee_id")) # wait for pending tasks of terminate primitives if tasks_dict_info: - self.logger.debug(logging_text + 'Waiting for tasks {}'.format(list(tasks_dict_info.keys()))) - error_list = await self._wait_for_tasks(logging_text, tasks_dict_info, - min(self.timeout_charm_delete, timeout_ns_terminate), - stage, nslcmop_id) + self.logger.debug( + logging_text + + "Waiting for tasks {}".format(list(tasks_dict_info.keys())) + ) + error_list = await self._wait_for_tasks( + logging_text, + tasks_dict_info, + min(self.timeout.charm_delete, timeout_ns_terminate), + stage, + nslcmop_id, + ) tasks_dict_info.clear() if error_list: - return # raise LcmException("; ".join(error_list)) + return # raise LcmException("; ".join(error_list)) # remove All execution environments at once stage[0] = "Stage 3/3 delete all." @@ -3129,11 +4536,26 @@ class NsLcm(LcmBase): if nsr_deployed.get("VCA"): stage[1] = "Deleting all execution environments." self.logger.debug(logging_text + stage[1]) - task_delete_ee = asyncio.ensure_future(asyncio.wait_for(self._delete_all_N2VC(db_nsr=db_nsr), - timeout=self.timeout_charm_delete)) + vca_id = self.get_vca_id({}, db_nsr) + task_delete_ee = asyncio.ensure_future( + asyncio.wait_for( + self._delete_all_N2VC(db_nsr=db_nsr, vca_id=vca_id), + timeout=self.timeout.charm_delete, + ) + ) # task_delete_ee = asyncio.ensure_future(self.n2vc.delete_namespace(namespace="." + nsr_id)) tasks_dict_info[task_delete_ee] = "Terminating all VCA" + # Delete Namespace and Certificates if necessary + if check_helm_ee_in_ns(list(db_vnfds_from_member_index.values())): + await self.vca_map["helm-v3"].delete_tls_certificate( + namespace=db_nslcmop["nsInstanceId"], + certificate_name=self.EE_TLS_NAME, + ) + await self.vca_map["helm-v3"].delete_namespace( + namespace=db_nslcmop["nsInstanceId"], + ) + # Delete from k8scluster stage[1] = "Deleting KDUs." self.logger.debug(logging_text + stage[1]) @@ -3143,37 +4565,59 @@ class NsLcm(LcmBase): continue kdu_instance = kdu.get("kdu-instance") if kdu.get("k8scluster-type") in self.k8scluster_map: + # TODO: Uninstall kdu instances taking into account they could be deployed in different VIMs + vca_id = self.get_vca_id({}, db_nsr) task_delete_kdu_instance = asyncio.ensure_future( self.k8scluster_map[kdu["k8scluster-type"]].uninstall( cluster_uuid=kdu.get("k8scluster-uuid"), - kdu_instance=kdu_instance)) + kdu_instance=kdu_instance, + vca_id=vca_id, + namespace=kdu.get("namespace"), + ) + ) else: - self.logger.error(logging_text + "Unknown k8s deployment type {}". - format(kdu.get("k8scluster-type"))) + self.logger.error( + logging_text + + "Unknown k8s deployment type {}".format( + kdu.get("k8scluster-type") + ) + ) continue - tasks_dict_info[task_delete_kdu_instance] = "Terminating KDU '{}'".format(kdu.get("kdu-name")) + tasks_dict_info[ + task_delete_kdu_instance + ] = "Terminating KDU '{}'".format(kdu.get("kdu-name")) # remove from RO stage[1] = "Deleting ns from VIM." - if self.ng_ro: + if self.ro_config.ng: task_delete_ro = asyncio.ensure_future( - self._terminate_ng_ro(logging_text, nsr_deployed, nsr_id, nslcmop_id, stage)) - else: - task_delete_ro = asyncio.ensure_future( - self._terminate_RO(logging_text, nsr_deployed, nsr_id, nslcmop_id, stage)) - tasks_dict_info[task_delete_ro] = "Removing deployment from VIM" + self._terminate_ng_ro( + logging_text, nsr_deployed, nsr_id, nslcmop_id, stage + ) + ) + tasks_dict_info[task_delete_ro] = "Removing deployment from VIM" # rest of staff will be done at finally - except (ROclient.ROClientException, DbException, LcmException, N2VCException) as e: + except ( + ROclient.ROClientException, + DbException, + LcmException, + N2VCException, + ) as e: self.logger.error(logging_text + "Exit Exception {}".format(e)) exc = e except asyncio.CancelledError: - self.logger.error(logging_text + "Cancelled Exception while '{}'".format(stage[1])) + self.logger.error( + logging_text + "Cancelled Exception while '{}'".format(stage[1]) + ) exc = "Operation was cancelled" except Exception as e: exc = traceback.format_exc() - self.logger.critical(logging_text + "Exit Exception while '{}': {}".format(stage[1], e), exc_info=True) + self.logger.critical( + logging_text + "Exit Exception while '{}': {}".format(stage[1], e), + exc_info=True, + ) finally: if exc: error_list.append(str(exc)) @@ -3182,8 +4626,13 @@ class NsLcm(LcmBase): if tasks_dict_info: stage[1] = "Waiting for terminate pending tasks." self.logger.debug(logging_text + stage[1]) - error_list += await self._wait_for_tasks(logging_text, tasks_dict_info, timeout_ns_terminate, - stage, nslcmop_id) + error_list += await self._wait_for_tasks( + logging_text, + tasks_dict_info, + timeout_ns_terminate, + stage, + nslcmop_id, + ) stage[1] = stage[2] = "" except asyncio.CancelledError: error_list.append("Cancelled") @@ -3194,11 +4643,17 @@ class NsLcm(LcmBase): if error_list: error_detail = "; ".join(error_list) # self.logger.error(logging_text + error_detail) - error_description_nslcmop = '{} Detail: {}'.format(stage[0], error_detail) - error_description_nsr = 'Operation: TERMINATING.{}, {}.'.format(nslcmop_id, stage[0]) + error_description_nslcmop = "{} Detail: {}".format( + stage[0], error_detail + ) + error_description_nsr = "Operation: TERMINATING.{}, {}.".format( + nslcmop_id, stage[0] + ) db_nsr_update["operational-status"] = "failed" - db_nsr_update["detailed-status"] = error_description_nsr + " Detail: " + error_detail + db_nsr_update["detailed-status"] = ( + error_description_nsr + " Detail: " + error_detail + ) db_nslcmop_update["detailed-status"] = error_detail nslcmop_operation_state = "FAILED" ns_state = "BROKEN" @@ -3220,7 +4675,7 @@ class NsLcm(LcmBase): current_operation_id=None, error_description=error_description_nsr, error_detail=error_detail, - other_update=db_nsr_update + other_update=db_nsr_update, ) self._write_op_status( op_id=nslcmop_id, @@ -3231,25 +4686,45 @@ class NsLcm(LcmBase): ) if ns_state == "NOT_INSTANTIATED": try: - self.db.set_list("vnfrs", {"nsr-id-ref": nsr_id}, {"_admin.nsState": "NOT_INSTANTIATED"}) + self.db.set_list( + "vnfrs", + {"nsr-id-ref": nsr_id}, + {"_admin.nsState": "NOT_INSTANTIATED"}, + ) except DbException as e: - self.logger.warn(logging_text + 'Error writing VNFR status for nsr-id-ref: {} -> {}'. - format(nsr_id, e)) + self.logger.warn( + logging_text + + "Error writing VNFR status for nsr-id-ref: {} -> {}".format( + nsr_id, e + ) + ) if operation_params: autoremove = operation_params.get("autoremove", False) if nslcmop_operation_state: try: - await self.msg.aiowrite("ns", "terminated", {"nsr_id": nsr_id, "nslcmop_id": nslcmop_id, - "operationState": nslcmop_operation_state, - "autoremove": autoremove}, - loop=self.loop) + await self.msg.aiowrite( + "ns", + "terminated", + { + "nsr_id": nsr_id, + "nslcmop_id": nslcmop_id, + "operationState": nslcmop_operation_state, + "autoremove": autoremove, + }, + ) except Exception as e: - self.logger.error(logging_text + "kafka_write notification Exception {}".format(e)) + self.logger.error( + logging_text + "kafka_write notification Exception {}".format(e) + ) + self.logger.debug(f"Deleting alerts: ns_id={nsr_id}") + self.db.del_list("alerts", {"tags.ns_id": nsr_id}) self.logger.debug(logging_text + "Exit") self.lcm_tasks.remove("ns", nsr_id, nslcmop_id, "ns_terminate") - async def _wait_for_tasks(self, logging_text, created_tasks_info, timeout, stage, nslcmop_id, nsr_id=None): + async def _wait_for_tasks( + self, logging_text, created_tasks_info, timeout, stage, nslcmop_id, nsr_id=None + ): time_start = time() error_detail_list = [] error_list = [] @@ -3261,10 +4736,11 @@ class NsLcm(LcmBase): while pending_tasks: new_error = None _timeout = timeout + time_start - time() - done, pending_tasks = await asyncio.wait(pending_tasks, timeout=_timeout, - return_when=asyncio.FIRST_COMPLETED) + done, pending_tasks = await asyncio.wait( + pending_tasks, timeout=_timeout, return_when=asyncio.FIRST_COMPLETED + ) num_done += len(done) - if not done: # Timeout + if not done: # Timeout for task in pending_tasks: new_error = created_tasks_info[task] + ": Timeout" error_detail_list.append(new_error) @@ -3281,20 +4757,45 @@ class NsLcm(LcmBase): new_error = created_tasks_info[task] + ": {}".format(exc) error_list.append(created_tasks_info[task]) error_detail_list.append(new_error) - if isinstance(exc, (str, DbException, N2VCException, ROclient.ROClientException, LcmException, - K8sException, NgRoException)): + if isinstance( + exc, + ( + str, + DbException, + N2VCException, + ROclient.ROClientException, + LcmException, + K8sException, + NgRoException, + ), + ): self.logger.error(logging_text + new_error) else: - exc_traceback = "".join(traceback.format_exception(None, exc, exc.__traceback__)) - self.logger.error(logging_text + created_tasks_info[task] + " " + exc_traceback) + exc_traceback = "".join( + traceback.format_exception(None, exc, exc.__traceback__) + ) + self.logger.error( + logging_text + + created_tasks_info[task] + + " " + + exc_traceback + ) else: - self.logger.debug(logging_text + created_tasks_info[task] + ": Done") + self.logger.debug( + logging_text + created_tasks_info[task] + ": Done" + ) stage[1] = "{}/{}.".format(num_done, num_tasks) if new_error: stage[1] += " Errors: " + ". ".join(error_detail_list) + "." if nsr_id: # update also nsr - self.update_db_2("nsrs", nsr_id, {"errorDescription": "Error at: " + ", ".join(error_list), - "errorDetail": ". ".join(error_detail_list)}) + self.update_db_2( + "nsrs", + nsr_id, + { + "errorDescription": "Error at: " + ", ".join(error_list), + "errorDetail": ". ".join(error_detail_list), + }, + ) self._write_op_status(nslcmop_id, stage) return error_detail_list @@ -3318,46 +4819,77 @@ class NsLcm(LcmBase): calculated_params[param_name] = parameter["value"] else: calculated_params[param_name] = parameter["default-value"] - if isinstance(calculated_params[param_name], str) and calculated_params[param_name].startswith("<") \ - and calculated_params[param_name].endswith(">"): + if ( + isinstance(calculated_params[param_name], str) + and calculated_params[param_name].startswith("<") + and calculated_params[param_name].endswith(">") + ): if calculated_params[param_name][1:-1] in instantiation_params: - calculated_params[param_name] = instantiation_params[calculated_params[param_name][1:-1]] + calculated_params[param_name] = instantiation_params[ + calculated_params[param_name][1:-1] + ] else: - raise LcmException("Parameter {} needed to execute primitive {} not provided". - format(calculated_params[param_name], primitive_desc["name"])) + raise LcmException( + "Parameter {} needed to execute primitive {} not provided".format( + calculated_params[param_name], primitive_desc["name"] + ) + ) else: - raise LcmException("Parameter {} needed to execute primitive {} not provided". - format(param_name, primitive_desc["name"])) + raise LcmException( + "Parameter {} needed to execute primitive {} not provided".format( + param_name, primitive_desc["name"] + ) + ) if isinstance(calculated_params[param_name], (dict, list, tuple)): - calculated_params[param_name] = yaml.safe_dump(calculated_params[param_name], - default_flow_style=True, width=256) - elif isinstance(calculated_params[param_name], str) and calculated_params[param_name].startswith("!!yaml "): + calculated_params[param_name] = yaml.safe_dump( + calculated_params[param_name], default_flow_style=True, width=256 + ) + elif isinstance(calculated_params[param_name], str) and calculated_params[ + param_name + ].startswith("!!yaml "): calculated_params[param_name] = calculated_params[param_name][7:] if parameter.get("data-type") == "INTEGER": try: calculated_params[param_name] = int(calculated_params[param_name]) except ValueError: # error converting string to int raise LcmException( - "Parameter {} of primitive {} must be integer".format(param_name, primitive_desc["name"])) + "Parameter {} of primitive {} must be integer".format( + param_name, primitive_desc["name"] + ) + ) elif parameter.get("data-type") == "BOOLEAN": - calculated_params[param_name] = not ((str(calculated_params[param_name])).lower() == 'false') + calculated_params[param_name] = not ( + (str(calculated_params[param_name])).lower() == "false" + ) # add always ns_config_info if primitive name is config if primitive_desc["name"] == "config": if "ns_config_info" in instantiation_params: - calculated_params["ns_config_info"] = instantiation_params["ns_config_info"] + calculated_params["ns_config_info"] = instantiation_params[ + "ns_config_info" + ] return calculated_params - def _look_for_deployed_vca(self, deployed_vca, member_vnf_index, vdu_id, vdu_count_index, kdu_name=None, - ee_descriptor_id=None): + def _look_for_deployed_vca( + self, + deployed_vca, + member_vnf_index, + vdu_id, + vdu_count_index, + kdu_name=None, + ee_descriptor_id=None, + ): # find vca_deployed record for this action. Raise LcmException if not found or there is not any id. for vca in deployed_vca: if not vca: continue if member_vnf_index != vca["member-vnf-index"] or vdu_id != vca["vdu_id"]: continue - if vdu_count_index is not None and vdu_count_index != vca["vdu_count_index"]: + if ( + vdu_count_index is not None + and vdu_count_index != vca["vdu_count_index"] + ): continue if kdu_name and kdu_name != vca["kdu_name"]: continue @@ -3366,20 +4898,42 @@ class NsLcm(LcmBase): break else: # vca_deployed not found - raise LcmException("charm for member_vnf_index={} vdu_id={}.{} kdu_name={} execution-environment-list.id={}" - " is not deployed".format(member_vnf_index, vdu_id, vdu_count_index, kdu_name, - ee_descriptor_id)) + raise LcmException( + "charm for member_vnf_index={} vdu_id={}.{} kdu_name={} execution-environment-list.id={}" + " is not deployed".format( + member_vnf_index, + vdu_id, + vdu_count_index, + kdu_name, + ee_descriptor_id, + ) + ) # get ee_id ee_id = vca.get("ee_id") - vca_type = vca.get("type", "lxc_proxy_charm") # default value for backward compatibility - proxy charm + vca_type = vca.get( + "type", "lxc_proxy_charm" + ) # default value for backward compatibility - proxy charm if not ee_id: - raise LcmException("charm for member_vnf_index={} vdu_id={} kdu_name={} vdu_count_index={} has not " - "execution environment" - .format(member_vnf_index, vdu_id, kdu_name, vdu_count_index)) + raise LcmException( + "charm for member_vnf_index={} vdu_id={} kdu_name={} vdu_count_index={} has not " + "execution environment".format( + member_vnf_index, vdu_id, kdu_name, vdu_count_index + ) + ) return ee_id, vca_type - async def _ns_execute_primitive(self, ee_id, primitive, primitive_params, retries=0, retries_interval=30, - timeout=None, vca_type=None, db_dict=None) -> (str, str): + async def _ns_execute_primitive( + self, + ee_id, + primitive, + primitive_params, + retries=0, + retries_interval=30, + timeout=None, + vca_type=None, + db_dict=None, + vca_id: str = None, + ) -> (str, str): try: if primitive == "config": primitive_params = {"params": primitive_params} @@ -3393,31 +4947,41 @@ class NsLcm(LcmBase): ee_id=ee_id, primitive_name=primitive, params_dict=primitive_params, - progress_timeout=self.timeout_progress_primitive, - total_timeout=self.timeout_primitive, - db_dict=db_dict), - timeout=timeout or self.timeout_primitive) + progress_timeout=self.timeout.progress_primitive, + total_timeout=self.timeout.primitive, + db_dict=db_dict, + vca_id=vca_id, + vca_type=vca_type, + ), + timeout=timeout or self.timeout.primitive, + ) # execution was OK break except asyncio.CancelledError: raise - except Exception as e: # asyncio.TimeoutError - if isinstance(e, asyncio.TimeoutError): - e = "Timeout" + except Exception as e: retries -= 1 if retries >= 0: - self.logger.debug('Error executing action {} on {} -> {}'.format(primitive, ee_id, e)) + self.logger.debug( + "Error executing action {} on {} -> {}".format( + primitive, ee_id, e + ) + ) # wait and retry - await asyncio.sleep(retries_interval, loop=self.loop) + await asyncio.sleep(retries_interval) else: - return 'FAILED', str(e) + if isinstance(e, asyncio.TimeoutError): + e = N2VCException( + message="Timed out waiting for action to complete" + ) + return "FAILED", getattr(e, "message", repr(e)) - return 'COMPLETED', output + return "COMPLETED", output except (LcmException, asyncio.CancelledError): raise except Exception as e: - return 'FAIL', 'Error executing action {}: {}'.format(primitive, e) + return "FAIL", "Error executing action {}: {}".format(primitive, e) async def vca_status_refresh(self, nsr_id, nslcmop_id): """ @@ -3429,12 +4993,23 @@ class NsLcm(LcmBase): self.logger.debug("Task ns={} action={} Enter".format(nsr_id, nslcmop_id)) db_nsr = self.db.get_one("nsrs", {"_id": nsr_id}) - if db_nsr['_admin']['deployed']['K8s']: - for k8s_index, k8s in enumerate(db_nsr['_admin']['deployed']['K8s']): - cluster_uuid, kdu_instance = k8s["k8scluster-uuid"], k8s["kdu-instance"] - await self._on_update_k8s_db(cluster_uuid, kdu_instance, filter={'_id': nsr_id}) + vca_id = self.get_vca_id({}, db_nsr) + if db_nsr["_admin"]["deployed"]["K8s"]: + for _, k8s in enumerate(db_nsr["_admin"]["deployed"]["K8s"]): + cluster_uuid, kdu_instance, cluster_type = ( + k8s["k8scluster-uuid"], + k8s["kdu-instance"], + k8s["k8scluster-type"], + ) + await self._on_update_k8s_db( + cluster_uuid=cluster_uuid, + kdu_instance=kdu_instance, + filter={"_id": nsr_id}, + vca_id=vca_id, + cluster_type=cluster_type, + ) else: - for vca_index, _ in enumerate(db_nsr['_admin']['deployed']['VCA']): + for vca_index, _ in enumerate(db_nsr["_admin"]["deployed"]["VCA"]): table, filter = "nsrs", {"_id": nsr_id} path = "_admin.deployed.VCA.{}.".format(vca_index) await self._on_update_n2vc_db(table, filter, path, {}) @@ -3444,7 +5019,7 @@ class NsLcm(LcmBase): async def action(self, nsr_id, nslcmop_id): # Try to lock HA task here - task_is_locked_by_me = self.lcm_tasks.lock_HA('ns', 'nslcmops', nslcmop_id) + task_is_locked_by_me = self.lcm_tasks.lock_HA("ns", "nslcmops", nslcmop_id) if not task_is_locked_by_me: return @@ -3458,21 +5033,26 @@ class NsLcm(LcmBase): nslcmop_operation_state = None error_description_nslcmop = None exc = None + step = "" try: # wait for any previous tasks in process step = "Waiting for previous operations to terminate" - await self.lcm_tasks.waitfor_related_HA('ns', 'nslcmops', nslcmop_id) + await self.lcm_tasks.waitfor_related_HA("ns", "nslcmops", nslcmop_id) self._write_ns_status( nsr_id=nsr_id, ns_state=None, current_operation="RUNNING ACTION", - current_operation_id=nslcmop_id + current_operation_id=nslcmop_id, ) step = "Getting information from database" db_nslcmop = self.db.get_one("nslcmops", {"_id": nslcmop_id}) db_nsr = self.db.get_one("nsrs", {"_id": nsr_id}) + if db_nslcmop["operationParams"].get("primitive_params"): + db_nslcmop["operationParams"]["primitive_params"] = json.loads( + db_nslcmop["operationParams"]["primitive_params"] + ) nsr_deployed = db_nsr["_admin"].get("deployed") vnf_index = db_nslcmop["operationParams"].get("member_vnf_index") @@ -3481,17 +5061,34 @@ class NsLcm(LcmBase): vdu_count_index = db_nslcmop["operationParams"].get("vdu_count_index") primitive = db_nslcmop["operationParams"]["primitive"] primitive_params = db_nslcmop["operationParams"]["primitive_params"] - timeout_ns_action = db_nslcmop["operationParams"].get("timeout_ns_action", self.timeout_primitive) + timeout_ns_action = db_nslcmop["operationParams"].get( + "timeout_ns_action", self.timeout.primitive + ) if vnf_index: step = "Getting vnfr from database" - db_vnfr = self.db.get_one("vnfrs", {"member-vnf-index-ref": vnf_index, "nsr-id-ref": nsr_id}) + db_vnfr = self.db.get_one( + "vnfrs", {"member-vnf-index-ref": vnf_index, "nsr-id-ref": nsr_id} + ) + if db_vnfr.get("kdur"): + kdur_list = [] + for kdur in db_vnfr["kdur"]: + if kdur.get("additionalParams"): + kdur["additionalParams"] = json.loads( + kdur["additionalParams"] + ) + kdur_list.append(kdur) + db_vnfr["kdur"] = kdur_list step = "Getting vnfd from database" db_vnfd = self.db.get_one("vnfds", {"_id": db_vnfr["vnfd-id"]}) + + # Sync filesystem before running a primitive + self.fs.sync(db_vnfr["vnfd-id"]) else: step = "Getting nsd from database" db_nsd = self.db.get_one("nsds", {"_id": db_nsr["nsd-id"]}) + vca_id = self.get_vca_id(db_vnfr, db_nsr) # for backward compatibility if nsr_deployed and isinstance(nsr_deployed.get("VCA"), dict): nsr_deployed["VCA"] = list(nsr_deployed["VCA"].values()) @@ -3509,7 +5106,9 @@ class NsLcm(LcmBase): else: descriptor_configuration = db_nsd.get("ns-configuration") - if descriptor_configuration and descriptor_configuration.get("config-primitive"): + if descriptor_configuration and descriptor_configuration.get( + "config-primitive" + ): for config_primitive in descriptor_configuration["config-primitive"]: if config_primitive["name"] == primitive: config_primitive_desc = config_primitive @@ -3517,23 +5116,36 @@ class NsLcm(LcmBase): if not config_primitive_desc: if not (kdu_name and primitive in ("upgrade", "rollback", "status")): - raise LcmException("Primitive {} not found at [ns|vnf|vdu]-configuration:config-primitive ". - format(primitive)) + raise LcmException( + "Primitive {} not found at [ns|vnf|vdu]-configuration:config-primitive ".format( + primitive + ) + ) primitive_name = primitive ee_descriptor_id = None else: - primitive_name = config_primitive_desc.get("execution-environment-primitive", primitive) - ee_descriptor_id = config_primitive_desc.get("execution-environment-ref") + primitive_name = config_primitive_desc.get( + "execution-environment-primitive", primitive + ) + ee_descriptor_id = config_primitive_desc.get( + "execution-environment-ref" + ) if vnf_index: if vdu_id: - vdur = next((x for x in db_vnfr["vdur"] if x["vdu-id-ref"] == vdu_id), None) + vdur = next( + (x for x in db_vnfr["vdur"] if x["vdu-id-ref"] == vdu_id), None + ) desc_params = parse_yaml_strings(vdur.get("additionalParams")) elif kdu_name: - kdur = next((x for x in db_vnfr["kdur"] if x["kdu-name"] == kdu_name), None) + kdur = next( + (x for x in db_vnfr["kdur"] if x["kdu-name"] == kdu_name), None + ) desc_params = parse_yaml_strings(kdur.get("additionalParams")) else: - desc_params = parse_yaml_strings(db_vnfr.get("additionalParamsForVnf")) + desc_params = parse_yaml_strings( + db_vnfr.get("additionalParamsForVnf") + ) else: desc_params = parse_yaml_strings(db_nsr.get("additionalParamsForNs")) if kdu_name and get_configuration(db_vnfd, kdu_name): @@ -3543,28 +5155,52 @@ class NsLcm(LcmBase): actions.add(primitive["name"]) for primitive in kdu_configuration.get("config-primitive", []): actions.add(primitive["name"]) - kdu_action = True if primitive_name in actions else False + kdu = find_in_list( + nsr_deployed["K8s"], + lambda kdu: kdu_name == kdu["kdu-name"] + and kdu["member-vnf-index"] == vnf_index, + ) + kdu_action = ( + True + if primitive_name in actions + and kdu["k8scluster-type"] not in ("helm-chart", "helm-chart-v3") + else False + ) # TODO check if ns is in a proper status - if kdu_name and (primitive_name in ("upgrade", "rollback", "status") or kdu_action): + if kdu_name and ( + primitive_name in ("upgrade", "rollback", "status") or kdu_action + ): # kdur and desc_params already set from before if primitive_params: desc_params.update(primitive_params) # TODO Check if we will need something at vnf level for index, kdu in enumerate(get_iterable(nsr_deployed, "K8s")): - if kdu_name == kdu["kdu-name"] and kdu["member-vnf-index"] == vnf_index: + if ( + kdu_name == kdu["kdu-name"] + and kdu["member-vnf-index"] == vnf_index + ): break else: - raise LcmException("KDU '{}' for vnf '{}' not deployed".format(kdu_name, vnf_index)) + raise LcmException( + "KDU '{}' for vnf '{}' not deployed".format(kdu_name, vnf_index) + ) if kdu.get("k8scluster-type") not in self.k8scluster_map: - msg = "unknown k8scluster-type '{}'".format(kdu.get("k8scluster-type")) + msg = "unknown k8scluster-type '{}'".format( + kdu.get("k8scluster-type") + ) raise LcmException(msg) - db_dict = {"collection": "nsrs", - "filter": {"_id": nsr_id}, - "path": "_admin.deployed.K8s.{}".format(index)} - self.logger.debug(logging_text + "Exec k8s {} on {}.{}".format(primitive_name, vnf_index, kdu_name)) + db_dict = { + "collection": "nsrs", + "filter": {"_id": nsr_id}, + "path": "_admin.deployed.K8s.{}".format(index), + } + self.logger.debug( + logging_text + + "Exec k8s {} on {}.{}".format(primitive_name, vnf_index, kdu_name) + ) step = "Executing kdu {}".format(primitive_name) if primitive_name == "upgrade": if desc_params.get("kdu_model"): @@ -3572,146 +5208,1053 @@ class NsLcm(LcmBase): del desc_params["kdu_model"] else: kdu_model = kdu.get("kdu-model") - parts = kdu_model.split(sep=":") - if len(parts) == 2: - kdu_model = parts[0] + if kdu_model.count("/") < 2: # helm chart is not embedded + parts = kdu_model.split(sep=":") + if len(parts) == 2: + kdu_model = parts[0] + if desc_params.get("kdu_atomic_upgrade"): + atomic_upgrade = desc_params.get( + "kdu_atomic_upgrade" + ).lower() in ("yes", "true", "1") + del desc_params["kdu_atomic_upgrade"] + else: + atomic_upgrade = True detailed_status = await asyncio.wait_for( self.k8scluster_map[kdu["k8scluster-type"]].upgrade( cluster_uuid=kdu.get("k8scluster-uuid"), kdu_instance=kdu.get("kdu-instance"), - atomic=True, kdu_model=kdu_model, - params=desc_params, db_dict=db_dict, - timeout=timeout_ns_action), - timeout=timeout_ns_action + 10) - self.logger.debug(logging_text + " Upgrade of kdu {} done".format(detailed_status)) + atomic=atomic_upgrade, + kdu_model=kdu_model, + params=desc_params, + db_dict=db_dict, + timeout=timeout_ns_action, + ), + timeout=timeout_ns_action + 10, + ) + self.logger.debug( + logging_text + " Upgrade of kdu {} done".format(detailed_status) + ) elif primitive_name == "rollback": detailed_status = await asyncio.wait_for( self.k8scluster_map[kdu["k8scluster-type"]].rollback( cluster_uuid=kdu.get("k8scluster-uuid"), kdu_instance=kdu.get("kdu-instance"), - db_dict=db_dict), - timeout=timeout_ns_action) + db_dict=db_dict, + ), + timeout=timeout_ns_action, + ) elif primitive_name == "status": detailed_status = await asyncio.wait_for( self.k8scluster_map[kdu["k8scluster-type"]].status_kdu( cluster_uuid=kdu.get("k8scluster-uuid"), - kdu_instance=kdu.get("kdu-instance")), - timeout=timeout_ns_action) + kdu_instance=kdu.get("kdu-instance"), + vca_id=vca_id, + ), + timeout=timeout_ns_action, + ) else: - kdu_instance = kdu.get("kdu-instance") or "{}-{}".format(kdu["kdu-name"], nsr_id) - params = self._map_primitive_params(config_primitive_desc, primitive_params, desc_params) + kdu_instance = kdu.get("kdu-instance") or "{}-{}".format( + kdu["kdu-name"], nsr_id + ) + params = self._map_primitive_params( + config_primitive_desc, primitive_params, desc_params + ) detailed_status = await asyncio.wait_for( self.k8scluster_map[kdu["k8scluster-type"]].exec_primitive( cluster_uuid=kdu.get("k8scluster-uuid"), kdu_instance=kdu_instance, primitive_name=primitive_name, - params=params, db_dict=db_dict, - timeout=timeout_ns_action), - timeout=timeout_ns_action) + params=params, + db_dict=db_dict, + timeout=timeout_ns_action, + vca_id=vca_id, + ), + timeout=timeout_ns_action, + ) if detailed_status: - nslcmop_operation_state = 'COMPLETED' + nslcmop_operation_state = "COMPLETED" else: - detailed_status = '' - nslcmop_operation_state = 'FAILED' + detailed_status = "" + nslcmop_operation_state = "FAILED" else: - ee_id, vca_type = self._look_for_deployed_vca(nsr_deployed["VCA"], member_vnf_index=vnf_index, - vdu_id=vdu_id, vdu_count_index=vdu_count_index, - ee_descriptor_id=ee_descriptor_id) - for vca_index, vca_deployed in enumerate(db_nsr['_admin']['deployed']['VCA']): + ee_id, vca_type = self._look_for_deployed_vca( + nsr_deployed["VCA"], + member_vnf_index=vnf_index, + vdu_id=vdu_id, + vdu_count_index=vdu_count_index, + ee_descriptor_id=ee_descriptor_id, + ) + for vca_index, vca_deployed in enumerate( + db_nsr["_admin"]["deployed"]["VCA"] + ): if vca_deployed.get("member-vnf-index") == vnf_index: - db_dict = {"collection": "nsrs", - "filter": {"_id": nsr_id}, - "path": "_admin.deployed.VCA.{}.".format(vca_index)} + db_dict = { + "collection": "nsrs", + "filter": {"_id": nsr_id}, + "path": "_admin.deployed.VCA.{}.".format(vca_index), + } break - nslcmop_operation_state, detailed_status = await self._ns_execute_primitive( + ( + nslcmop_operation_state, + detailed_status, + ) = await self._ns_execute_primitive( ee_id, primitive=primitive_name, - primitive_params=self._map_primitive_params(config_primitive_desc, primitive_params, desc_params), + primitive_params=self._map_primitive_params( + config_primitive_desc, primitive_params, desc_params + ), timeout=timeout_ns_action, vca_type=vca_type, - db_dict=db_dict) + db_dict=db_dict, + vca_id=vca_id, + ) db_nslcmop_update["detailed-status"] = detailed_status - error_description_nslcmop = detailed_status if nslcmop_operation_state == "FAILED" else "" - self.logger.debug(logging_text + " task Done with result {} {}".format(nslcmop_operation_state, - detailed_status)) + error_description_nslcmop = ( + detailed_status if nslcmop_operation_state == "FAILED" else "" + ) + self.logger.debug( + logging_text + + "Done with result {} {}".format( + nslcmop_operation_state, detailed_status + ) + ) return # database update is called inside finally except (DbException, LcmException, N2VCException, K8sException) as e: self.logger.error(logging_text + "Exit Exception {}".format(e)) exc = e except asyncio.CancelledError: - self.logger.error(logging_text + "Cancelled Exception while '{}'".format(step)) + self.logger.error( + logging_text + "Cancelled Exception while '{}'".format(step) + ) exc = "Operation was cancelled" except asyncio.TimeoutError: self.logger.error(logging_text + "Timeout while '{}'".format(step)) exc = "Timeout" except Exception as e: exc = traceback.format_exc() - self.logger.critical(logging_text + "Exit Exception {} {}".format(type(e).__name__, e), exc_info=True) + self.logger.critical( + logging_text + "Exit Exception {} {}".format(type(e).__name__, e), + exc_info=True, + ) finally: if exc: - db_nslcmop_update["detailed-status"] = detailed_status = error_description_nslcmop = \ - "FAILED {}: {}".format(step, exc) + db_nslcmop_update[ + "detailed-status" + ] = ( + detailed_status + ) = error_description_nslcmop = "FAILED {}: {}".format(step, exc) nslcmop_operation_state = "FAILED" if db_nsr: self._write_ns_status( nsr_id=nsr_id, - ns_state=db_nsr["nsState"], # TODO check if degraded. For the moment use previous status + ns_state=db_nsr[ + "nsState" + ], # TODO check if degraded. For the moment use previous status current_operation="IDLE", current_operation_id=None, # error_description=error_description_nsr, # error_detail=error_detail, - other_update=db_nsr_update + other_update=db_nsr_update, ) - self._write_op_status(op_id=nslcmop_id, stage="", error_message=error_description_nslcmop, - operation_state=nslcmop_operation_state, other_update=db_nslcmop_update) + self._write_op_status( + op_id=nslcmop_id, + stage="", + error_message=error_description_nslcmop, + operation_state=nslcmop_operation_state, + other_update=db_nslcmop_update, + ) if nslcmop_operation_state: try: - await self.msg.aiowrite("ns", "actioned", {"nsr_id": nsr_id, "nslcmop_id": nslcmop_id, - "operationState": nslcmop_operation_state}, - loop=self.loop) + await self.msg.aiowrite( + "ns", + "actioned", + { + "nsr_id": nsr_id, + "nslcmop_id": nslcmop_id, + "operationState": nslcmop_operation_state, + }, + ) except Exception as e: - self.logger.error(logging_text + "kafka_write notification Exception {}".format(e)) + self.logger.error( + logging_text + "kafka_write notification Exception {}".format(e) + ) self.logger.debug(logging_text + "Exit") self.lcm_tasks.remove("ns", nsr_id, nslcmop_id, "ns_action") return nslcmop_operation_state, detailed_status - async def scale(self, nsr_id, nslcmop_id): + async def terminate_vdus( + self, db_vnfr, member_vnf_index, db_nsr, update_db_nslcmops, stage, logging_text + ): + """This method terminates VDUs + + Args: + db_vnfr: VNF instance record + member_vnf_index: VNF index to identify the VDUs to be removed + db_nsr: NS instance record + update_db_nslcmops: Nslcmop update record + """ + vca_scaling_info = [] + scaling_info = {"scaling_group_name": "vdu_autoscale", "vdu": [], "kdu": []} + scaling_info["scaling_direction"] = "IN" + scaling_info["vdu-delete"] = {} + scaling_info["kdu-delete"] = {} + db_vdur = db_vnfr.get("vdur") + vdur_list = copy(db_vdur) + count_index = 0 + for index, vdu in enumerate(vdur_list): + vca_scaling_info.append( + { + "osm_vdu_id": vdu["vdu-id-ref"], + "member-vnf-index": member_vnf_index, + "type": "delete", + "vdu_index": count_index, + } + ) + scaling_info["vdu-delete"][vdu["vdu-id-ref"]] = count_index + scaling_info["vdu"].append( + { + "name": vdu.get("name") or vdu.get("vdu-name"), + "vdu_id": vdu["vdu-id-ref"], + "interface": [], + } + ) + for interface in vdu["interfaces"]: + scaling_info["vdu"][index]["interface"].append( + { + "name": interface["name"], + "ip_address": interface["ip-address"], + "mac_address": interface.get("mac-address"), + } + ) + self.logger.info("NS update scaling info{}".format(scaling_info)) + stage[2] = "Terminating VDUs" + if scaling_info.get("vdu-delete"): + # scale_process = "RO" + if self.ro_config.ng: + await self._scale_ng_ro( + logging_text, + db_nsr, + update_db_nslcmops, + db_vnfr, + scaling_info, + stage, + ) + + async def remove_vnf(self, nsr_id, nslcmop_id, vnf_instance_id): + """This method is to Remove VNF instances from NS. + + Args: + nsr_id: NS instance id + nslcmop_id: nslcmop id of update + vnf_instance_id: id of the VNF instance to be removed + + Returns: + result: (str, str) COMPLETED/FAILED, details + """ + try: + db_nsr_update = {} + logging_text = "Task ns={} update ".format(nsr_id) + check_vnfr_count = len(self.db.get_list("vnfrs", {"nsr-id-ref": nsr_id})) + self.logger.info("check_vnfr_count {}".format(check_vnfr_count)) + if check_vnfr_count > 1: + stage = ["", "", ""] + step = "Getting nslcmop from database" + self.logger.debug( + step + " after having waited for previous tasks to be completed" + ) + # db_nslcmop = self.db.get_one("nslcmops", {"_id": nslcmop_id}) + db_nsr = self.db.get_one("nsrs", {"_id": nsr_id}) + db_vnfr = self.db.get_one("vnfrs", {"_id": vnf_instance_id}) + member_vnf_index = db_vnfr["member-vnf-index-ref"] + """ db_vnfr = self.db.get_one( + "vnfrs", {"member-vnf-index-ref": member_vnf_index, "nsr-id-ref": nsr_id}) """ + + update_db_nslcmops = self.db.get_one("nslcmops", {"_id": nslcmop_id}) + await self.terminate_vdus( + db_vnfr, + member_vnf_index, + db_nsr, + update_db_nslcmops, + stage, + logging_text, + ) + + constituent_vnfr = db_nsr.get("constituent-vnfr-ref") + constituent_vnfr.remove(db_vnfr.get("_id")) + db_nsr_update["constituent-vnfr-ref"] = db_nsr.get( + "constituent-vnfr-ref" + ) + self.update_db_2("nsrs", nsr_id, db_nsr_update) + self.db.del_one("vnfrs", {"_id": db_vnfr.get("_id")}) + self.update_db_2("nsrs", nsr_id, db_nsr_update) + return "COMPLETED", "Done" + else: + step = "Terminate VNF Failed with" + raise LcmException( + "{} Cannot terminate the last VNF in this NS.".format( + vnf_instance_id + ) + ) + except (LcmException, asyncio.CancelledError): + raise + except Exception as e: + self.logger.debug("Error removing VNF {}".format(e)) + return "FAILED", "Error removing VNF {}".format(e) + + async def _ns_redeploy_vnf( + self, + nsr_id, + nslcmop_id, + db_vnfd, + db_vnfr, + db_nsr, + ): + """This method updates and redeploys VNF instances + + Args: + nsr_id: NS instance id + nslcmop_id: nslcmop id + db_vnfd: VNF descriptor + db_vnfr: VNF instance record + db_nsr: NS instance record + + Returns: + result: (str, str) COMPLETED/FAILED, details + """ + try: + count_index = 0 + stage = ["", "", ""] + logging_text = "Task ns={} update ".format(nsr_id) + latest_vnfd_revision = db_vnfd["_admin"].get("revision") + member_vnf_index = db_vnfr["member-vnf-index-ref"] + + # Terminate old VNF resources + update_db_nslcmops = self.db.get_one("nslcmops", {"_id": nslcmop_id}) + await self.terminate_vdus( + db_vnfr, + member_vnf_index, + db_nsr, + update_db_nslcmops, + stage, + logging_text, + ) + + # old_vnfd_id = db_vnfr["vnfd-id"] + # new_db_vnfd = self.db.get_one("vnfds", {"_id": vnfd_id}) + new_db_vnfd = db_vnfd + # new_vnfd_ref = new_db_vnfd["id"] + # new_vnfd_id = vnfd_id + + # Create VDUR + new_vnfr_cp = [] + for cp in new_db_vnfd.get("ext-cpd", ()): + vnf_cp = { + "name": cp.get("id"), + "connection-point-id": cp.get("int-cpd", {}).get("cpd"), + "connection-point-vdu-id": cp.get("int-cpd", {}).get("vdu-id"), + "id": cp.get("id"), + } + new_vnfr_cp.append(vnf_cp) + new_vdur = update_db_nslcmops["operationParams"]["newVdur"] + # new_vdur = self._create_vdur_descriptor_from_vnfd(db_nsd, db_vnfd, old_db_vnfd, vnfd_id, db_nsr, member_vnf_index) + # new_vnfr_update = {"vnfd-ref": new_vnfd_ref, "vnfd-id": new_vnfd_id, "connection-point": new_vnfr_cp, "vdur": new_vdur, "ip-address": ""} + new_vnfr_update = { + "revision": latest_vnfd_revision, + "connection-point": new_vnfr_cp, + "vdur": new_vdur, + "ip-address": "", + } + self.update_db_2("vnfrs", db_vnfr["_id"], new_vnfr_update) + updated_db_vnfr = self.db.get_one( + "vnfrs", + {"member-vnf-index-ref": member_vnf_index, "nsr-id-ref": nsr_id}, + ) + + # Instantiate new VNF resources + # update_db_nslcmops = self.db.get_one("nslcmops", {"_id": nslcmop_id}) + vca_scaling_info = [] + scaling_info = {"scaling_group_name": "vdu_autoscale", "vdu": [], "kdu": []} + scaling_info["scaling_direction"] = "OUT" + scaling_info["vdu-create"] = {} + scaling_info["kdu-create"] = {} + vdud_instantiate_list = db_vnfd["vdu"] + for index, vdud in enumerate(vdud_instantiate_list): + cloud_init_text = self._get_vdu_cloud_init_content(vdud, db_vnfd) + if cloud_init_text: + additional_params = ( + self._get_vdu_additional_params(updated_db_vnfr, vdud["id"]) + or {} + ) + cloud_init_list = [] + if cloud_init_text: + # TODO Information of its own ip is not available because db_vnfr is not updated. + additional_params["OSM"] = get_osm_params( + updated_db_vnfr, vdud["id"], 1 + ) + cloud_init_list.append( + self._parse_cloud_init( + cloud_init_text, + additional_params, + db_vnfd["id"], + vdud["id"], + ) + ) + vca_scaling_info.append( + { + "osm_vdu_id": vdud["id"], + "member-vnf-index": member_vnf_index, + "type": "create", + "vdu_index": count_index, + } + ) + scaling_info["vdu-create"][vdud["id"]] = count_index + if self.ro_config.ng: + self.logger.debug( + "New Resources to be deployed: {}".format(scaling_info) + ) + await self._scale_ng_ro( + logging_text, + db_nsr, + update_db_nslcmops, + updated_db_vnfr, + scaling_info, + stage, + ) + return "COMPLETED", "Done" + except (LcmException, asyncio.CancelledError): + raise + except Exception as e: + self.logger.debug("Error updating VNF {}".format(e)) + return "FAILED", "Error updating VNF {}".format(e) + + async def _ns_charm_upgrade( + self, + ee_id, + charm_id, + charm_type, + path, + timeout: float = None, + ) -> (str, str): + """This method upgrade charms in VNF instances + + Args: + ee_id: Execution environment id + path: Local path to the charm + charm_id: charm-id + charm_type: Charm type can be lxc-proxy-charm, native-charm or k8s-proxy-charm + timeout: (Float) Timeout for the ns update operation + + Returns: + result: (str, str) COMPLETED/FAILED, details + """ + try: + charm_type = charm_type or "lxc_proxy_charm" + output = await self.vca_map[charm_type].upgrade_charm( + ee_id=ee_id, + path=path, + charm_id=charm_id, + charm_type=charm_type, + timeout=timeout or self.timeout.ns_update, + ) + + if output: + return "COMPLETED", output + + except (LcmException, asyncio.CancelledError): + raise + + except Exception as e: + self.logger.debug("Error upgrading charm {}".format(path)) + + return "FAILED", "Error upgrading charm {}: {}".format(path, e) + + async def update(self, nsr_id, nslcmop_id): + """Update NS according to different update types + + This method performs upgrade of VNF instances then updates the revision + number in VNF record + + Args: + nsr_id: Network service will be updated + nslcmop_id: ns lcm operation id + + Returns: + It may raise DbException, LcmException, N2VCException, K8sException + + """ # Try to lock HA task here - task_is_locked_by_me = self.lcm_tasks.lock_HA('ns', 'nslcmops', nslcmop_id) + task_is_locked_by_me = self.lcm_tasks.lock_HA("ns", "nslcmops", nslcmop_id) if not task_is_locked_by_me: return - logging_text = "Task ns={} scale={} ".format(nsr_id, nslcmop_id) - stage = ['', '', ''] - tasks_dict_info = {} - # ^ stage, step, VIM progress + logging_text = "Task ns={} update={} ".format(nsr_id, nslcmop_id) self.logger.debug(logging_text + "Enter") - # get all needed from database + + # Set the required variables to be filled up later db_nsr = None db_nslcmop_update = {} + vnfr_update = {} + nslcmop_operation_state = None db_nsr_update = {} + error_description_nslcmop = "" exc = None - # in case of error, indicates what part of scale was failed to put nsr at error status - scale_process = None - old_operational_status = "" - old_config_status = "" - nsi_id = None + change_type = "updated" + detailed_status = "" + member_vnf_index = None + try: # wait for any previous tasks in process step = "Waiting for previous operations to terminate" - await self.lcm_tasks.waitfor_related_HA('ns', 'nslcmops', nslcmop_id) - self._write_ns_status(nsr_id=nsr_id, ns_state=None, - current_operation="SCALING", current_operation_id=nslcmop_id) - + await self.lcm_tasks.waitfor_related_HA("ns", "nslcmops", nslcmop_id) + self._write_ns_status( + nsr_id=nsr_id, + ns_state=None, + current_operation="UPDATING", + current_operation_id=nslcmop_id, + ) + + step = "Getting nslcmop from database" + db_nslcmop = self.db.get_one( + "nslcmops", {"_id": nslcmop_id}, fail_on_empty=False + ) + update_type = db_nslcmop["operationParams"]["updateType"] + + step = "Getting nsr from database" + db_nsr = self.db.get_one("nsrs", {"_id": nsr_id}) + old_operational_status = db_nsr["operational-status"] + db_nsr_update["operational-status"] = "updating" + self.update_db_2("nsrs", nsr_id, db_nsr_update) + nsr_deployed = db_nsr["_admin"].get("deployed") + + if update_type == "CHANGE_VNFPKG": + # Get the input parameters given through update request + vnf_instance_id = db_nslcmop["operationParams"][ + "changeVnfPackageData" + ].get("vnfInstanceId") + + vnfd_id = db_nslcmop["operationParams"]["changeVnfPackageData"].get( + "vnfdId" + ) + timeout_seconds = db_nslcmop["operationParams"].get("timeout_ns_update") + + step = "Getting vnfr from database" + db_vnfr = self.db.get_one( + "vnfrs", {"_id": vnf_instance_id}, fail_on_empty=False + ) + + step = "Getting vnfds from database" + # Latest VNFD + latest_vnfd = self.db.get_one( + "vnfds", {"_id": vnfd_id}, fail_on_empty=False + ) + latest_vnfd_revision = latest_vnfd["_admin"].get("revision") + + # Current VNFD + current_vnf_revision = db_vnfr.get("revision", 1) + current_vnfd = self.db.get_one( + "vnfds_revisions", + {"_id": vnfd_id + ":" + str(current_vnf_revision)}, + fail_on_empty=False, + ) + # Charm artifact paths will be filled up later + ( + current_charm_artifact_path, + target_charm_artifact_path, + charm_artifact_paths, + helm_artifacts, + ) = ([], [], [], []) + + step = "Checking if revision has changed in VNFD" + if current_vnf_revision != latest_vnfd_revision: + change_type = "policy_updated" + + # There is new revision of VNFD, update operation is required + current_vnfd_path = vnfd_id + ":" + str(current_vnf_revision) + latest_vnfd_path = vnfd_id + ":" + str(latest_vnfd_revision) + + step = "Removing the VNFD packages if they exist in the local path" + shutil.rmtree(self.fs.path + current_vnfd_path, ignore_errors=True) + shutil.rmtree(self.fs.path + latest_vnfd_path, ignore_errors=True) + + step = "Get the VNFD packages from FSMongo" + self.fs.sync(from_path=latest_vnfd_path) + self.fs.sync(from_path=current_vnfd_path) + + step = ( + "Get the charm-type, charm-id, ee-id if there is deployed VCA" + ) + current_base_folder = current_vnfd["_admin"]["storage"] + latest_base_folder = latest_vnfd["_admin"]["storage"] + + for vca_index, vca_deployed in enumerate( + get_iterable(nsr_deployed, "VCA") + ): + vnf_index = db_vnfr.get("member-vnf-index-ref") + + # Getting charm-id and charm-type + if vca_deployed.get("member-vnf-index") == vnf_index: + vca_id = self.get_vca_id(db_vnfr, db_nsr) + vca_type = vca_deployed.get("type") + vdu_count_index = vca_deployed.get("vdu_count_index") + + # Getting ee-id + ee_id = vca_deployed.get("ee_id") + + step = "Getting descriptor config" + if current_vnfd.get("kdu"): + search_key = "kdu_name" + else: + search_key = "vnfd_id" + + entity_id = vca_deployed.get(search_key) + + descriptor_config = get_configuration( + current_vnfd, entity_id + ) + + if "execution-environment-list" in descriptor_config: + ee_list = descriptor_config.get( + "execution-environment-list", [] + ) + else: + ee_list = [] + + # There could be several charm used in the same VNF + for ee_item in ee_list: + if ee_item.get("juju"): + step = "Getting charm name" + charm_name = ee_item["juju"].get("charm") + + step = "Setting Charm artifact paths" + current_charm_artifact_path.append( + get_charm_artifact_path( + current_base_folder, + charm_name, + vca_type, + current_vnf_revision, + ) + ) + target_charm_artifact_path.append( + get_charm_artifact_path( + latest_base_folder, + charm_name, + vca_type, + latest_vnfd_revision, + ) + ) + elif ee_item.get("helm-chart"): + # add chart to list and all parameters + step = "Getting helm chart name" + chart_name = ee_item.get("helm-chart") + if ( + ee_item.get("helm-version") + and ee_item.get("helm-version") == "v2" + ): + vca_type = "helm" + else: + vca_type = "helm-v3" + step = "Setting Helm chart artifact paths" + + helm_artifacts.append( + { + "current_artifact_path": get_charm_artifact_path( + current_base_folder, + chart_name, + vca_type, + current_vnf_revision, + ), + "target_artifact_path": get_charm_artifact_path( + latest_base_folder, + chart_name, + vca_type, + latest_vnfd_revision, + ), + "ee_id": ee_id, + "vca_index": vca_index, + "vdu_index": vdu_count_index, + } + ) + + charm_artifact_paths = zip( + current_charm_artifact_path, target_charm_artifact_path + ) + + step = "Checking if software version has changed in VNFD" + if find_software_version(current_vnfd) != find_software_version( + latest_vnfd + ): + step = "Checking if existing VNF has charm" + for current_charm_path, target_charm_path in list( + charm_artifact_paths + ): + if current_charm_path: + raise LcmException( + "Software version change is not supported as VNF instance {} has charm.".format( + vnf_instance_id + ) + ) + + # There is no change in the charm package, then redeploy the VNF + # based on new descriptor + step = "Redeploying VNF" + member_vnf_index = db_vnfr["member-vnf-index-ref"] + (result, detailed_status) = await self._ns_redeploy_vnf( + nsr_id, nslcmop_id, latest_vnfd, db_vnfr, db_nsr + ) + if result == "FAILED": + nslcmop_operation_state = result + error_description_nslcmop = detailed_status + db_nslcmop_update["detailed-status"] = detailed_status + self.logger.debug( + logging_text + + " step {} Done with result {} {}".format( + step, nslcmop_operation_state, detailed_status + ) + ) + + else: + step = "Checking if any charm package has changed or not" + for current_charm_path, target_charm_path in list( + charm_artifact_paths + ): + if ( + current_charm_path + and target_charm_path + and self.check_charm_hash_changed( + current_charm_path, target_charm_path + ) + ): + step = "Checking whether VNF uses juju bundle" + if check_juju_bundle_existence(current_vnfd): + raise LcmException( + "Charm upgrade is not supported for the instance which" + " uses juju-bundle: {}".format( + check_juju_bundle_existence(current_vnfd) + ) + ) + + step = "Upgrading Charm" + ( + result, + detailed_status, + ) = await self._ns_charm_upgrade( + ee_id=ee_id, + charm_id=vca_id, + charm_type=vca_type, + path=self.fs.path + target_charm_path, + timeout=timeout_seconds, + ) + + if result == "FAILED": + nslcmop_operation_state = result + error_description_nslcmop = detailed_status + + db_nslcmop_update["detailed-status"] = detailed_status + self.logger.debug( + logging_text + + " step {} Done with result {} {}".format( + step, nslcmop_operation_state, detailed_status + ) + ) + + step = "Updating policies" + member_vnf_index = db_vnfr["member-vnf-index-ref"] + result = "COMPLETED" + detailed_status = "Done" + db_nslcmop_update["detailed-status"] = "Done" + + # helm base EE + for item in helm_artifacts: + if not ( + item["current_artifact_path"] + and item["target_artifact_path"] + and self.check_charm_hash_changed( + item["current_artifact_path"], + item["target_artifact_path"], + ) + ): + continue + db_update_entry = "_admin.deployed.VCA.{}.".format( + item["vca_index"] + ) + vnfr_id = db_vnfr["_id"] + osm_config = {"osm": {"ns_id": nsr_id, "vnf_id": vnfr_id}} + db_dict = { + "collection": "nsrs", + "filter": {"_id": nsr_id}, + "path": db_update_entry, + } + vca_type, namespace, helm_id = get_ee_id_parts(item["ee_id"]) + await self.vca_map[vca_type].upgrade_execution_environment( + namespace=namespace, + helm_id=helm_id, + db_dict=db_dict, + config=osm_config, + artifact_path=item["target_artifact_path"], + vca_type=vca_type, + ) + vnf_id = db_vnfr.get("vnfd-ref") + config_descriptor = get_configuration(latest_vnfd, vnf_id) + self.logger.debug("get ssh key block") + rw_mgmt_ip = None + if deep_get( + config_descriptor, + ("config-access", "ssh-access", "required"), + ): + # Needed to inject a ssh key + user = deep_get( + config_descriptor, + ("config-access", "ssh-access", "default-user"), + ) + step = ( + "Install configuration Software, getting public ssh key" + ) + pub_key = await self.vca_map[ + vca_type + ].get_ee_ssh_public__key( + ee_id=ee_id, db_dict=db_dict, vca_id=vca_id + ) + + step = ( + "Insert public key into VM user={} ssh_key={}".format( + user, pub_key + ) + ) + self.logger.debug(logging_text + step) + + # wait for RO (ip-address) Insert pub_key into VM + rw_mgmt_ip = await self.wait_vm_up_insert_key_ro( + logging_text, + nsr_id, + vnfr_id, + None, + item["vdu_index"], + user=user, + pub_key=pub_key, + ) + + initial_config_primitive_list = config_descriptor.get( + "initial-config-primitive" + ) + config_primitive = next( + ( + p + for p in initial_config_primitive_list + if p["name"] == "config" + ), + None, + ) + if not config_primitive: + continue + + deploy_params = {"OSM": get_osm_params(db_vnfr)} + if rw_mgmt_ip: + deploy_params["rw_mgmt_ip"] = rw_mgmt_ip + if db_vnfr.get("additionalParamsForVnf"): + deploy_params.update( + parse_yaml_strings( + db_vnfr["additionalParamsForVnf"].copy() + ) + ) + primitive_params_ = self._map_primitive_params( + config_primitive, {}, deploy_params + ) + + step = "execute primitive '{}' params '{}'".format( + config_primitive["name"], primitive_params_ + ) + self.logger.debug(logging_text + step) + await self.vca_map[vca_type].exec_primitive( + ee_id=ee_id, + primitive_name=config_primitive["name"], + params_dict=primitive_params_, + db_dict=db_dict, + vca_id=vca_id, + vca_type=vca_type, + ) + + step = "Updating policies" + member_vnf_index = db_vnfr["member-vnf-index-ref"] + detailed_status = "Done" + db_nslcmop_update["detailed-status"] = "Done" + + # If nslcmop_operation_state is None, so any operation is not failed. + if not nslcmop_operation_state: + nslcmop_operation_state = "COMPLETED" + + # If update CHANGE_VNFPKG nslcmop_operation is successful + # vnf revision need to be updated + vnfr_update["revision"] = latest_vnfd_revision + self.update_db_2("vnfrs", db_vnfr["_id"], vnfr_update) + + self.logger.debug( + logging_text + + " task Done with result {} {}".format( + nslcmop_operation_state, detailed_status + ) + ) + elif update_type == "REMOVE_VNF": + # This part is included in https://osm.etsi.org/gerrit/11876 + vnf_instance_id = db_nslcmop["operationParams"]["removeVnfInstanceId"] + db_vnfr = self.db.get_one("vnfrs", {"_id": vnf_instance_id}) + member_vnf_index = db_vnfr["member-vnf-index-ref"] + step = "Removing VNF" + (result, detailed_status) = await self.remove_vnf( + nsr_id, nslcmop_id, vnf_instance_id + ) + if result == "FAILED": + nslcmop_operation_state = result + error_description_nslcmop = detailed_status + db_nslcmop_update["detailed-status"] = detailed_status + change_type = "vnf_terminated" + if not nslcmop_operation_state: + nslcmop_operation_state = "COMPLETED" + self.logger.debug( + logging_text + + " task Done with result {} {}".format( + nslcmop_operation_state, detailed_status + ) + ) + + elif update_type == "OPERATE_VNF": + vnf_id = db_nslcmop["operationParams"]["operateVnfData"][ + "vnfInstanceId" + ] + operation_type = db_nslcmop["operationParams"]["operateVnfData"][ + "changeStateTo" + ] + additional_param = db_nslcmop["operationParams"]["operateVnfData"][ + "additionalParam" + ] + (result, detailed_status) = await self.rebuild_start_stop( + nsr_id, nslcmop_id, vnf_id, additional_param, operation_type + ) + if result == "FAILED": + nslcmop_operation_state = result + error_description_nslcmop = detailed_status + db_nslcmop_update["detailed-status"] = detailed_status + if not nslcmop_operation_state: + nslcmop_operation_state = "COMPLETED" + self.logger.debug( + logging_text + + " task Done with result {} {}".format( + nslcmop_operation_state, detailed_status + ) + ) + + # If nslcmop_operation_state is None, so any operation is not failed. + # All operations are executed in overall. + if not nslcmop_operation_state: + nslcmop_operation_state = "COMPLETED" + db_nsr_update["operational-status"] = old_operational_status + + except (DbException, LcmException, N2VCException, K8sException) as e: + self.logger.error(logging_text + "Exit Exception {}".format(e)) + exc = e + except asyncio.CancelledError: + self.logger.error( + logging_text + "Cancelled Exception while '{}'".format(step) + ) + exc = "Operation was cancelled" + except asyncio.TimeoutError: + self.logger.error(logging_text + "Timeout while '{}'".format(step)) + exc = "Timeout" + except Exception as e: + exc = traceback.format_exc() + self.logger.critical( + logging_text + "Exit Exception {} {}".format(type(e).__name__, e), + exc_info=True, + ) + finally: + if exc: + db_nslcmop_update[ + "detailed-status" + ] = ( + detailed_status + ) = error_description_nslcmop = "FAILED {}: {}".format(step, exc) + nslcmop_operation_state = "FAILED" + db_nsr_update["operational-status"] = old_operational_status + if db_nsr: + self._write_ns_status( + nsr_id=nsr_id, + ns_state=db_nsr["nsState"], + current_operation="IDLE", + current_operation_id=None, + other_update=db_nsr_update, + ) + + self._write_op_status( + op_id=nslcmop_id, + stage="", + error_message=error_description_nslcmop, + operation_state=nslcmop_operation_state, + other_update=db_nslcmop_update, + ) + + if nslcmop_operation_state: + try: + msg = { + "nsr_id": nsr_id, + "nslcmop_id": nslcmop_id, + "operationState": nslcmop_operation_state, + } + if ( + change_type in ("vnf_terminated", "policy_updated") + and member_vnf_index + ): + msg.update({"vnf_member_index": member_vnf_index}) + await self.msg.aiowrite("ns", change_type, msg) + except Exception as e: + self.logger.error( + logging_text + "kafka_write notification Exception {}".format(e) + ) + self.logger.debug(logging_text + "Exit") + self.lcm_tasks.remove("ns", nsr_id, nslcmop_id, "ns_update") + return nslcmop_operation_state, detailed_status + + async def scale(self, nsr_id, nslcmop_id): + # Try to lock HA task here + task_is_locked_by_me = self.lcm_tasks.lock_HA("ns", "nslcmops", nslcmop_id) + if not task_is_locked_by_me: + return + + logging_text = "Task ns={} scale={} ".format(nsr_id, nslcmop_id) + stage = ["", "", ""] + tasks_dict_info = {} + # ^ stage, step, VIM progress + self.logger.debug(logging_text + "Enter") + # get all needed from database + db_nsr = None + db_nslcmop_update = {} + db_nsr_update = {} + exc = None + # in case of error, indicates what part of scale was failed to put nsr at error status + scale_process = None + old_operational_status = "" + old_config_status = "" + nsi_id = None + try: + # wait for any previous tasks in process + step = "Waiting for previous operations to terminate" + await self.lcm_tasks.waitfor_related_HA("ns", "nslcmops", nslcmop_id) + self._write_ns_status( + nsr_id=nsr_id, + ns_state=None, + current_operation="SCALING", + current_operation_id=nslcmop_id, + ) + step = "Getting nslcmop from database" - self.logger.debug(step + " after having waited for previous tasks to be completed") + self.logger.debug( + step + " after having waited for previous tasks to be completed" + ) db_nslcmop = self.db.get_one("nslcmops", {"_id": nslcmop_id}) step = "Getting nsr from database" @@ -3724,16 +6267,12 @@ class NsLcm(LcmBase): self.update_db_2("nsrs", nsr_id, db_nsr_update) nsr_deployed = db_nsr["_admin"].get("deployed") - ####### - nsr_deployed = db_nsr["_admin"].get("deployed") - vnf_index = db_nslcmop["operationParams"].get("member_vnf_index") - # vdu_id = db_nslcmop["operationParams"].get("vdu_id") - # vdu_count_index = db_nslcmop["operationParams"].get("vdu_count_index") - # vdu_name = db_nslcmop["operationParams"].get("vdu_name") - ####### - - vnf_index = db_nslcmop["operationParams"]["scaleVnfData"]["scaleByStepData"]["member-vnf-index"] - scaling_group = db_nslcmop["operationParams"]["scaleVnfData"]["scaleByStepData"]["scaling-group-descriptor"] + vnf_index = db_nslcmop["operationParams"]["scaleVnfData"][ + "scaleByStepData" + ]["member-vnf-index"] + scaling_group = db_nslcmop["operationParams"]["scaleVnfData"][ + "scaleByStepData" + ]["scaling-group-descriptor"] scaling_type = db_nslcmop["operationParams"]["scaleVnfData"]["scaleVnfType"] # for backward compatibility if nsr_deployed and isinstance(nsr_deployed.get("VCA"), dict): @@ -3742,7 +6281,11 @@ class NsLcm(LcmBase): self.update_db_2("nsrs", nsr_id, db_nsr_update) step = "Getting vnfr from database" - db_vnfr = self.db.get_one("vnfrs", {"member-vnf-index-ref": vnf_index, "nsr-id-ref": nsr_id}) + db_vnfr = self.db.get_one( + "vnfrs", {"member-vnf-index-ref": vnf_index, "nsr-id-ref": nsr_id} + ) + + vca_id = self.get_vca_id(db_vnfr, db_nsr) step = "Getting vnfd from database" db_vnfd = self.db.get_one("vnfds", {"_id": db_vnfr["vnfd-id"]}) @@ -3751,32 +6294,44 @@ class NsLcm(LcmBase): step = "Getting scaling-group-descriptor" scaling_descriptor = find_in_list( - get_scaling_aspect( - db_vnfd - ), - lambda scale_desc: scale_desc["name"] == scaling_group + get_scaling_aspect(db_vnfd), + lambda scale_desc: scale_desc["name"] == scaling_group, ) if not scaling_descriptor: - raise LcmException("input parameter 'scaleByStepData':'scaling-group-descriptor':'{}' is not present " - "at vnfd:scaling-group-descriptor".format(scaling_group)) + raise LcmException( + "input parameter 'scaleByStepData':'scaling-group-descriptor':'{}' is not present " + "at vnfd:scaling-group-descriptor".format(scaling_group) + ) step = "Sending scale order to VIM" # TODO check if ns is in a proper status nb_scale_op = 0 if not db_nsr["_admin"].get("scaling-group"): - self.update_db_2("nsrs", nsr_id, {"_admin.scaling-group": [{"name": scaling_group, "nb-scale-op": 0}]}) + self.update_db_2( + "nsrs", + nsr_id, + { + "_admin.scaling-group": [ + {"name": scaling_group, "nb-scale-op": 0} + ] + }, + ) admin_scale_index = 0 else: - for admin_scale_index, admin_scale_info in enumerate(db_nsr["_admin"]["scaling-group"]): + for admin_scale_index, admin_scale_info in enumerate( + db_nsr["_admin"]["scaling-group"] + ): if admin_scale_info["name"] == scaling_group: nb_scale_op = admin_scale_info.get("nb-scale-op", 0) break else: # not found, set index one plus last element and add new entry with the name admin_scale_index += 1 - db_nsr_update["_admin.scaling-group.{}.name".format(admin_scale_index)] = scaling_group - RO_scaling_info = [] - VCA_scaling_info = [] - vdu_scaling_info = {"scaling_group_name": scaling_group, "vdu": []} + db_nsr_update[ + "_admin.scaling-group.{}.name".format(admin_scale_index) + ] = scaling_group + + vca_scaling_info = [] + scaling_info = {"scaling_group_name": scaling_group, "vdu": [], "kdu": []} if scaling_type == "SCALE_OUT": if "aspect-delta-details" not in scaling_descriptor: raise LcmException( @@ -3787,194 +6342,439 @@ class NsLcm(LcmBase): # count if max-instance-count is reached deltas = scaling_descriptor.get("aspect-delta-details")["deltas"] - vdu_scaling_info["scaling_direction"] = "OUT" - vdu_scaling_info["vdu-create"] = {} + scaling_info["scaling_direction"] = "OUT" + scaling_info["vdu-create"] = {} + scaling_info["kdu-create"] = {} for delta in deltas: - for vdu_delta in delta["vdu-delta"]: + for vdu_delta in delta.get("vdu-delta", {}): vdud = get_vdu(db_vnfd, vdu_delta["id"]) - vdu_index = get_vdur_index(db_vnfr, vdu_delta) - cloud_init_text = self._get_vdu_cloud_init_content(vdud, db_vnfd) + # vdu_index also provides the number of instance of the targeted vdu + vdu_count = vdu_index = get_vdur_index(db_vnfr, vdu_delta) + cloud_init_text = self._get_vdu_cloud_init_content( + vdud, db_vnfd + ) if cloud_init_text: - additional_params = self._get_vdu_additional_params(db_vnfr, vdud["id"]) or {} + additional_params = ( + self._get_vdu_additional_params(db_vnfr, vdud["id"]) + or {} + ) cloud_init_list = [] vdu_profile = get_vdu_profile(db_vnfd, vdu_delta["id"]) max_instance_count = 10 if vdu_profile and "max-number-of-instances" in vdu_profile: - max_instance_count = vdu_profile.get("max-number-of-instances", 10) - - default_instance_num = get_number_of_instances(db_vnfd, vdud["id"]) + max_instance_count = vdu_profile.get( + "max-number-of-instances", 10 + ) - nb_scale_op += vdu_delta.get("number-of-instances", 1) + default_instance_num = get_number_of_instances( + db_vnfd, vdud["id"] + ) + instances_number = vdu_delta.get("number-of-instances", 1) + nb_scale_op += instances_number + + new_instance_count = nb_scale_op + default_instance_num + # Control if new count is over max and vdu count is less than max. + # Then assign new instance count + if new_instance_count > max_instance_count > vdu_count: + instances_number = new_instance_count - max_instance_count + else: + instances_number = instances_number - if nb_scale_op + default_instance_num > max_instance_count: + if new_instance_count > max_instance_count: raise LcmException( "reached the limit of {} (max-instance-count) " "scaling-out operations for the " - "scaling-group-descriptor '{}'".format(nb_scale_op, scaling_group) + "scaling-group-descriptor '{}'".format( + nb_scale_op, scaling_group + ) ) for x in range(vdu_delta.get("number-of-instances", 1)): if cloud_init_text: # TODO Information of its own ip is not available because db_vnfr is not updated. additional_params["OSM"] = get_osm_params( - db_vnfr, - vdu_delta["id"], - vdu_index + x + db_vnfr, vdu_delta["id"], vdu_index + x ) cloud_init_list.append( self._parse_cloud_init( cloud_init_text, additional_params, db_vnfd["id"], - vdud["id"] + vdud["id"], ) ) - VCA_scaling_info.append( + vca_scaling_info.append( { "osm_vdu_id": vdu_delta["id"], "member-vnf-index": vnf_index, "type": "create", - "vdu_index": vdu_index + x + "vdu_index": vdu_index + x, } ) - RO_scaling_info.append( + scaling_info["vdu-create"][vdu_delta["id"]] = instances_number + for kdu_delta in delta.get("kdu-resource-delta", {}): + kdu_profile = get_kdu_resource_profile(db_vnfd, kdu_delta["id"]) + kdu_name = kdu_profile["kdu-name"] + resource_name = kdu_profile.get("resource-name", "") + + # Might have different kdus in the same delta + # Should have list for each kdu + if not scaling_info["kdu-create"].get(kdu_name, None): + scaling_info["kdu-create"][kdu_name] = [] + + kdur = get_kdur(db_vnfr, kdu_name) + if kdur.get("helm-chart"): + k8s_cluster_type = "helm-chart-v3" + self.logger.debug("kdur: {}".format(kdur)) + if ( + kdur.get("helm-version") + and kdur.get("helm-version") == "v2" + ): + k8s_cluster_type = "helm-chart" + elif kdur.get("juju-bundle"): + k8s_cluster_type = "juju-bundle" + else: + raise LcmException( + "kdu type for kdu='{}.{}' is neither helm-chart nor " + "juju-bundle. Maybe an old NBI version is running".format( + db_vnfr["member-vnf-index-ref"], kdu_name + ) + ) + + max_instance_count = 10 + if kdu_profile and "max-number-of-instances" in kdu_profile: + max_instance_count = kdu_profile.get( + "max-number-of-instances", 10 + ) + + nb_scale_op += kdu_delta.get("number-of-instances", 1) + deployed_kdu, _ = get_deployed_kdu( + nsr_deployed, kdu_name, vnf_index + ) + if deployed_kdu is None: + raise LcmException( + "KDU '{}' for vnf '{}' not deployed".format( + kdu_name, vnf_index + ) + ) + kdu_instance = deployed_kdu.get("kdu-instance") + instance_num = await self.k8scluster_map[ + k8s_cluster_type + ].get_scale_count( + resource_name, + kdu_instance, + vca_id=vca_id, + cluster_uuid=deployed_kdu.get("k8scluster-uuid"), + kdu_model=deployed_kdu.get("kdu-model"), + ) + kdu_replica_count = instance_num + kdu_delta.get( + "number-of-instances", 1 + ) + + # Control if new count is over max and instance_num is less than max. + # Then assign max instance number to kdu replica count + if kdu_replica_count > max_instance_count > instance_num: + kdu_replica_count = max_instance_count + if kdu_replica_count > max_instance_count: + raise LcmException( + "reached the limit of {} (max-instance-count) " + "scaling-out operations for the " + "scaling-group-descriptor '{}'".format( + instance_num, scaling_group + ) + ) + + for x in range(kdu_delta.get("number-of-instances", 1)): + vca_scaling_info.append( + { + "osm_kdu_id": kdu_name, + "member-vnf-index": vnf_index, + "type": "create", + "kdu_index": instance_num + x - 1, + } + ) + scaling_info["kdu-create"][kdu_name].append( { - "osm_vdu_id": vdu_delta["id"], "member-vnf-index": vnf_index, "type": "create", - "count": vdu_delta.get("number-of-instances", 1) + "k8s-cluster-type": k8s_cluster_type, + "resource-name": resource_name, + "scale": kdu_replica_count, } ) - if cloud_init_list: - RO_scaling_info[-1]["cloud_init"] = cloud_init_list - vdu_scaling_info["vdu-create"][vdu_delta["id"]] = vdu_delta.get("number-of-instances", 1) - elif scaling_type == "SCALE_IN": - if "min-instance-count" in scaling_descriptor and scaling_descriptor["min-instance-count"] is not None: - min_instance_count = int(scaling_descriptor["min-instance-count"]) - - vdu_scaling_info["scaling_direction"] = "IN" - vdu_scaling_info["vdu-delete"] = {} deltas = scaling_descriptor.get("aspect-delta-details")["deltas"] + + scaling_info["scaling_direction"] = "IN" + scaling_info["vdu-delete"] = {} + scaling_info["kdu-delete"] = {} + for delta in deltas: - for vdu_delta in delta["vdu-delta"]: - vdu_index = get_vdur_index(db_vnfr, vdu_delta) + for vdu_delta in delta.get("vdu-delta", {}): + vdu_count = vdu_index = get_vdur_index(db_vnfr, vdu_delta) min_instance_count = 0 vdu_profile = get_vdu_profile(db_vnfd, vdu_delta["id"]) if vdu_profile and "min-number-of-instances" in vdu_profile: min_instance_count = vdu_profile["min-number-of-instances"] - default_instance_num = get_number_of_instances(db_vnfd, vdu_delta["id"]) + default_instance_num = get_number_of_instances( + db_vnfd, vdu_delta["id"] + ) + instance_num = vdu_delta.get("number-of-instances", 1) + nb_scale_op -= instance_num + + new_instance_count = nb_scale_op + default_instance_num + + if new_instance_count < min_instance_count < vdu_count: + instances_number = min_instance_count - new_instance_count + else: + instances_number = instance_num - nb_scale_op -= vdu_delta.get("number-of-instances", 1) - if nb_scale_op + default_instance_num < min_instance_count: + if new_instance_count < min_instance_count: raise LcmException( "reached the limit of {} (min-instance-count) scaling-in operations for the " - "scaling-group-descriptor '{}'".format(nb_scale_op, scaling_group) + "scaling-group-descriptor '{}'".format( + nb_scale_op, scaling_group + ) ) - RO_scaling_info.append({"osm_vdu_id": vdu_delta["id"], "member-vnf-index": vnf_index, - "type": "delete", "count": vdu_delta.get("number-of-instances", 1), - "vdu_index": vdu_index - 1}) for x in range(vdu_delta.get("number-of-instances", 1)): - VCA_scaling_info.append( + vca_scaling_info.append( { "osm_vdu_id": vdu_delta["id"], "member-vnf-index": vnf_index, "type": "delete", - "vdu_index": vdu_index - 1 - x + "vdu_index": vdu_index - 1 - x, + } + ) + scaling_info["vdu-delete"][vdu_delta["id"]] = instances_number + for kdu_delta in delta.get("kdu-resource-delta", {}): + kdu_profile = get_kdu_resource_profile(db_vnfd, kdu_delta["id"]) + kdu_name = kdu_profile["kdu-name"] + resource_name = kdu_profile.get("resource-name", "") + + if not scaling_info["kdu-delete"].get(kdu_name, None): + scaling_info["kdu-delete"][kdu_name] = [] + + kdur = get_kdur(db_vnfr, kdu_name) + if kdur.get("helm-chart"): + k8s_cluster_type = "helm-chart-v3" + self.logger.debug("kdur: {}".format(kdur)) + if ( + kdur.get("helm-version") + and kdur.get("helm-version") == "v2" + ): + k8s_cluster_type = "helm-chart" + elif kdur.get("juju-bundle"): + k8s_cluster_type = "juju-bundle" + else: + raise LcmException( + "kdu type for kdu='{}.{}' is neither helm-chart nor " + "juju-bundle. Maybe an old NBI version is running".format( + db_vnfr["member-vnf-index-ref"], kdur["kdu-name"] + ) + ) + + min_instance_count = 0 + if kdu_profile and "min-number-of-instances" in kdu_profile: + min_instance_count = kdu_profile["min-number-of-instances"] + + nb_scale_op -= kdu_delta.get("number-of-instances", 1) + deployed_kdu, _ = get_deployed_kdu( + nsr_deployed, kdu_name, vnf_index + ) + if deployed_kdu is None: + raise LcmException( + "KDU '{}' for vnf '{}' not deployed".format( + kdu_name, vnf_index + ) + ) + kdu_instance = deployed_kdu.get("kdu-instance") + instance_num = await self.k8scluster_map[ + k8s_cluster_type + ].get_scale_count( + resource_name, + kdu_instance, + vca_id=vca_id, + cluster_uuid=deployed_kdu.get("k8scluster-uuid"), + kdu_model=deployed_kdu.get("kdu-model"), + ) + kdu_replica_count = instance_num - kdu_delta.get( + "number-of-instances", 1 + ) + + if kdu_replica_count < min_instance_count < instance_num: + kdu_replica_count = min_instance_count + if kdu_replica_count < min_instance_count: + raise LcmException( + "reached the limit of {} (min-instance-count) scaling-in operations for the " + "scaling-group-descriptor '{}'".format( + instance_num, scaling_group + ) + ) + + for x in range(kdu_delta.get("number-of-instances", 1)): + vca_scaling_info.append( + { + "osm_kdu_id": kdu_name, + "member-vnf-index": vnf_index, + "type": "delete", + "kdu_index": instance_num - x - 1, } ) - vdu_scaling_info["vdu-delete"][vdu_delta["id"]] = vdu_delta.get("number-of-instances", 1) + scaling_info["kdu-delete"][kdu_name].append( + { + "member-vnf-index": vnf_index, + "type": "delete", + "k8s-cluster-type": k8s_cluster_type, + "resource-name": resource_name, + "scale": kdu_replica_count, + } + ) # update VDU_SCALING_INFO with the VDUs to delete ip_addresses - vdu_delete = copy(vdu_scaling_info.get("vdu-delete")) - if vdu_scaling_info["scaling_direction"] == "IN": + vdu_delete = copy(scaling_info.get("vdu-delete")) + if scaling_info["scaling_direction"] == "IN": for vdur in reversed(db_vnfr["vdur"]): if vdu_delete.get(vdur["vdu-id-ref"]): vdu_delete[vdur["vdu-id-ref"]] -= 1 - vdu_scaling_info["vdu"].append({ - "name": vdur.get("name") or vdur.get("vdu-name"), - "vdu_id": vdur["vdu-id-ref"], - "interface": [] - }) + scaling_info["vdu"].append( + { + "name": vdur.get("name") or vdur.get("vdu-name"), + "vdu_id": vdur["vdu-id-ref"], + "interface": [], + } + ) for interface in vdur["interfaces"]: - vdu_scaling_info["vdu"][-1]["interface"].append({ - "name": interface["name"], - "ip_address": interface["ip-address"], - "mac_address": interface.get("mac-address"), - }) + scaling_info["vdu"][-1]["interface"].append( + { + "name": interface["name"], + "ip_address": interface["ip-address"], + "mac_address": interface.get("mac-address"), + } + ) # vdu_delete = vdu_scaling_info.pop("vdu-delete") # PRE-SCALE BEGIN step = "Executing pre-scale vnf-config-primitive" if scaling_descriptor.get("scaling-config-action"): - for scaling_config_action in scaling_descriptor["scaling-config-action"]: - if (scaling_config_action.get("trigger") == "pre-scale-in" and scaling_type == "SCALE_IN") \ - or (scaling_config_action.get("trigger") == "pre-scale-out" and scaling_type == "SCALE_OUT"): - vnf_config_primitive = scaling_config_action["vnf-config-primitive-name-ref"] - step = db_nslcmop_update["detailed-status"] = \ - "executing pre-scale scaling-config-action '{}'".format(vnf_config_primitive) + for scaling_config_action in scaling_descriptor[ + "scaling-config-action" + ]: + if ( + scaling_config_action.get("trigger") == "pre-scale-in" + and scaling_type == "SCALE_IN" + ) or ( + scaling_config_action.get("trigger") == "pre-scale-out" + and scaling_type == "SCALE_OUT" + ): + vnf_config_primitive = scaling_config_action[ + "vnf-config-primitive-name-ref" + ] + step = db_nslcmop_update[ + "detailed-status" + ] = "executing pre-scale scaling-config-action '{}'".format( + vnf_config_primitive + ) # look for primitive - for config_primitive in (get_configuration( - db_vnfd, db_vnfd["id"] - ) or {}).get("config-primitive", ()): + for config_primitive in ( + get_configuration(db_vnfd, db_vnfd["id"]) or {} + ).get("config-primitive", ()): if config_primitive["name"] == vnf_config_primitive: break else: raise LcmException( "Invalid vnfd descriptor at scaling-group-descriptor[name='{}']:scaling-config-action" "[vnf-config-primitive-name-ref='{}'] does not match any vnf-configuration:config-" - "primitive".format(scaling_group, vnf_config_primitive)) + "primitive".format(scaling_group, vnf_config_primitive) + ) - vnfr_params = {"VDU_SCALE_INFO": vdu_scaling_info} + vnfr_params = {"VDU_SCALE_INFO": scaling_info} if db_vnfr.get("additionalParamsForVnf"): vnfr_params.update(db_vnfr["additionalParamsForVnf"]) scale_process = "VCA" db_nsr_update["config-status"] = "configuring pre-scaling" - primitive_params = self._map_primitive_params(config_primitive, {}, vnfr_params) + primitive_params = self._map_primitive_params( + config_primitive, {}, vnfr_params + ) # Pre-scale retry check: Check if this sub-operation has been executed before op_index = self._check_or_add_scale_suboperation( - db_nslcmop, nslcmop_id, vnf_index, vnf_config_primitive, primitive_params, 'PRE-SCALE') + db_nslcmop, + vnf_index, + vnf_config_primitive, + primitive_params, + "PRE-SCALE", + ) if op_index == self.SUBOPERATION_STATUS_SKIP: # Skip sub-operation - result = 'COMPLETED' - result_detail = 'Done' - self.logger.debug(logging_text + - "vnf_config_primitive={} Skipped sub-operation, result {} {}".format( - vnf_config_primitive, result, result_detail)) + result = "COMPLETED" + result_detail = "Done" + self.logger.debug( + logging_text + + "vnf_config_primitive={} Skipped sub-operation, result {} {}".format( + vnf_config_primitive, result, result_detail + ) + ) else: if op_index == self.SUBOPERATION_STATUS_NEW: # New sub-operation: Get index of this sub-operation - op_index = len(db_nslcmop.get('_admin', {}).get('operations')) - 1 - self.logger.debug(logging_text + "vnf_config_primitive={} New sub-operation". - format(vnf_config_primitive)) - else: + op_index = ( + len(db_nslcmop.get("_admin", {}).get("operations")) + - 1 + ) + self.logger.debug( + logging_text + + "vnf_config_primitive={} New sub-operation".format( + vnf_config_primitive + ) + ) + else: # retry: Get registered params for this existing sub-operation - op = db_nslcmop.get('_admin', {}).get('operations', [])[op_index] - vnf_index = op.get('member_vnf_index') - vnf_config_primitive = op.get('primitive') - primitive_params = op.get('primitive_params') - self.logger.debug(logging_text + "vnf_config_primitive={} Sub-operation retry". - format(vnf_config_primitive)) + op = db_nslcmop.get("_admin", {}).get("operations", [])[ + op_index + ] + vnf_index = op.get("member_vnf_index") + vnf_config_primitive = op.get("primitive") + primitive_params = op.get("primitive_params") + self.logger.debug( + logging_text + + "vnf_config_primitive={} Sub-operation retry".format( + vnf_config_primitive + ) + ) # Execute the primitive, either with new (first-time) or registered (reintent) args - ee_descriptor_id = config_primitive.get("execution-environment-ref") - primitive_name = config_primitive.get("execution-environment-primitive", - vnf_config_primitive) - ee_id, vca_type = self._look_for_deployed_vca(nsr_deployed["VCA"], - member_vnf_index=vnf_index, - vdu_id=None, - vdu_count_index=None, - ee_descriptor_id=ee_descriptor_id) + ee_descriptor_id = config_primitive.get( + "execution-environment-ref" + ) + primitive_name = config_primitive.get( + "execution-environment-primitive", vnf_config_primitive + ) + ee_id, vca_type = self._look_for_deployed_vca( + nsr_deployed["VCA"], + member_vnf_index=vnf_index, + vdu_id=None, + vdu_count_index=None, + ee_descriptor_id=ee_descriptor_id, + ) result, result_detail = await self._ns_execute_primitive( - ee_id, primitive_name, primitive_params, vca_type=vca_type) - self.logger.debug(logging_text + "vnf_config_primitive={} Done with result {} {}".format( - vnf_config_primitive, result, result_detail)) + ee_id, + primitive_name, + primitive_params, + vca_type=vca_type, + vca_id=vca_id, + ) + self.logger.debug( + logging_text + + "vnf_config_primitive={} Done with result {} {}".format( + vnf_config_primitive, result, result_detail + ) + ) # Update operationState = COMPLETED | FAILED self._update_suboperation_status( - db_nslcmop, op_index, result, result_detail) + db_nslcmop, op_index, result, result_detail + ) if result == "FAILED": raise LcmException(result_detail) @@ -3982,191 +6782,240 @@ class NsLcm(LcmBase): scale_process = None # PRE-SCALE END - db_nsr_update["_admin.scaling-group.{}.nb-scale-op".format(admin_scale_index)] = nb_scale_op - db_nsr_update["_admin.scaling-group.{}.time".format(admin_scale_index)] = time() + db_nsr_update[ + "_admin.scaling-group.{}.nb-scale-op".format(admin_scale_index) + ] = nb_scale_op + db_nsr_update[ + "_admin.scaling-group.{}.time".format(admin_scale_index) + ] = time() # SCALE-IN VCA - BEGIN - if VCA_scaling_info: - step = db_nslcmop_update["detailed-status"] = \ - "Deleting the execution environments" + if vca_scaling_info: + step = db_nslcmop_update[ + "detailed-status" + ] = "Deleting the execution environments" scale_process = "VCA" - for vdu_info in VCA_scaling_info: - if vdu_info["type"] == "delete": - member_vnf_index = str(vdu_info["member-vnf-index"]) - self.logger.debug(logging_text + "vdu info: {}".format(vdu_info)) - vdu_id = vdu_info["osm_vdu_id"] - vdu_index = int(vdu_info["vdu_index"]) - stage[1] = "Scaling member_vnf_index={}, vdu_id={}, vdu_index={} ".format( - member_vnf_index, vdu_id, vdu_index) - stage[2] = step = "Scaling in VCA" - self._write_op_status( - op_id=nslcmop_id, - stage=stage + for vca_info in vca_scaling_info: + if vca_info["type"] == "delete" and not vca_info.get("osm_kdu_id"): + member_vnf_index = str(vca_info["member-vnf-index"]) + self.logger.debug( + logging_text + "vdu info: {}".format(vca_info) ) + if vca_info.get("osm_vdu_id"): + vdu_id = vca_info["osm_vdu_id"] + vdu_index = int(vca_info["vdu_index"]) + stage[ + 1 + ] = "Scaling member_vnf_index={}, vdu_id={}, vdu_index={} ".format( + member_vnf_index, vdu_id, vdu_index + ) + stage[2] = step = "Scaling in VCA" + self._write_op_status(op_id=nslcmop_id, stage=stage) vca_update = db_nsr["_admin"]["deployed"]["VCA"] config_update = db_nsr["configurationStatus"] for vca_index, vca in enumerate(vca_update): - if (vca or vca.get("ee_id")) and vca["member-vnf-index"] == member_vnf_index and \ - vca["vdu_count_index"] == vdu_index: + if ( + (vca or vca.get("ee_id")) + and vca["member-vnf-index"] == member_vnf_index + and vca["vdu_count_index"] == vdu_index + ): if vca.get("vdu_id"): - config_descriptor = get_configuration(db_vnfd, vca.get("vdu_id")) + config_descriptor = get_configuration( + db_vnfd, vca.get("vdu_id") + ) elif vca.get("kdu_name"): - config_descriptor = get_configuration(db_vnfd, vca.get("kdu_name")) + config_descriptor = get_configuration( + db_vnfd, vca.get("kdu_name") + ) else: - config_descriptor = get_configuration(db_vnfd, db_vnfd["id"]) - operation_params = db_nslcmop.get("operationParams") or {} - exec_terminate_primitives = (not operation_params.get("skip_terminate_primitives") and - vca.get("needed_terminate")) - task = asyncio.ensure_future(asyncio.wait_for( - self.destroy_N2VC(logging_text, db_nslcmop, vca, config_descriptor, - vca_index, destroy_ee=True, - exec_primitives=exec_terminate_primitives, - scaling_in=True), timeout=self.timeout_charm_delete)) - # wait before next removal - await asyncio.sleep(30) - tasks_dict_info[task] = "Terminating VCA {}".format(vca.get("ee_id")) + config_descriptor = get_configuration( + db_vnfd, db_vnfd["id"] + ) + operation_params = ( + db_nslcmop.get("operationParams") or {} + ) + exec_terminate_primitives = not operation_params.get( + "skip_terminate_primitives" + ) and vca.get("needed_terminate") + task = asyncio.ensure_future( + asyncio.wait_for( + self.destroy_N2VC( + logging_text, + db_nslcmop, + vca, + config_descriptor, + vca_index, + destroy_ee=True, + exec_primitives=exec_terminate_primitives, + scaling_in=True, + vca_id=vca_id, + ), + timeout=self.timeout.charm_delete, + ) + ) + tasks_dict_info[task] = "Terminating VCA {}".format( + vca.get("ee_id") + ) del vca_update[vca_index] del config_update[vca_index] # wait for pending tasks of terminate primitives if tasks_dict_info: - self.logger.debug(logging_text + - 'Waiting for tasks {}'.format(list(tasks_dict_info.keys()))) - error_list = await self._wait_for_tasks(logging_text, tasks_dict_info, - min(self.timeout_charm_delete, - self.timeout_ns_terminate), - stage, nslcmop_id) + self.logger.debug( + logging_text + + "Waiting for tasks {}".format( + list(tasks_dict_info.keys()) + ) + ) + error_list = await self._wait_for_tasks( + logging_text, + tasks_dict_info, + min( + self.timeout.charm_delete, self.timeout.ns_terminate + ), + stage, + nslcmop_id, + ) tasks_dict_info.clear() if error_list: raise LcmException("; ".join(error_list)) db_vca_and_config_update = { "_admin.deployed.VCA": vca_update, - "configurationStatus": config_update + "configurationStatus": config_update, } - self.update_db_2("nsrs", db_nsr["_id"], db_vca_and_config_update) + self.update_db_2( + "nsrs", db_nsr["_id"], db_vca_and_config_update + ) scale_process = None # SCALE-IN VCA - END # SCALE RO - BEGIN - if RO_scaling_info: + if scaling_info.get("vdu-create") or scaling_info.get("vdu-delete"): scale_process = "RO" - if self.ro_config.get("ng"): - await self._scale_ng_ro(logging_text, db_nsr, db_nslcmop, db_vnfr, vdu_scaling_info, stage) - vdu_scaling_info.pop("vdu-create", None) - vdu_scaling_info.pop("vdu-delete", None) + if self.ro_config.ng: + await self._scale_ng_ro( + logging_text, db_nsr, db_nslcmop, db_vnfr, scaling_info, stage + ) + scaling_info.pop("vdu-create", None) + scaling_info.pop("vdu-delete", None) + + scale_process = None + # SCALE RO - END + + # SCALE KDU - BEGIN + if scaling_info.get("kdu-create") or scaling_info.get("kdu-delete"): + scale_process = "KDU" + await self._scale_kdu( + logging_text, nsr_id, nsr_deployed, db_vnfd, vca_id, scaling_info + ) + scaling_info.pop("kdu-create", None) + scaling_info.pop("kdu-delete", None) scale_process = None + # SCALE KDU - END + if db_nsr_update: self.update_db_2("nsrs", nsr_id, db_nsr_update) - # SCALE RO - END # SCALE-UP VCA - BEGIN - if VCA_scaling_info: - step = db_nslcmop_update["detailed-status"] = \ - "Creating new execution environments" + if vca_scaling_info: + step = db_nslcmop_update[ + "detailed-status" + ] = "Creating new execution environments" scale_process = "VCA" - for vdu_info in VCA_scaling_info: - if vdu_info["type"] == "create": - member_vnf_index = str(vdu_info["member-vnf-index"]) - self.logger.debug(logging_text + "vdu info: {}".format(vdu_info)) + for vca_info in vca_scaling_info: + if vca_info["type"] == "create" and not vca_info.get("osm_kdu_id"): + member_vnf_index = str(vca_info["member-vnf-index"]) + self.logger.debug( + logging_text + "vdu info: {}".format(vca_info) + ) vnfd_id = db_vnfr["vnfd-ref"] - vdu_index = int(vdu_info["vdu_index"]) - deploy_params = {"OSM": get_osm_params(db_vnfr)} - if db_vnfr.get("additionalParamsForVnf"): - deploy_params.update(parse_yaml_strings(db_vnfr["additionalParamsForVnf"].copy())) - descriptor_config = get_configuration(db_vnfd, db_vnfd["id"]) - if descriptor_config: - vdu_id = None - vdu_name = None - kdu_name = None - self._deploy_n2vc( - logging_text=logging_text + "member_vnf_index={} ".format(member_vnf_index), - db_nsr=db_nsr, - db_vnfr=db_vnfr, - nslcmop_id=nslcmop_id, - nsr_id=nsr_id, - nsi_id=nsi_id, - vnfd_id=vnfd_id, - vdu_id=vdu_id, - kdu_name=kdu_name, - member_vnf_index=member_vnf_index, - vdu_index=vdu_index, - vdu_name=vdu_name, - deploy_params=deploy_params, - descriptor_config=descriptor_config, - base_folder=base_folder, - task_instantiation_info=tasks_dict_info, - stage=stage - ) - vdu_id = vdu_info["osm_vdu_id"] - vdur = find_in_list(db_vnfr["vdur"], lambda vdu: vdu["vdu-id-ref"] == vdu_id) - descriptor_config = get_configuration(db_vnfd, vdu_id) - if vdur.get("additionalParams"): - deploy_params_vdu = parse_yaml_strings(vdur["additionalParams"]) - else: - deploy_params_vdu = deploy_params - deploy_params_vdu["OSM"] = get_osm_params(db_vnfr, vdu_id, vdu_count_index=vdu_index) - if descriptor_config: - vdu_name = None - kdu_name = None - stage[1] = "Scaling member_vnf_index={}, vdu_id={}, vdu_index={} ".format( - member_vnf_index, vdu_id, vdu_index) - stage[2] = step = "Scaling out VCA" - self._write_op_status( - op_id=nslcmop_id, - stage=stage + if vca_info.get("osm_vdu_id"): + vdu_index = int(vca_info["vdu_index"]) + deploy_params = {"OSM": get_osm_params(db_vnfr)} + if db_vnfr.get("additionalParamsForVnf"): + deploy_params.update( + parse_yaml_strings( + db_vnfr["additionalParamsForVnf"].copy() + ) + ) + descriptor_config = get_configuration( + db_vnfd, db_vnfd["id"] ) - self._deploy_n2vc( - logging_text=logging_text + "member_vnf_index={}, vdu_id={}, vdu_index={} ".format( - member_vnf_index, vdu_id, vdu_index), - db_nsr=db_nsr, - db_vnfr=db_vnfr, - nslcmop_id=nslcmop_id, - nsr_id=nsr_id, - nsi_id=nsi_id, - vnfd_id=vnfd_id, - vdu_id=vdu_id, - kdu_name=kdu_name, - member_vnf_index=member_vnf_index, - vdu_index=vdu_index, - vdu_name=vdu_name, - deploy_params=deploy_params_vdu, - descriptor_config=descriptor_config, - base_folder=base_folder, - task_instantiation_info=tasks_dict_info, - stage=stage + if descriptor_config: + vdu_id = None + vdu_name = None + kdu_name = None + kdu_index = None + self._deploy_n2vc( + logging_text=logging_text + + "member_vnf_index={} ".format(member_vnf_index), + db_nsr=db_nsr, + db_vnfr=db_vnfr, + nslcmop_id=nslcmop_id, + nsr_id=nsr_id, + nsi_id=nsi_id, + vnfd_id=vnfd_id, + vdu_id=vdu_id, + kdu_name=kdu_name, + kdu_index=kdu_index, + member_vnf_index=member_vnf_index, + vdu_index=vdu_index, + vdu_name=vdu_name, + deploy_params=deploy_params, + descriptor_config=descriptor_config, + base_folder=base_folder, + task_instantiation_info=tasks_dict_info, + stage=stage, + ) + vdu_id = vca_info["osm_vdu_id"] + vdur = find_in_list( + db_vnfr["vdur"], lambda vdu: vdu["vdu-id-ref"] == vdu_id ) - # TODO: scaling for kdu is not implemented yet. - kdu_name = vdu_info["osm_vdu_id"] - descriptor_config = get_configuration(db_vnfd, kdu_name) - if descriptor_config: - vdu_id = None - vdu_index = vdu_index - vdu_name = None - kdur = next(x for x in db_vnfr["kdur"] if x["kdu-name"] == kdu_name) - deploy_params_kdu = {"OSM": get_osm_params(db_vnfr)} - if kdur.get("additionalParams"): - deploy_params_kdu = parse_yaml_strings(kdur["additionalParams"]) - - self._deploy_n2vc( - logging_text=logging_text, - db_nsr=db_nsr, - db_vnfr=db_vnfr, - nslcmop_id=nslcmop_id, - nsr_id=nsr_id, - nsi_id=nsi_id, - vnfd_id=vnfd_id, - vdu_id=vdu_id, - kdu_name=kdu_name, - member_vnf_index=member_vnf_index, - vdu_index=vdu_index, - vdu_name=vdu_name, - deploy_params=deploy_params_kdu, - descriptor_config=descriptor_config, - base_folder=base_folder, - task_instantiation_info=tasks_dict_info, - stage=stage + descriptor_config = get_configuration(db_vnfd, vdu_id) + if vdur.get("additionalParams"): + deploy_params_vdu = parse_yaml_strings( + vdur["additionalParams"] + ) + else: + deploy_params_vdu = deploy_params + deploy_params_vdu["OSM"] = get_osm_params( + db_vnfr, vdu_id, vdu_count_index=vdu_index ) + if descriptor_config: + vdu_name = None + kdu_name = None + kdu_index = None + stage[ + 1 + ] = "Scaling member_vnf_index={}, vdu_id={}, vdu_index={} ".format( + member_vnf_index, vdu_id, vdu_index + ) + stage[2] = step = "Scaling out VCA" + self._write_op_status(op_id=nslcmop_id, stage=stage) + self._deploy_n2vc( + logging_text=logging_text + + "member_vnf_index={}, vdu_id={}, vdu_index={} ".format( + member_vnf_index, vdu_id, vdu_index + ), + db_nsr=db_nsr, + db_vnfr=db_vnfr, + nslcmop_id=nslcmop_id, + nsr_id=nsr_id, + nsi_id=nsi_id, + vnfd_id=vnfd_id, + vdu_id=vdu_id, + kdu_name=kdu_name, + member_vnf_index=member_vnf_index, + vdu_index=vdu_index, + kdu_index=kdu_index, + vdu_name=vdu_name, + deploy_params=deploy_params_vdu, + descriptor_config=descriptor_config, + base_folder=base_folder, + task_instantiation_info=tasks_dict_info, + stage=stage, + ) # SCALE-UP VCA - END scale_process = None @@ -4174,14 +7023,26 @@ class NsLcm(LcmBase): # execute primitive service POST-SCALING step = "Executing post-scale vnf-config-primitive" if scaling_descriptor.get("scaling-config-action"): - for scaling_config_action in scaling_descriptor["scaling-config-action"]: - if (scaling_config_action.get("trigger") == "post-scale-in" and scaling_type == "SCALE_IN") \ - or (scaling_config_action.get("trigger") == "post-scale-out" and scaling_type == "SCALE_OUT"): - vnf_config_primitive = scaling_config_action["vnf-config-primitive-name-ref"] - step = db_nslcmop_update["detailed-status"] = \ - "executing post-scale scaling-config-action '{}'".format(vnf_config_primitive) - - vnfr_params = {"VDU_SCALE_INFO": vdu_scaling_info} + for scaling_config_action in scaling_descriptor[ + "scaling-config-action" + ]: + if ( + scaling_config_action.get("trigger") == "post-scale-in" + and scaling_type == "SCALE_IN" + ) or ( + scaling_config_action.get("trigger") == "post-scale-out" + and scaling_type == "SCALE_OUT" + ): + vnf_config_primitive = scaling_config_action[ + "vnf-config-primitive-name-ref" + ] + step = db_nslcmop_update[ + "detailed-status" + ] = "executing post-scale scaling-config-action '{}'".format( + vnf_config_primitive + ) + + vnfr_params = {"VDU_SCALE_INFO": scaling_info} if db_vnfr.get("additionalParamsForVnf"): vnfr_params.update(db_vnfr["additionalParamsForVnf"]) @@ -4195,51 +7056,92 @@ class NsLcm(LcmBase): raise LcmException( "Invalid vnfd descriptor at scaling-group-descriptor[name='{}']:scaling-config-" "action[vnf-config-primitive-name-ref='{}'] does not match any vnf-configuration:" - "config-primitive".format(scaling_group, vnf_config_primitive)) + "config-primitive".format( + scaling_group, vnf_config_primitive + ) + ) scale_process = "VCA" db_nsr_update["config-status"] = "configuring post-scaling" - primitive_params = self._map_primitive_params(config_primitive, {}, vnfr_params) + primitive_params = self._map_primitive_params( + config_primitive, {}, vnfr_params + ) # Post-scale retry check: Check if this sub-operation has been executed before op_index = self._check_or_add_scale_suboperation( - db_nslcmop, nslcmop_id, vnf_index, vnf_config_primitive, primitive_params, 'POST-SCALE') + db_nslcmop, + vnf_index, + vnf_config_primitive, + primitive_params, + "POST-SCALE", + ) if op_index == self.SUBOPERATION_STATUS_SKIP: # Skip sub-operation - result = 'COMPLETED' - result_detail = 'Done' - self.logger.debug(logging_text + - "vnf_config_primitive={} Skipped sub-operation, result {} {}". - format(vnf_config_primitive, result, result_detail)) + result = "COMPLETED" + result_detail = "Done" + self.logger.debug( + logging_text + + "vnf_config_primitive={} Skipped sub-operation, result {} {}".format( + vnf_config_primitive, result, result_detail + ) + ) else: if op_index == self.SUBOPERATION_STATUS_NEW: # New sub-operation: Get index of this sub-operation - op_index = len(db_nslcmop.get('_admin', {}).get('operations')) - 1 - self.logger.debug(logging_text + "vnf_config_primitive={} New sub-operation". - format(vnf_config_primitive)) + op_index = ( + len(db_nslcmop.get("_admin", {}).get("operations")) + - 1 + ) + self.logger.debug( + logging_text + + "vnf_config_primitive={} New sub-operation".format( + vnf_config_primitive + ) + ) else: # retry: Get registered params for this existing sub-operation - op = db_nslcmop.get('_admin', {}).get('operations', [])[op_index] - vnf_index = op.get('member_vnf_index') - vnf_config_primitive = op.get('primitive') - primitive_params = op.get('primitive_params') - self.logger.debug(logging_text + "vnf_config_primitive={} Sub-operation retry". - format(vnf_config_primitive)) + op = db_nslcmop.get("_admin", {}).get("operations", [])[ + op_index + ] + vnf_index = op.get("member_vnf_index") + vnf_config_primitive = op.get("primitive") + primitive_params = op.get("primitive_params") + self.logger.debug( + logging_text + + "vnf_config_primitive={} Sub-operation retry".format( + vnf_config_primitive + ) + ) # Execute the primitive, either with new (first-time) or registered (reintent) args - ee_descriptor_id = config_primitive.get("execution-environment-ref") - primitive_name = config_primitive.get("execution-environment-primitive", - vnf_config_primitive) - ee_id, vca_type = self._look_for_deployed_vca(nsr_deployed["VCA"], - member_vnf_index=vnf_index, - vdu_id=None, - vdu_count_index=None, - ee_descriptor_id=ee_descriptor_id) + ee_descriptor_id = config_primitive.get( + "execution-environment-ref" + ) + primitive_name = config_primitive.get( + "execution-environment-primitive", vnf_config_primitive + ) + ee_id, vca_type = self._look_for_deployed_vca( + nsr_deployed["VCA"], + member_vnf_index=vnf_index, + vdu_id=None, + vdu_count_index=None, + ee_descriptor_id=ee_descriptor_id, + ) result, result_detail = await self._ns_execute_primitive( - ee_id, primitive_name, primitive_params, vca_type=vca_type) - self.logger.debug(logging_text + "vnf_config_primitive={} Done with result {} {}".format( - vnf_config_primitive, result, result_detail)) + ee_id, + primitive_name, + primitive_params, + vca_type=vca_type, + vca_id=vca_id, + ) + self.logger.debug( + logging_text + + "vnf_config_primitive={} Done with result {} {}".format( + vnf_config_primitive, result, result_detail + ) + ) # Update operationState = COMPLETED | FAILED self._update_suboperation_status( - db_nslcmop, op_index, result, result_detail) + db_nslcmop, op_index, result, result_detail + ) if result == "FAILED": raise LcmException(result_detail) @@ -4247,29 +7149,57 @@ class NsLcm(LcmBase): scale_process = None # POST-SCALE END - db_nsr_update["detailed-status"] = "" # "scaled {} {}".format(scaling_group, scaling_type) - db_nsr_update["operational-status"] = "running" if old_operational_status == "failed" \ + db_nsr_update[ + "detailed-status" + ] = "" # "scaled {} {}".format(scaling_group, scaling_type) + db_nsr_update["operational-status"] = ( + "running" + if old_operational_status == "failed" else old_operational_status + ) db_nsr_update["config-status"] = old_config_status return - except (ROclient.ROClientException, DbException, LcmException, NgRoException) as e: + except ( + ROclient.ROClientException, + DbException, + LcmException, + NgRoException, + ) as e: self.logger.error(logging_text + "Exit Exception {}".format(e)) exc = e except asyncio.CancelledError: - self.logger.error(logging_text + "Cancelled Exception while '{}'".format(step)) + self.logger.error( + logging_text + "Cancelled Exception while '{}'".format(step) + ) exc = "Operation was cancelled" except Exception as e: exc = traceback.format_exc() - self.logger.critical(logging_text + "Exit Exception {} {}".format(type(e).__name__, e), exc_info=True) + self.logger.critical( + logging_text + "Exit Exception {} {}".format(type(e).__name__, e), + exc_info=True, + ) finally: - self._write_ns_status(nsr_id=nsr_id, ns_state=None, current_operation="IDLE", current_operation_id=None) + self._write_ns_status( + nsr_id=nsr_id, + ns_state=None, + current_operation="IDLE", + current_operation_id=None, + ) if tasks_dict_info: stage[1] = "Waiting for instantiate pending tasks." self.logger.debug(logging_text + stage[1]) - exc = await self._wait_for_tasks(logging_text, tasks_dict_info, self.timeout_ns_deploy, - stage, nslcmop_id, nsr_id=nsr_id) + exc = await self._wait_for_tasks( + logging_text, + tasks_dict_info, + self.timeout.ns_deploy, + stage, + nslcmop_id, + nsr_id=nsr_id, + ) if exc: - db_nslcmop_update["detailed-status"] = error_description_nslcmop = "FAILED {}: {}".format(step, exc) + db_nslcmop_update[ + "detailed-status" + ] = error_description_nslcmop = "FAILED {}: {}".format(step, exc) nslcmop_operation_state = "FAILED" if db_nsr: db_nsr_update["operational-status"] = old_operational_status @@ -4280,29 +7210,160 @@ class NsLcm(LcmBase): db_nsr_update["config-status"] = "failed" if "RO" in scale_process: db_nsr_update["operational-status"] = "failed" - db_nsr_update["detailed-status"] = "FAILED scaling nslcmop={} {}: {}".format(nslcmop_id, step, - exc) + db_nsr_update[ + "detailed-status" + ] = "FAILED scaling nslcmop={} {}: {}".format( + nslcmop_id, step, exc + ) else: error_description_nslcmop = None nslcmop_operation_state = "COMPLETED" db_nslcmop_update["detailed-status"] = "Done" - self._write_op_status(op_id=nslcmop_id, stage="", error_message=error_description_nslcmop, - operation_state=nslcmop_operation_state, other_update=db_nslcmop_update) + self._write_op_status( + op_id=nslcmop_id, + stage="", + error_message=error_description_nslcmop, + operation_state=nslcmop_operation_state, + other_update=db_nslcmop_update, + ) if db_nsr: - self._write_ns_status(nsr_id=nsr_id, ns_state=None, current_operation="IDLE", - current_operation_id=None, other_update=db_nsr_update) + self._write_ns_status( + nsr_id=nsr_id, + ns_state=None, + current_operation="IDLE", + current_operation_id=None, + other_update=db_nsr_update, + ) if nslcmop_operation_state: try: - msg = {"nsr_id": nsr_id, "nslcmop_id": nslcmop_id, "operationState": nslcmop_operation_state} - await self.msg.aiowrite("ns", "scaled", msg, loop=self.loop) + msg = { + "nsr_id": nsr_id, + "nslcmop_id": nslcmop_id, + "operationState": nslcmop_operation_state, + } + await self.msg.aiowrite("ns", "scaled", msg) except Exception as e: - self.logger.error(logging_text + "kafka_write notification Exception {}".format(e)) + self.logger.error( + logging_text + "kafka_write notification Exception {}".format(e) + ) self.logger.debug(logging_text + "Exit") self.lcm_tasks.remove("ns", nsr_id, nslcmop_id, "ns_scale") - async def _scale_ng_ro(self, logging_text, db_nsr, db_nslcmop, db_vnfr, vdu_scaling_info, stage): + async def _scale_kdu( + self, logging_text, nsr_id, nsr_deployed, db_vnfd, vca_id, scaling_info + ): + _scaling_info = scaling_info.get("kdu-create") or scaling_info.get("kdu-delete") + for kdu_name in _scaling_info: + for kdu_scaling_info in _scaling_info[kdu_name]: + deployed_kdu, index = get_deployed_kdu( + nsr_deployed, kdu_name, kdu_scaling_info["member-vnf-index"] + ) + cluster_uuid = deployed_kdu["k8scluster-uuid"] + kdu_instance = deployed_kdu["kdu-instance"] + kdu_model = deployed_kdu.get("kdu-model") + scale = int(kdu_scaling_info["scale"]) + k8s_cluster_type = kdu_scaling_info["k8s-cluster-type"] + + db_dict = { + "collection": "nsrs", + "filter": {"_id": nsr_id}, + "path": "_admin.deployed.K8s.{}".format(index), + } + + step = "scaling application {}".format( + kdu_scaling_info["resource-name"] + ) + self.logger.debug(logging_text + step) + + if kdu_scaling_info["type"] == "delete": + kdu_config = get_configuration(db_vnfd, kdu_name) + if ( + kdu_config + and kdu_config.get("terminate-config-primitive") + and get_juju_ee_ref(db_vnfd, kdu_name) is None + ): + terminate_config_primitive_list = kdu_config.get( + "terminate-config-primitive" + ) + terminate_config_primitive_list.sort( + key=lambda val: int(val["seq"]) + ) + + for ( + terminate_config_primitive + ) in terminate_config_primitive_list: + primitive_params_ = self._map_primitive_params( + terminate_config_primitive, {}, {} + ) + step = "execute terminate config primitive" + self.logger.debug(logging_text + step) + await asyncio.wait_for( + self.k8scluster_map[k8s_cluster_type].exec_primitive( + cluster_uuid=cluster_uuid, + kdu_instance=kdu_instance, + primitive_name=terminate_config_primitive["name"], + params=primitive_params_, + db_dict=db_dict, + total_timeout=self.timeout.primitive, + vca_id=vca_id, + ), + timeout=self.timeout.primitive + * self.timeout.primitive_outer_factor, + ) + + await asyncio.wait_for( + self.k8scluster_map[k8s_cluster_type].scale( + kdu_instance=kdu_instance, + scale=scale, + resource_name=kdu_scaling_info["resource-name"], + total_timeout=self.timeout.scale_on_error, + vca_id=vca_id, + cluster_uuid=cluster_uuid, + kdu_model=kdu_model, + atomic=True, + db_dict=db_dict, + ), + timeout=self.timeout.scale_on_error + * self.timeout.scale_on_error_outer_factor, + ) + + if kdu_scaling_info["type"] == "create": + kdu_config = get_configuration(db_vnfd, kdu_name) + if ( + kdu_config + and kdu_config.get("initial-config-primitive") + and get_juju_ee_ref(db_vnfd, kdu_name) is None + ): + initial_config_primitive_list = kdu_config.get( + "initial-config-primitive" + ) + initial_config_primitive_list.sort( + key=lambda val: int(val["seq"]) + ) + + for initial_config_primitive in initial_config_primitive_list: + primitive_params_ = self._map_primitive_params( + initial_config_primitive, {}, {} + ) + step = "execute initial config primitive" + self.logger.debug(logging_text + step) + await asyncio.wait_for( + self.k8scluster_map[k8s_cluster_type].exec_primitive( + cluster_uuid=cluster_uuid, + kdu_instance=kdu_instance, + primitive_name=initial_config_primitive["name"], + params=primitive_params_, + db_dict=db_dict, + vca_id=vca_id, + ), + timeout=600, + ) + + async def _scale_ng_ro( + self, logging_text, db_nsr, db_nslcmop, db_vnfr, vdu_scaling_info, stage + ): nsr_id = db_nslcmop["nsInstanceId"] db_nsd = self.db.get_one("nsds", {"_id": db_nsr["nsd-id"]}) db_vnfrs = {} @@ -4321,66 +7382,1430 @@ class NsLcm(LcmBase): db_vnfds.append(vnfd) n2vc_key = self.n2vc.get_public_key() n2vc_key_list = [n2vc_key] - self.scale_vnfr(db_vnfr, vdu_scaling_info.get("vdu-create"), vdu_scaling_info.get("vdu-delete"), - mark_delete=True) + self.scale_vnfr( + db_vnfr, + vdu_scaling_info.get("vdu-create"), + vdu_scaling_info.get("vdu-delete"), + mark_delete=True, + ) # db_vnfr has been updated, update db_vnfrs to use it db_vnfrs[db_vnfr["member-vnf-index-ref"]] = db_vnfr - await self._instantiate_ng_ro(logging_text, nsr_id, db_nsd, db_nsr, db_nslcmop, db_vnfrs, - db_vnfds, n2vc_key_list, stage=stage, start_deploy=time(), - timeout_ns_deploy=self.timeout_ns_deploy) + await self._instantiate_ng_ro( + logging_text, + nsr_id, + db_nsd, + db_nsr, + db_nslcmop, + db_vnfrs, + db_vnfds, + n2vc_key_list, + stage=stage, + start_deploy=time(), + timeout_ns_deploy=self.timeout.ns_deploy, + ) if vdu_scaling_info.get("vdu-delete"): - self.scale_vnfr(db_vnfr, None, vdu_scaling_info["vdu-delete"], mark_delete=False) + self.scale_vnfr( + db_vnfr, None, vdu_scaling_info["vdu-delete"], mark_delete=False + ) + + async def extract_prometheus_scrape_jobs( + self, + ee_id: str, + artifact_path: str, + ee_config_descriptor: dict, + vnfr_id: str, + nsr_id: str, + target_ip: str, + element_type: str, + vnf_member_index: str = "", + vdu_id: str = "", + vdu_index: int = None, + kdu_name: str = "", + kdu_index: int = None, + ) -> dict: + """Method to extract prometheus scrape jobs from EE's Prometheus template job file + This method will wait until the corresponding VDU or KDU is fully instantiated + + Args: + ee_id (str): Execution Environment ID + artifact_path (str): Path where the EE's content is (including the Prometheus template file) + ee_config_descriptor (dict): Execution Environment's configuration descriptor + vnfr_id (str): VNFR ID where this EE applies + nsr_id (str): NSR ID where this EE applies + target_ip (str): VDU/KDU instance IP address + element_type (str): NS or VNF or VDU or KDU + vnf_member_index (str, optional): VNF index where this EE applies. Defaults to "". + vdu_id (str, optional): VDU ID where this EE applies. Defaults to "". + vdu_index (int, optional): VDU index where this EE applies. Defaults to None. + kdu_name (str, optional): KDU name where this EE applies. Defaults to "". + kdu_index (int, optional): KDU index where this EE applies. Defaults to None. + + Raises: + LcmException: When the VDU or KDU instance was not found in an hour + + Returns: + _type_: Prometheus jobs + """ + # default the vdur and kdur names to an empty string, to avoid any later + # problem with Prometheus when the element type is not VDU or KDU + vdur_name = "" + kdur_name = "" - async def add_prometheus_metrics(self, ee_id, artifact_path, ee_config_descriptor, vnfr_id, nsr_id, target_ip): - if not self.prometheus: - return # look if exist a file called 'prometheus*.j2' and artifact_content = self.fs.dir_ls(artifact_path) - job_file = next((f for f in artifact_content if f.startswith("prometheus") and f.endswith(".j2")), None) + job_file = next( + ( + f + for f in artifact_content + if f.startswith("prometheus") and f.endswith(".j2") + ), + None, + ) if not job_file: return + self.logger.debug("Artifact path{}".format(artifact_path)) + self.logger.debug("job file{}".format(job_file)) with self.fs.file_open((artifact_path, job_file), "r") as f: job_data = f.read() - # TODO get_service - _, _, service = ee_id.partition(".") # remove prefix "namespace." - host_name = "{}-{}".format(service, ee_config_descriptor["metric-service"]) - host_port = "80" - vnfr_id = vnfr_id.replace("-", "") - variables = { - "JOB_NAME": vnfr_id, - "TARGET_IP": target_ip, - "EXPORTER_POD_IP": host_name, - "EXPORTER_POD_PORT": host_port, - } - job_list = self.prometheus.parse_job(job_data, variables) + # obtain the VDUR or KDUR, if the element type is VDU or KDU + if element_type in ("VDU", "KDU"): + for _ in range(360): + db_vnfr = self.db.get_one("vnfrs", {"_id": vnfr_id}) + if vdu_id and vdu_index is not None: + vdur = next( + ( + x + for x in get_iterable(db_vnfr, "vdur") + if ( + x.get("vdu-id-ref") == vdu_id + and x.get("count-index") == vdu_index + ) + ), + {}, + ) + if vdur.get("name"): + vdur_name = vdur.get("name") + break + if kdu_name and kdu_index is not None: + kdur = next( + ( + x + for x in get_iterable(db_vnfr, "kdur") + if ( + x.get("kdu-name") == kdu_name + and x.get("count-index") == kdu_index + ) + ), + {}, + ) + if kdur.get("name"): + kdur_name = kdur.get("name") + break + + await asyncio.sleep(10) + else: + if vdu_id and vdu_index is not None: + raise LcmException( + f"Timeout waiting VDU with name={vdu_id} and index={vdu_index} to be intantiated" + ) + if kdu_name and kdu_index is not None: + raise LcmException( + f"Timeout waiting KDU with name={kdu_name} and index={kdu_index} to be intantiated" + ) + + if ee_id is not None: + _, namespace, helm_id = get_ee_id_parts( + ee_id + ) # get namespace and EE gRPC service name + host_name = f'{helm_id}-{ee_config_descriptor["metric-service"]}.{namespace}.svc' # svc_name.namespace.svc + host_port = "80" + vnfr_id = vnfr_id.replace("-", "") + variables = { + "JOB_NAME": vnfr_id, + "TARGET_IP": target_ip, + "EXPORTER_POD_IP": host_name, + "EXPORTER_POD_PORT": host_port, + "NSR_ID": nsr_id, + "VNF_MEMBER_INDEX": vnf_member_index, + "VDUR_NAME": vdur_name, + "KDUR_NAME": kdur_name, + "ELEMENT_TYPE": element_type, + } + else: + metric_path = ee_config_descriptor["metric-path"] + target_port = ee_config_descriptor["metric-port"] + vnfr_id = vnfr_id.replace("-", "") + variables = { + "JOB_NAME": vnfr_id, + "TARGET_IP": target_ip, + "TARGET_PORT": target_port, + "METRIC_PATH": metric_path, + } + + job_list = parse_job(job_data, variables) # ensure job_name is using the vnfr_id. Adding the metadata nsr_id for job in job_list: - if not isinstance(job.get("job_name"), str) or vnfr_id not in job["job_name"]: - job["job_name"] = vnfr_id + "_" + str(randint(1, 10000)) + if ( + not isinstance(job.get("job_name"), str) + or vnfr_id not in job["job_name"] + ): + job["job_name"] = vnfr_id + "_" + str(SystemRandom().randint(1, 10000)) job["nsr_id"] = nsr_id - job_dict = {jl["job_name"]: jl for jl in job_list} - if await self.prometheus.update(job_dict): - return list(job_dict.keys()) + job["vnfr_id"] = vnfr_id + return job_list + + async def rebuild_start_stop( + self, nsr_id, nslcmop_id, vnf_id, additional_param, operation_type + ): + logging_text = "Task ns={} {}={} ".format(nsr_id, operation_type, nslcmop_id) + self.logger.info(logging_text + "Enter") + stage = ["Preparing the environment", ""] + # database nsrs record + db_nsr_update = {} + vdu_vim_name = None + vim_vm_id = None + # in case of error, indicates what part of scale was failed to put nsr at error status + start_deploy = time() + try: + db_vnfr = self.db.get_one("vnfrs", {"_id": vnf_id}) + vim_account_id = db_vnfr.get("vim-account-id") + vim_info_key = "vim:" + vim_account_id + vdu_id = additional_param["vdu_id"] + vdurs = [item for item in db_vnfr["vdur"] if item["vdu-id-ref"] == vdu_id] + vdur = find_in_list( + vdurs, lambda vdu: vdu["count-index"] == additional_param["count-index"] + ) + if vdur: + vdu_vim_name = vdur["name"] + vim_vm_id = vdur["vim_info"][vim_info_key]["vim_id"] + target_vim, _ = next(k_v for k_v in vdur["vim_info"].items()) + else: + raise LcmException("Target vdu is not found") + self.logger.info("vdu_vim_name >> {} ".format(vdu_vim_name)) + # wait for any previous tasks in process + stage[1] = "Waiting for previous operations to terminate" + self.logger.info(stage[1]) + await self.lcm_tasks.waitfor_related_HA("ns", "nslcmops", nslcmop_id) - def get_vca_cloud_and_credentials(self, vim_account_id: str) -> (str, str): - """ - Get VCA Cloud and VCA Cloud Credentials for the VIM account + stage[1] = "Reading from database." + self.logger.info(stage[1]) + self._write_ns_status( + nsr_id=nsr_id, + ns_state=None, + current_operation=operation_type.upper(), + current_operation_id=nslcmop_id, + ) + self._write_op_status(op_id=nslcmop_id, stage=stage, queuePosition=0) - :param: vim_account_id: VIM Account ID + # read from db: ns + stage[1] = "Getting nsr={} from db.".format(nsr_id) + db_nsr_update["operational-status"] = operation_type + self.update_db_2("nsrs", nsr_id, db_nsr_update) + # Payload for RO + desc = { + operation_type: { + "vim_vm_id": vim_vm_id, + "vnf_id": vnf_id, + "vdu_index": additional_param["count-index"], + "vdu_id": vdur["id"], + "target_vim": target_vim, + "vim_account_id": vim_account_id, + } + } + stage[1] = "Sending rebuild request to RO... {}".format(desc) + self._write_op_status(op_id=nslcmop_id, stage=stage, queuePosition=0) + self.logger.info("ro nsr id: {}".format(nsr_id)) + result_dict = await self.RO.operate(nsr_id, desc, operation_type) + self.logger.info("response from RO: {}".format(result_dict)) + action_id = result_dict["action_id"] + await self._wait_ng_ro( + nsr_id, + action_id, + nslcmop_id, + start_deploy, + self.timeout.operate, + None, + "start_stop_rebuild", + ) + return "COMPLETED", "Done" + except (ROclient.ROClientException, DbException, LcmException) as e: + self.logger.error("Exit Exception {}".format(e)) + exc = e + except asyncio.CancelledError: + self.logger.error("Cancelled Exception while '{}'".format(stage)) + exc = "Operation was cancelled" + except Exception as e: + exc = traceback.format_exc() + self.logger.critical( + "Exit Exception {} {}".format(type(e).__name__, e), exc_info=True + ) + return "FAILED", "Error in operate VNF {}".format(exc) - :return: (cloud_name, cloud_credential) + async def migrate(self, nsr_id, nslcmop_id): """ - config = VimAccountDB.get_vim_account_with_id(vim_account_id).get("config", {}) - return config.get("vca_cloud"), config.get("vca_cloud_credential") + Migrate VNFs and VDUs instances in a NS + + :param: nsr_id: NS Instance ID + :param: nslcmop_id: nslcmop ID of migrate - def get_vca_k8s_cloud_and_credentials(self, vim_account_id: str) -> (str, str): """ - Get VCA K8s Cloud and VCA K8s Cloud Credentials for the VIM account + # Try to lock HA task here + task_is_locked_by_me = self.lcm_tasks.lock_HA("ns", "nslcmops", nslcmop_id) + if not task_is_locked_by_me: + return + logging_text = "Task ns={} migrate ".format(nsr_id) + self.logger.debug(logging_text + "Enter") + # get all needed from database + db_nslcmop = None + db_nslcmop_update = {} + nslcmop_operation_state = None + db_nsr_update = {} + target = {} + exc = None + # in case of error, indicates what part of scale was failed to put nsr at error status + start_deploy = time() - :param: vim_account_id: VIM Account ID + try: + # wait for any previous tasks in process + step = "Waiting for previous operations to terminate" + await self.lcm_tasks.waitfor_related_HA("ns", "nslcmops", nslcmop_id) - :return: (cloud_name, cloud_credential) - """ - config = VimAccountDB.get_vim_account_with_id(vim_account_id).get("config", {}) - return config.get("vca_k8s_cloud"), config.get("vca_k8s_cloud_credential") + self._write_ns_status( + nsr_id=nsr_id, + ns_state=None, + current_operation="MIGRATING", + current_operation_id=nslcmop_id, + ) + step = "Getting nslcmop from database" + self.logger.debug( + step + " after having waited for previous tasks to be completed" + ) + db_nslcmop = self.db.get_one("nslcmops", {"_id": nslcmop_id}) + migrate_params = db_nslcmop.get("operationParams") + + target = {} + target.update(migrate_params) + desc = await self.RO.migrate(nsr_id, target) + self.logger.debug("RO return > {}".format(desc)) + action_id = desc["action_id"] + await self._wait_ng_ro( + nsr_id, + action_id, + nslcmop_id, + start_deploy, + self.timeout.migrate, + operation="migrate", + ) + except (ROclient.ROClientException, DbException, LcmException) as e: + self.logger.error("Exit Exception {}".format(e)) + exc = e + except asyncio.CancelledError: + self.logger.error("Cancelled Exception while '{}'".format(step)) + exc = "Operation was cancelled" + except Exception as e: + exc = traceback.format_exc() + self.logger.critical( + "Exit Exception {} {}".format(type(e).__name__, e), exc_info=True + ) + finally: + self._write_ns_status( + nsr_id=nsr_id, + ns_state=None, + current_operation="IDLE", + current_operation_id=None, + ) + if exc: + db_nslcmop_update["detailed-status"] = "FAILED {}: {}".format(step, exc) + nslcmop_operation_state = "FAILED" + else: + nslcmop_operation_state = "COMPLETED" + db_nslcmop_update["detailed-status"] = "Done" + db_nsr_update["detailed-status"] = "Done" + + self._write_op_status( + op_id=nslcmop_id, + stage="", + error_message="", + operation_state=nslcmop_operation_state, + other_update=db_nslcmop_update, + ) + if nslcmop_operation_state: + try: + msg = { + "nsr_id": nsr_id, + "nslcmop_id": nslcmop_id, + "operationState": nslcmop_operation_state, + } + await self.msg.aiowrite("ns", "migrated", msg) + except Exception as e: + self.logger.error( + logging_text + "kafka_write notification Exception {}".format(e) + ) + self.logger.debug(logging_text + "Exit") + self.lcm_tasks.remove("ns", nsr_id, nslcmop_id, "ns_migrate") + + async def heal(self, nsr_id, nslcmop_id): + """ + Heal NS + + :param nsr_id: ns instance to heal + :param nslcmop_id: operation to run + :return: + """ + + # Try to lock HA task here + task_is_locked_by_me = self.lcm_tasks.lock_HA("ns", "nslcmops", nslcmop_id) + if not task_is_locked_by_me: + return + + logging_text = "Task ns={} heal={} ".format(nsr_id, nslcmop_id) + stage = ["", "", ""] + tasks_dict_info = {} + # ^ stage, step, VIM progress + self.logger.debug(logging_text + "Enter") + # get all needed from database + db_nsr = None + db_nslcmop_update = {} + db_nsr_update = {} + db_vnfrs = {} # vnf's info indexed by _id + exc = None + old_operational_status = "" + old_config_status = "" + nsi_id = None + try: + # wait for any previous tasks in process + step = "Waiting for previous operations to terminate" + await self.lcm_tasks.waitfor_related_HA("ns", "nslcmops", nslcmop_id) + self._write_ns_status( + nsr_id=nsr_id, + ns_state=None, + current_operation="HEALING", + current_operation_id=nslcmop_id, + ) + + step = "Getting nslcmop from database" + self.logger.debug( + step + " after having waited for previous tasks to be completed" + ) + db_nslcmop = self.db.get_one("nslcmops", {"_id": nslcmop_id}) + + step = "Getting nsr from database" + db_nsr = self.db.get_one("nsrs", {"_id": nsr_id}) + old_operational_status = db_nsr["operational-status"] + old_config_status = db_nsr["config-status"] + + db_nsr_update = { + "_admin.deployed.RO.operational-status": "healing", + } + self.update_db_2("nsrs", nsr_id, db_nsr_update) + + step = "Sending heal order to VIM" + await self.heal_RO( + logging_text=logging_text, + nsr_id=nsr_id, + db_nslcmop=db_nslcmop, + stage=stage, + ) + # VCA tasks + # read from db: nsd + stage[1] = "Getting nsd={} from db.".format(db_nsr["nsd-id"]) + self.logger.debug(logging_text + stage[1]) + nsd = self.db.get_one("nsds", {"_id": db_nsr["nsd-id"]}) + self.fs.sync(db_nsr["nsd-id"]) + db_nsr["nsd"] = nsd + # read from db: vnfr's of this ns + step = "Getting vnfrs from db" + db_vnfrs_list = self.db.get_list("vnfrs", {"nsr-id-ref": nsr_id}) + for vnfr in db_vnfrs_list: + db_vnfrs[vnfr["_id"]] = vnfr + self.logger.debug("ns.heal db_vnfrs={}".format(db_vnfrs)) + + # Check for each target VNF + target_list = db_nslcmop.get("operationParams", {}).get("healVnfData", {}) + for target_vnf in target_list: + # Find this VNF in the list from DB + vnfr_id = target_vnf.get("vnfInstanceId", None) + if vnfr_id: + db_vnfr = db_vnfrs[vnfr_id] + vnfd_id = db_vnfr.get("vnfd-id") + vnfd_ref = db_vnfr.get("vnfd-ref") + vnfd = self.db.get_one("vnfds", {"_id": vnfd_id}) + base_folder = vnfd["_admin"]["storage"] + vdu_id = None + vdu_index = 0 + vdu_name = None + kdu_name = None + nsi_id = None # TODO put nsi_id when this nsr belongs to a NSI + member_vnf_index = db_vnfr.get("member-vnf-index-ref") + + # Check each target VDU and deploy N2VC + target_vdu_list = target_vnf.get("additionalParams", {}).get( + "vdu", [] + ) + if not target_vdu_list: + # Codigo nuevo para crear diccionario + target_vdu_list = [] + for existing_vdu in db_vnfr.get("vdur"): + vdu_name = existing_vdu.get("vdu-name", None) + vdu_index = existing_vdu.get("count-index", 0) + vdu_run_day1 = target_vnf.get("additionalParams", {}).get( + "run-day1", False + ) + vdu_to_be_healed = { + "vdu-id": vdu_name, + "count-index": vdu_index, + "run-day1": vdu_run_day1, + } + target_vdu_list.append(vdu_to_be_healed) + for target_vdu in target_vdu_list: + deploy_params_vdu = target_vdu + # Set run-day1 vnf level value if not vdu level value exists + if not deploy_params_vdu.get("run-day1") and target_vnf.get( + "additionalParams", {} + ).get("run-day1"): + deploy_params_vdu["run-day1"] = target_vnf[ + "additionalParams" + ].get("run-day1") + vdu_name = target_vdu.get("vdu-id", None) + # TODO: Get vdu_id from vdud. + vdu_id = vdu_name + # For multi instance VDU count-index is mandatory + # For single session VDU count-indes is 0 + vdu_index = target_vdu.get("count-index", 0) + + # n2vc_redesign STEP 3 to 6 Deploy N2VC + stage[1] = "Deploying Execution Environments." + self.logger.debug(logging_text + stage[1]) + + # VNF Level charm. Normal case when proxy charms. + # If target instance is management machine continue with actions: recreate EE for native charms or reinject juju key for proxy charms. + descriptor_config = get_configuration(vnfd, vnfd_ref) + if descriptor_config: + # Continue if healed machine is management machine + vnf_ip_address = db_vnfr.get("ip-address") + target_instance = None + for instance in db_vnfr.get("vdur", None): + if ( + instance["vdu-name"] == vdu_name + and instance["count-index"] == vdu_index + ): + target_instance = instance + break + if vnf_ip_address == target_instance.get("ip-address"): + self._heal_n2vc( + logging_text=logging_text + + "member_vnf_index={}, vdu_name={}, vdu_index={} ".format( + member_vnf_index, vdu_name, vdu_index + ), + db_nsr=db_nsr, + db_vnfr=db_vnfr, + nslcmop_id=nslcmop_id, + nsr_id=nsr_id, + nsi_id=nsi_id, + vnfd_id=vnfd_ref, + vdu_id=None, + kdu_name=None, + member_vnf_index=member_vnf_index, + vdu_index=0, + vdu_name=None, + deploy_params=deploy_params_vdu, + descriptor_config=descriptor_config, + base_folder=base_folder, + task_instantiation_info=tasks_dict_info, + stage=stage, + ) + + # VDU Level charm. Normal case with native charms. + descriptor_config = get_configuration(vnfd, vdu_name) + if descriptor_config: + self._heal_n2vc( + logging_text=logging_text + + "member_vnf_index={}, vdu_name={}, vdu_index={} ".format( + member_vnf_index, vdu_name, vdu_index + ), + db_nsr=db_nsr, + db_vnfr=db_vnfr, + nslcmop_id=nslcmop_id, + nsr_id=nsr_id, + nsi_id=nsi_id, + vnfd_id=vnfd_ref, + vdu_id=vdu_id, + kdu_name=kdu_name, + member_vnf_index=member_vnf_index, + vdu_index=vdu_index, + vdu_name=vdu_name, + deploy_params=deploy_params_vdu, + descriptor_config=descriptor_config, + base_folder=base_folder, + task_instantiation_info=tasks_dict_info, + stage=stage, + ) + + except ( + ROclient.ROClientException, + DbException, + LcmException, + NgRoException, + ) as e: + self.logger.error(logging_text + "Exit Exception {}".format(e)) + exc = e + except asyncio.CancelledError: + self.logger.error( + logging_text + "Cancelled Exception while '{}'".format(step) + ) + exc = "Operation was cancelled" + except Exception as e: + exc = traceback.format_exc() + self.logger.critical( + logging_text + "Exit Exception {} {}".format(type(e).__name__, e), + exc_info=True, + ) + finally: + if tasks_dict_info: + stage[1] = "Waiting for healing pending tasks." + self.logger.debug(logging_text + stage[1]) + exc = await self._wait_for_tasks( + logging_text, + tasks_dict_info, + self.timeout.ns_deploy, + stage, + nslcmop_id, + nsr_id=nsr_id, + ) + if exc: + db_nslcmop_update[ + "detailed-status" + ] = error_description_nslcmop = "FAILED {}: {}".format(step, exc) + nslcmop_operation_state = "FAILED" + if db_nsr: + db_nsr_update["operational-status"] = old_operational_status + db_nsr_update["config-status"] = old_config_status + db_nsr_update[ + "detailed-status" + ] = "FAILED healing nslcmop={} {}: {}".format(nslcmop_id, step, exc) + for task, task_name in tasks_dict_info.items(): + if not task.done() or task.cancelled() or task.exception(): + if task_name.startswith(self.task_name_deploy_vca): + # A N2VC task is pending + db_nsr_update["config-status"] = "failed" + else: + # RO task is pending + db_nsr_update["operational-status"] = "failed" + else: + error_description_nslcmop = None + nslcmop_operation_state = "COMPLETED" + db_nslcmop_update["detailed-status"] = "Done" + db_nsr_update["detailed-status"] = "Done" + db_nsr_update["operational-status"] = "running" + db_nsr_update["config-status"] = "configured" + + self._write_op_status( + op_id=nslcmop_id, + stage="", + error_message=error_description_nslcmop, + operation_state=nslcmop_operation_state, + other_update=db_nslcmop_update, + ) + if db_nsr: + self._write_ns_status( + nsr_id=nsr_id, + ns_state=None, + current_operation="IDLE", + current_operation_id=None, + other_update=db_nsr_update, + ) + + if nslcmop_operation_state: + try: + msg = { + "nsr_id": nsr_id, + "nslcmop_id": nslcmop_id, + "operationState": nslcmop_operation_state, + } + await self.msg.aiowrite("ns", "healed", msg) + except Exception as e: + self.logger.error( + logging_text + "kafka_write notification Exception {}".format(e) + ) + self.logger.debug(logging_text + "Exit") + self.lcm_tasks.remove("ns", nsr_id, nslcmop_id, "ns_heal") + + async def heal_RO( + self, + logging_text, + nsr_id, + db_nslcmop, + stage, + ): + """ + Heal at RO + :param logging_text: preffix text to use at logging + :param nsr_id: nsr identity + :param db_nslcmop: database content of ns operation, in this case, 'instantiate' + :param stage: list with 3 items: [general stage, tasks, vim_specific]. This task will write over vim_specific + :return: None or exception + """ + + def get_vim_account(vim_account_id): + nonlocal db_vims + if vim_account_id in db_vims: + return db_vims[vim_account_id] + db_vim = self.db.get_one("vim_accounts", {"_id": vim_account_id}) + db_vims[vim_account_id] = db_vim + return db_vim + + try: + start_heal = time() + ns_params = db_nslcmop.get("operationParams") + if ns_params and ns_params.get("timeout_ns_heal"): + timeout_ns_heal = ns_params["timeout_ns_heal"] + else: + timeout_ns_heal = self.timeout.ns_heal + + db_vims = {} + + nslcmop_id = db_nslcmop["_id"] + target = { + "action_id": nslcmop_id, + } + self.logger.warning( + "db_nslcmop={} and timeout_ns_heal={}".format( + db_nslcmop, timeout_ns_heal + ) + ) + target.update(db_nslcmop.get("operationParams", {})) + + self.logger.debug("Send to RO > nsr_id={} target={}".format(nsr_id, target)) + desc = await self.RO.recreate(nsr_id, target) + self.logger.debug("RO return > {}".format(desc)) + action_id = desc["action_id"] + # waits for RO to complete because Reinjecting juju key at ro can find VM in state Deleted + await self._wait_ng_ro( + nsr_id, + action_id, + nslcmop_id, + start_heal, + timeout_ns_heal, + stage, + operation="healing", + ) + + # Updating NSR + db_nsr_update = { + "_admin.deployed.RO.operational-status": "running", + "detailed-status": " ".join(stage), + } + self.update_db_2("nsrs", nsr_id, db_nsr_update) + self._write_op_status(nslcmop_id, stage) + self.logger.debug( + logging_text + "ns healed at RO. RO_id={}".format(action_id) + ) + + except Exception as e: + stage[2] = "ERROR healing at VIM" + # self.set_vnfr_at_error(db_vnfrs, str(e)) + self.logger.error( + "Error healing at VIM {}".format(e), + exc_info=not isinstance( + e, + ( + ROclient.ROClientException, + LcmException, + DbException, + NgRoException, + ), + ), + ) + raise + + def _heal_n2vc( + self, + logging_text, + db_nsr, + db_vnfr, + nslcmop_id, + nsr_id, + nsi_id, + vnfd_id, + vdu_id, + kdu_name, + member_vnf_index, + vdu_index, + vdu_name, + deploy_params, + descriptor_config, + base_folder, + task_instantiation_info, + stage, + ): + # launch instantiate_N2VC in a asyncio task and register task object + # Look where information of this charm is at database ._admin.deployed.VCA + # if not found, create one entry and update database + # fill db_nsr._admin.deployed.VCA. + + self.logger.debug( + logging_text + "_deploy_n2vc vnfd_id={}, vdu_id={}".format(vnfd_id, vdu_id) + ) + + charm_name = "" + get_charm_name = False + if "execution-environment-list" in descriptor_config: + ee_list = descriptor_config.get("execution-environment-list", []) + elif "juju" in descriptor_config: + ee_list = [descriptor_config] # ns charms + if "execution-environment-list" not in descriptor_config: + # charm name is only required for ns charms + get_charm_name = True + else: # other types as script are not supported + ee_list = [] + + for ee_item in ee_list: + self.logger.debug( + logging_text + + "_deploy_n2vc ee_item juju={}, helm={}".format( + ee_item.get("juju"), ee_item.get("helm-chart") + ) + ) + ee_descriptor_id = ee_item.get("id") + if ee_item.get("juju"): + vca_name = ee_item["juju"].get("charm") + if get_charm_name: + charm_name = self.find_charm_name(db_nsr, str(vca_name)) + vca_type = ( + "lxc_proxy_charm" + if ee_item["juju"].get("charm") is not None + else "native_charm" + ) + if ee_item["juju"].get("cloud") == "k8s": + vca_type = "k8s_proxy_charm" + elif ee_item["juju"].get("proxy") is False: + vca_type = "native_charm" + elif ee_item.get("helm-chart"): + vca_name = ee_item["helm-chart"] + if ee_item.get("helm-version") and ee_item.get("helm-version") == "v2": + vca_type = "helm" + else: + vca_type = "helm-v3" + else: + self.logger.debug( + logging_text + "skipping non juju neither charm configuration" + ) + continue + + vca_index = -1 + for vca_index, vca_deployed in enumerate( + db_nsr["_admin"]["deployed"]["VCA"] + ): + if not vca_deployed: + continue + if ( + vca_deployed.get("member-vnf-index") == member_vnf_index + and vca_deployed.get("vdu_id") == vdu_id + and vca_deployed.get("kdu_name") == kdu_name + and vca_deployed.get("vdu_count_index", 0) == vdu_index + and vca_deployed.get("ee_descriptor_id") == ee_descriptor_id + ): + break + else: + # not found, create one. + target = ( + "ns" if not member_vnf_index else "vnf/{}".format(member_vnf_index) + ) + if vdu_id: + target += "/vdu/{}/{}".format(vdu_id, vdu_index or 0) + elif kdu_name: + target += "/kdu/{}".format(kdu_name) + vca_deployed = { + "target_element": target, + # ^ target_element will replace member-vnf-index, kdu_name, vdu_id ... in a single string + "member-vnf-index": member_vnf_index, + "vdu_id": vdu_id, + "kdu_name": kdu_name, + "vdu_count_index": vdu_index, + "operational-status": "init", # TODO revise + "detailed-status": "", # TODO revise + "step": "initial-deploy", # TODO revise + "vnfd_id": vnfd_id, + "vdu_name": vdu_name, + "type": vca_type, + "ee_descriptor_id": ee_descriptor_id, + "charm_name": charm_name, + } + vca_index += 1 + + # create VCA and configurationStatus in db + db_dict = { + "_admin.deployed.VCA.{}".format(vca_index): vca_deployed, + "configurationStatus.{}".format(vca_index): dict(), + } + self.update_db_2("nsrs", nsr_id, db_dict) + + db_nsr["_admin"]["deployed"]["VCA"].append(vca_deployed) + + self.logger.debug("N2VC > NSR_ID > {}".format(nsr_id)) + self.logger.debug("N2VC > DB_NSR > {}".format(db_nsr)) + self.logger.debug("N2VC > VCA_DEPLOYED > {}".format(vca_deployed)) + + # Launch task + task_n2vc = asyncio.ensure_future( + self.heal_N2VC( + logging_text=logging_text, + vca_index=vca_index, + nsi_id=nsi_id, + db_nsr=db_nsr, + db_vnfr=db_vnfr, + vdu_id=vdu_id, + kdu_name=kdu_name, + vdu_index=vdu_index, + deploy_params=deploy_params, + config_descriptor=descriptor_config, + base_folder=base_folder, + nslcmop_id=nslcmop_id, + stage=stage, + vca_type=vca_type, + vca_name=vca_name, + ee_config_descriptor=ee_item, + ) + ) + self.lcm_tasks.register( + "ns", + nsr_id, + nslcmop_id, + "instantiate_N2VC-{}".format(vca_index), + task_n2vc, + ) + task_instantiation_info[ + task_n2vc + ] = self.task_name_deploy_vca + " {}.{}".format( + member_vnf_index or "", vdu_id or "" + ) + + async def heal_N2VC( + self, + logging_text, + vca_index, + nsi_id, + db_nsr, + db_vnfr, + vdu_id, + kdu_name, + vdu_index, + config_descriptor, + deploy_params, + base_folder, + nslcmop_id, + stage, + vca_type, + vca_name, + ee_config_descriptor, + ): + nsr_id = db_nsr["_id"] + db_update_entry = "_admin.deployed.VCA.{}.".format(vca_index) + vca_deployed_list = db_nsr["_admin"]["deployed"]["VCA"] + vca_deployed = db_nsr["_admin"]["deployed"]["VCA"][vca_index] + osm_config = {"osm": {"ns_id": db_nsr["_id"]}} + db_dict = { + "collection": "nsrs", + "filter": {"_id": nsr_id}, + "path": db_update_entry, + } + step = "" + try: + element_type = "NS" + element_under_configuration = nsr_id + + vnfr_id = None + if db_vnfr: + vnfr_id = db_vnfr["_id"] + osm_config["osm"]["vnf_id"] = vnfr_id + + namespace = "{nsi}.{ns}".format(nsi=nsi_id if nsi_id else "", ns=nsr_id) + + if vca_type == "native_charm": + index_number = 0 + else: + index_number = vdu_index or 0 + + if vnfr_id: + element_type = "VNF" + element_under_configuration = vnfr_id + namespace += ".{}-{}".format(vnfr_id, index_number) + if vdu_id: + namespace += ".{}-{}".format(vdu_id, index_number) + element_type = "VDU" + element_under_configuration = "{}-{}".format(vdu_id, index_number) + osm_config["osm"]["vdu_id"] = vdu_id + elif kdu_name: + namespace += ".{}".format(kdu_name) + element_type = "KDU" + element_under_configuration = kdu_name + osm_config["osm"]["kdu_name"] = kdu_name + + # Get artifact path + if base_folder["pkg-dir"]: + artifact_path = "{}/{}/{}/{}".format( + base_folder["folder"], + base_folder["pkg-dir"], + "charms" + if vca_type + in ("native_charm", "lxc_proxy_charm", "k8s_proxy_charm") + else "helm-charts", + vca_name, + ) + else: + artifact_path = "{}/Scripts/{}/{}/".format( + base_folder["folder"], + "charms" + if vca_type + in ("native_charm", "lxc_proxy_charm", "k8s_proxy_charm") + else "helm-charts", + vca_name, + ) + + self.logger.debug("Artifact path > {}".format(artifact_path)) + + # get initial_config_primitive_list that applies to this element + initial_config_primitive_list = config_descriptor.get( + "initial-config-primitive" + ) + + self.logger.debug( + "Initial config primitive list > {}".format( + initial_config_primitive_list + ) + ) + + # add config if not present for NS charm + ee_descriptor_id = ee_config_descriptor.get("id") + self.logger.debug("EE Descriptor > {}".format(ee_descriptor_id)) + initial_config_primitive_list = get_ee_sorted_initial_config_primitive_list( + initial_config_primitive_list, vca_deployed, ee_descriptor_id + ) + + self.logger.debug( + "Initial config primitive list #2 > {}".format( + initial_config_primitive_list + ) + ) + # n2vc_redesign STEP 3.1 + # find old ee_id if exists + ee_id = vca_deployed.get("ee_id") + + vca_id = self.get_vca_id(db_vnfr, db_nsr) + # create or register execution environment in VCA. Only for native charms when healing + if vca_type == "native_charm": + step = "Waiting to VM being up and getting IP address" + self.logger.debug(logging_text + step) + rw_mgmt_ip = await self.wait_vm_up_insert_key_ro( + logging_text, + nsr_id, + vnfr_id, + vdu_id, + vdu_index, + user=None, + pub_key=None, + ) + credentials = {"hostname": rw_mgmt_ip} + # get username + username = deep_get( + config_descriptor, ("config-access", "ssh-access", "default-user") + ) + # TODO remove this when changes on IM regarding config-access:ssh-access:default-user were + # merged. Meanwhile let's get username from initial-config-primitive + if not username and initial_config_primitive_list: + for config_primitive in initial_config_primitive_list: + for param in config_primitive.get("parameter", ()): + if param["name"] == "ssh-username": + username = param["value"] + break + if not username: + raise LcmException( + "Cannot determine the username neither with 'initial-config-primitive' nor with " + "'config-access.ssh-access.default-user'" + ) + credentials["username"] = username + + # n2vc_redesign STEP 3.2 + # TODO: Before healing at RO it is needed to destroy native charm units to be deleted. + self._write_configuration_status( + nsr_id=nsr_id, + vca_index=vca_index, + status="REGISTERING", + element_under_configuration=element_under_configuration, + element_type=element_type, + ) + + step = "register execution environment {}".format(credentials) + self.logger.debug(logging_text + step) + ee_id = await self.vca_map[vca_type].register_execution_environment( + credentials=credentials, + namespace=namespace, + db_dict=db_dict, + vca_id=vca_id, + ) + + # update ee_id en db + db_dict_ee_id = { + "_admin.deployed.VCA.{}.ee_id".format(vca_index): ee_id, + } + self.update_db_2("nsrs", nsr_id, db_dict_ee_id) + + # for compatibility with MON/POL modules, the need model and application name at database + # TODO ask MON/POL if needed to not assuming anymore the format "model_name.application_name" + # Not sure if this need to be done when healing + """ + ee_id_parts = ee_id.split(".") + db_nsr_update = {db_update_entry + "ee_id": ee_id} + if len(ee_id_parts) >= 2: + model_name = ee_id_parts[0] + application_name = ee_id_parts[1] + db_nsr_update[db_update_entry + "model"] = model_name + db_nsr_update[db_update_entry + "application"] = application_name + """ + + # n2vc_redesign STEP 3.3 + # Install configuration software. Only for native charms. + step = "Install configuration Software" + + self._write_configuration_status( + nsr_id=nsr_id, + vca_index=vca_index, + status="INSTALLING SW", + element_under_configuration=element_under_configuration, + element_type=element_type, + # other_update=db_nsr_update, + other_update=None, + ) + + # TODO check if already done + self.logger.debug(logging_text + step) + config = None + if vca_type == "native_charm": + config_primitive = next( + (p for p in initial_config_primitive_list if p["name"] == "config"), + None, + ) + if config_primitive: + config = self._map_primitive_params( + config_primitive, {}, deploy_params + ) + await self.vca_map[vca_type].install_configuration_sw( + ee_id=ee_id, + artifact_path=artifact_path, + db_dict=db_dict, + config=config, + num_units=1, + vca_id=vca_id, + vca_type=vca_type, + ) + + # write in db flag of configuration_sw already installed + self.update_db_2( + "nsrs", nsr_id, {db_update_entry + "config_sw_installed": True} + ) + + # Not sure if this need to be done when healing + """ + # add relations for this VCA (wait for other peers related with this VCA) + await self._add_vca_relations( + logging_text=logging_text, + nsr_id=nsr_id, + vca_type=vca_type, + vca_index=vca_index, + ) + """ + + # if SSH access is required, then get execution environment SSH public + # if native charm we have waited already to VM be UP + if vca_type in ("k8s_proxy_charm", "lxc_proxy_charm", "helm", "helm-v3"): + pub_key = None + user = None + # self.logger.debug("get ssh key block") + if deep_get( + config_descriptor, ("config-access", "ssh-access", "required") + ): + # self.logger.debug("ssh key needed") + # Needed to inject a ssh key + user = deep_get( + config_descriptor, + ("config-access", "ssh-access", "default-user"), + ) + step = "Install configuration Software, getting public ssh key" + pub_key = await self.vca_map[vca_type].get_ee_ssh_public__key( + ee_id=ee_id, db_dict=db_dict, vca_id=vca_id + ) + + step = "Insert public key into VM user={} ssh_key={}".format( + user, pub_key + ) + else: + # self.logger.debug("no need to get ssh key") + step = "Waiting to VM being up and getting IP address" + self.logger.debug(logging_text + step) + + # n2vc_redesign STEP 5.1 + # wait for RO (ip-address) Insert pub_key into VM + # IMPORTANT: We need do wait for RO to complete healing operation. + await self._wait_heal_ro(nsr_id, self.timeout.ns_heal) + if vnfr_id: + if kdu_name: + rw_mgmt_ip = await self.wait_kdu_up( + logging_text, nsr_id, vnfr_id, kdu_name + ) + else: + rw_mgmt_ip = await self.wait_vm_up_insert_key_ro( + logging_text, + nsr_id, + vnfr_id, + vdu_id, + vdu_index, + user=user, + pub_key=pub_key, + ) + else: + rw_mgmt_ip = None # This is for a NS configuration + + self.logger.debug(logging_text + " VM_ip_address={}".format(rw_mgmt_ip)) + + # store rw_mgmt_ip in deploy params for later replacement + deploy_params["rw_mgmt_ip"] = rw_mgmt_ip + + # Day1 operations. + # get run-day1 operation parameter + runDay1 = deploy_params.get("run-day1", False) + self.logger.debug( + "Healing vnf={}, vdu={}, runDay1 ={}".format(vnfr_id, vdu_id, runDay1) + ) + if runDay1: + # n2vc_redesign STEP 6 Execute initial config primitive + step = "execute initial config primitive" + + # wait for dependent primitives execution (NS -> VNF -> VDU) + if initial_config_primitive_list: + await self._wait_dependent_n2vc( + nsr_id, vca_deployed_list, vca_index + ) + + # stage, in function of element type: vdu, kdu, vnf or ns + my_vca = vca_deployed_list[vca_index] + if my_vca.get("vdu_id") or my_vca.get("kdu_name"): + # VDU or KDU + stage[0] = "Stage 3/5: running Day-1 primitives for VDU." + elif my_vca.get("member-vnf-index"): + # VNF + stage[0] = "Stage 4/5: running Day-1 primitives for VNF." + else: + # NS + stage[0] = "Stage 5/5: running Day-1 primitives for NS." + + self._write_configuration_status( + nsr_id=nsr_id, vca_index=vca_index, status="EXECUTING PRIMITIVE" + ) + + self._write_op_status(op_id=nslcmop_id, stage=stage) + + check_if_terminated_needed = True + for initial_config_primitive in initial_config_primitive_list: + # adding information on the vca_deployed if it is a NS execution environment + if not vca_deployed["member-vnf-index"]: + deploy_params["ns_config_info"] = json.dumps( + self._get_ns_config_info(nsr_id) + ) + # TODO check if already done + primitive_params_ = self._map_primitive_params( + initial_config_primitive, {}, deploy_params + ) + + step = "execute primitive '{}' params '{}'".format( + initial_config_primitive["name"], primitive_params_ + ) + self.logger.debug(logging_text + step) + await self.vca_map[vca_type].exec_primitive( + ee_id=ee_id, + primitive_name=initial_config_primitive["name"], + params_dict=primitive_params_, + db_dict=db_dict, + vca_id=vca_id, + vca_type=vca_type, + ) + # Once some primitive has been exec, check and write at db if it needs to exec terminated primitives + if check_if_terminated_needed: + if config_descriptor.get("terminate-config-primitive"): + self.update_db_2( + "nsrs", + nsr_id, + {db_update_entry + "needed_terminate": True}, + ) + check_if_terminated_needed = False + + # TODO register in database that primitive is done + + # STEP 7 Configure metrics + # Not sure if this need to be done when healing + """ + if vca_type == "helm" or vca_type == "helm-v3": + prometheus_jobs = await self.extract_prometheus_scrape_jobs( + ee_id=ee_id, + artifact_path=artifact_path, + ee_config_descriptor=ee_config_descriptor, + vnfr_id=vnfr_id, + nsr_id=nsr_id, + target_ip=rw_mgmt_ip, + ) + if prometheus_jobs: + self.update_db_2( + "nsrs", + nsr_id, + {db_update_entry + "prometheus_jobs": prometheus_jobs}, + ) + + for job in prometheus_jobs: + self.db.set_one( + "prometheus_jobs", + {"job_name": job["job_name"]}, + job, + upsert=True, + fail_on_empty=False, + ) + + """ + step = "instantiated at VCA" + self.logger.debug(logging_text + step) + + self._write_configuration_status( + nsr_id=nsr_id, vca_index=vca_index, status="READY" + ) + + except Exception as e: # TODO not use Exception but N2VC exception + # self.update_db_2("nsrs", nsr_id, {db_update_entry + "instantiation": "FAILED"}) + if not isinstance( + e, (DbException, N2VCException, LcmException, asyncio.CancelledError) + ): + self.logger.error( + "Exception while {} : {}".format(step, e), exc_info=True + ) + self._write_configuration_status( + nsr_id=nsr_id, vca_index=vca_index, status="BROKEN" + ) + raise LcmException("{} {}".format(step, e)) from e + + async def _wait_heal_ro( + self, + nsr_id, + timeout=600, + ): + start_time = time() + while time() <= start_time + timeout: + db_nsr = self.db.get_one("nsrs", {"_id": nsr_id}) + operational_status_ro = db_nsr["_admin"]["deployed"]["RO"][ + "operational-status" + ] + self.logger.debug("Wait Heal RO > {}".format(operational_status_ro)) + if operational_status_ro != "healing": + break + await asyncio.sleep(15) + else: # timeout_ns_deploy + raise NgRoException("Timeout waiting ns to deploy") + + async def vertical_scale(self, nsr_id, nslcmop_id): + """ + Vertical Scale the VDUs in a NS + + :param: nsr_id: NS Instance ID + :param: nslcmop_id: nslcmop ID of migrate + + """ + # Try to lock HA task here + task_is_locked_by_me = self.lcm_tasks.lock_HA("ns", "nslcmops", nslcmop_id) + if not task_is_locked_by_me: + return + logging_text = "Task ns={} vertical scale ".format(nsr_id) + self.logger.debug(logging_text + "Enter") + # get all needed from database + db_nslcmop = None + db_nslcmop_update = {} + nslcmop_operation_state = None + db_nsr_update = {} + target = {} + exc = None + # in case of error, indicates what part of scale was failed to put nsr at error status + start_deploy = time() + + try: + # wait for any previous tasks in process + step = "Waiting for previous operations to terminate" + await self.lcm_tasks.waitfor_related_HA("ns", "nslcmops", nslcmop_id) + + self._write_ns_status( + nsr_id=nsr_id, + ns_state=None, + current_operation="VerticalScale", + current_operation_id=nslcmop_id, + ) + step = "Getting nslcmop from database" + self.logger.debug( + step + " after having waited for previous tasks to be completed" + ) + db_nslcmop = self.db.get_one("nslcmops", {"_id": nslcmop_id}) + operationParams = db_nslcmop.get("operationParams") + target = {} + target.update(operationParams) + desc = await self.RO.vertical_scale(nsr_id, target) + self.logger.debug("RO return > {}".format(desc)) + action_id = desc["action_id"] + await self._wait_ng_ro( + nsr_id, + action_id, + nslcmop_id, + start_deploy, + self.timeout.verticalscale, + operation="verticalscale", + ) + except (ROclient.ROClientException, DbException, LcmException) as e: + self.logger.error("Exit Exception {}".format(e)) + exc = e + except asyncio.CancelledError: + self.logger.error("Cancelled Exception while '{}'".format(step)) + exc = "Operation was cancelled" + except Exception as e: + exc = traceback.format_exc() + self.logger.critical( + "Exit Exception {} {}".format(type(e).__name__, e), exc_info=True + ) + finally: + self._write_ns_status( + nsr_id=nsr_id, + ns_state=None, + current_operation="IDLE", + current_operation_id=None, + ) + if exc: + db_nslcmop_update["detailed-status"] = "FAILED {}: {}".format(step, exc) + nslcmop_operation_state = "FAILED" + else: + nslcmop_operation_state = "COMPLETED" + db_nslcmop_update["detailed-status"] = "Done" + db_nsr_update["detailed-status"] = "Done" + + self._write_op_status( + op_id=nslcmop_id, + stage="", + error_message="", + operation_state=nslcmop_operation_state, + other_update=db_nslcmop_update, + ) + if nslcmop_operation_state: + try: + msg = { + "nsr_id": nsr_id, + "nslcmop_id": nslcmop_id, + "operationState": nslcmop_operation_state, + } + await self.msg.aiowrite("ns", "verticalscaled", msg) + except Exception as e: + self.logger.error( + logging_text + "kafka_write notification Exception {}".format(e) + ) + self.logger.debug(logging_text + "Exit") + self.lcm_tasks.remove("ns", nsr_id, nslcmop_id, "ns_verticalscale")