X-Git-Url: https://osm.etsi.org/gitweb/?a=blobdiff_plain;f=osm_lcm%2Fns.py;h=87eaecd578b9b6f2fdf50d2b1343eb7a8f8a38c4;hb=refs%2Fchanges%2F95%2F8395%2F2;hp=a76c121fc4dc3652360445b9d830178e7ccae49d;hpb=5705ed31c013e5cab19c2c01ac4429ec2fcd04bd;p=osm%2FLCM.git diff --git a/osm_lcm/ns.py b/osm_lcm/ns.py index a76c121..87eaecd 100644 --- a/osm_lcm/ns.py +++ b/osm_lcm/ns.py @@ -21,6 +21,7 @@ import yaml import logging import logging.handlers import traceback +import json from jinja2 import Environment, Template, meta, TemplateError, TemplateNotFound, TemplateSyntaxError from osm_lcm import ROclient @@ -455,8 +456,7 @@ class NsLcm(LcmBase): if isinstance(vld_params["ns-net"], dict): for vld_id, instance_scenario_id in vld_params["ns-net"].items(): RO_vld_ns_net = {"instance_scenario_id": instance_scenario_id, "osm_id": vld_id} - if RO_vld_ns_net: - populate_dict(RO_ns_params, ("networks", vld_params["name"], "use-network"), RO_vld_ns_net) + populate_dict(RO_ns_params, ("networks", vld_params["name"], "use-network"), RO_vld_ns_net) if "vnfd-connection-point-ref" in vld_params: for cp_params in vld_params["vnfd-connection-point-ref"]: # look for interface @@ -629,15 +629,16 @@ class NsLcm(LcmBase): else: raise LcmException("ns_update_vnfr: Not found member_vnf_index={} from VIM info".format(vnf_index)) - @staticmethod - def _get_ns_config_info(vca_deployed_list): + def _get_ns_config_info(self, nsr_id): """ Generates a mapping between vnf,vdu elements and the N2VC id - :param vca_deployed_list: List of database _admin.deploy.VCA that contains this list + :param nsr_id: id of nsr to get last database _admin.deployed.VCA that contains this list :return: a dictionary with {osm-config-mapping: {}} where its element contains: "": for a vnf configuration, or "..": for a vdu configuration """ + db_nsr = self.db.get_one("nsrs", {"_id": nsr_id}) + vca_deployed_list = db_nsr["_admin"]["deployed"]["VCA"] mapping = {} ns_config_info = {"osm-config-mapping": mapping} for vca in vca_deployed_list: @@ -961,6 +962,33 @@ class NsLcm(LcmBase): return ip_address + async def _wait_dependent_n2vc(self, nsr_id, vca_deployed_list, vca_index): + """ + Wait until dependent VCA deployments have been finished. NS wait for VNFs and VDUs. VNFs for VDUs + """ + my_vca = vca_deployed_list[vca_index] + if my_vca.get("vdu_id") or my_vca.get("kdu_name"): + return + timeout = 300 + while timeout >= 0: + for index, vca_deployed in enumerate(vca_deployed_list): + if index == vca_index: + continue + if not my_vca.get("member-vnf-index") or \ + (vca_deployed.get("member-vnf-index") == my_vca.get("member-vnf-index")): + if not vca_deployed.get("instantiation"): + break # wait + if vca_deployed["instantiation"] == "FAILED": + raise LcmException("Configuration aborted because dependent charm/s has failed") + else: + return + await asyncio.sleep(10) + timeout -= 1 + db_nsr = self.db.get_one("nsrs", {"_id": nsr_id}) + vca_deployed_list = db_nsr["_admin"]["deployed"]["VCA"] + + 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): nsr_id = db_nsr["_id"] @@ -972,9 +1000,6 @@ class NsLcm(LcmBase): 'filter': {'_id': nsr_id}, 'path': db_update_entry } - logging_text += "member_vnf_index={} vdu_id={}, vdu_index={} ".format(db_vnfr["member-vnf-index-ref"], - vdu_id, vdu_index) - step = "" try: vnfr_id = None @@ -1065,15 +1090,18 @@ class NsLcm(LcmBase): step = "Install configuration Software, getting public ssh key" pub_key = await self.n2vc.get_ee_ssh_public__key(ee_id=ee_id, db_dict=db_dict) - step = "Insert public key into VM" + step = "Insert public key into VM user={} ssh_key={}".format(user, pub_key) else: 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 - 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) + if vnfr_id: + 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)) @@ -1093,11 +1121,12 @@ class NsLcm(LcmBase): # add config if not present for NS charm initial_config_primitive_list = self._get_initial_config_primitive_list(initial_config_primitive_list, vca_deployed) - + if initial_config_primitive_list: + await self._wait_dependent_n2vc(nsr_id, vca_deployed_list, vca_index) 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"] = self._get_ns_config_info(vca_deployed_list) + 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) @@ -1112,12 +1141,28 @@ class NsLcm(LcmBase): # TODO register in database that primitive is done step = "instantiated at VCA" + self.update_db_2("nsrs", nsr_id, {db_update_entry + "instantiation": "COMPLETED"}) self.logger.debug(logging_text + step) except Exception as e: # TODO not use Exception but N2VC exception + self.update_db_2("nsrs", nsr_id, {db_update_entry + "instantiation": "FAILED"}) raise Exception("{} {}".format(step, e)) from e # TODO raise N2VC exception with 'step' extra information + 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): + try: + db_dict = dict() + if ns_state: + db_dict["nsState"] = ns_state + db_dict["currentOperation"] = current_operation + db_dict["currentOperationID"] = current_operation_id + db_dict["errorDescription"] = error_description + db_dict["errorDetail"] = error_detail + self.update_db_2("nsrs", nsr_id, db_dict) + except Exception as e: + self.logger.warn('Error writing NS status: {}'.format(e)) + async def instantiate(self, nsr_id, nslcmop_id): """ @@ -1165,6 +1210,14 @@ class NsLcm(LcmBase): # STEP 0: Reading database (nslcmops, nsrs, nsds, vnfrs, vnfds) + # nsState="BUILDING", currentOperation="INSTANTIATING", currentOperationID=nslcmop_id + self._write_ns_status( + nsr_id=nsr_id, + ns_state="BUILDING", + current_operation="INSTANTIATING", + current_operation_id=nslcmop_id + ) + # read from db: operation step = "Getting nslcmop={} from db".format(nslcmop_id) db_nslcmop = self.db.get_one("nslcmops", {"_id": nslcmop_id}) @@ -1290,7 +1343,7 @@ class NsLcm(LcmBase): descriptor_config = vnfd.get("vnf-configuration") if descriptor_config and descriptor_config.get("juju"): self._deploy_n2vc( - logging_text=logging_text, + logging_text=logging_text + "member_vnf_index={} ".format(member_vnf_index), db_nsr=db_nsr, db_vnfr=db_vnfr, nslcmop_id=nslcmop_id, @@ -1332,7 +1385,8 @@ class NsLcm(LcmBase): for vdu_index in range(int(vdud.get("count", 1))): # TODO vnfr_params["rw_mgmt_ip"] = vdur["ip-address"] self._deploy_n2vc( - logging_text=logging_text, + 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, @@ -1427,27 +1481,56 @@ class NsLcm(LcmBase): # while time() <= start_deploy + self.total_deploy_timeout: error_text_list = [] - timeout = 3600 # time() - start_deploy + timeout = 3600 + + # let's begin with all OK + instantiated_ok = True + # let's begin with RO 'running' status (later we can change it) + db_nsr_update["operational-status"] = "running" + # let's begin with VCA 'configured' status (later we can change it) + db_nsr_update["config-status"] = "configured" + if task_instantiation_list: + # wait for all tasks completion done, pending = await asyncio.wait(task_instantiation_list, timeout=timeout) - if pending: - for task in pending: - error_text_list.append(task_instantiation_info[task] + ": Timeout") + + for task in pending: + instantiated_ok = False + if task == task_ro: + db_nsr_update["operational-status"] = "failed" + else: + db_nsr_update["config-status"] = "failed" + self.logger.error(logging_text + task_instantiation_info[task] + ": Timeout") + error_text_list.append(task_instantiation_info[task] + ": Timeout") for task in done: if task.cancelled(): + instantiated_ok = False + if task == task_ro: + db_nsr_update["operational-status"] = "failed" + else: + db_nsr_update["config-status"] = "failed" + self.logger.warn(logging_text + task_instantiation_info[task] + ": Cancelled") error_text_list.append(task_instantiation_info[task] + ": Cancelled") - elif task.done(): + else: exc = task.exception() if exc: + instantiated_ok = False + if task == task_ro: + db_nsr_update["operational-status"] = "failed" + else: + db_nsr_update["config-status"] = "failed" + self.logger.error(logging_text + task_instantiation_info[task] + ": Failed") if isinstance(exc, (N2VCException, ROclient.ROClientException)): error_text_list.append(task_instantiation_info[task] + ": {}".format(exc)) else: - error_text_list.append(task_instantiation_info[task] + ": " + "". - join(traceback.format_exception(None, exc, exc.__traceback__))) + exc_traceback = "".join(traceback.format_exception(None, exc, exc.__traceback__)) + self.logger.error(logging_text + task_instantiation_info[task] + exc_traceback) + error_text_list.append(task_instantiation_info[task] + ": " + exc_traceback) + else: + self.logger.debug(logging_text + task_instantiation_info[task] + ": Done") if error_text_list: error_text = "\n".join(error_text_list) - db_nsr_update["config-status"] = "failed" db_nsr_update["detailed-status"] = error_text db_nslcmop_update["operationState"] = nslcmop_operation_state = "FAILED_TEMP" db_nslcmop_update["detailed-status"] = error_text @@ -1457,7 +1540,6 @@ class NsLcm(LcmBase): db_nslcmop_update["operationState"] = nslcmop_operation_state = "COMPLETED" db_nslcmop_update["statusEnteredTime"] = time() db_nslcmop_update["detailed-status"] = "done" - db_nsr_update["config-status"] = "configured" db_nsr_update["detailed-status"] = "done" except (ROclient.ROClientException, DbException, LcmException) as e: @@ -1475,6 +1557,7 @@ class NsLcm(LcmBase): if db_nsr: db_nsr_update["detailed-status"] = "ERROR {}: {}".format(step, exc) db_nsr_update["operational-status"] = "failed" + db_nsr_update["config-status"] = "failed" if db_nslcmop: db_nslcmop_update["detailed-status"] = "FAILED {}: {}".format(step, exc) db_nslcmop_update["operationState"] = nslcmop_operation_state = "FAILED" @@ -1485,6 +1568,26 @@ class NsLcm(LcmBase): db_nsr_update["_admin.current-operation"] = None db_nsr_update["_admin.operation-type"] = None self.update_db_2("nsrs", nsr_id, db_nsr_update) + + # nsState="READY/BROKEN", currentOperation="IDLE", currentOperationID=None + ns_state = None + error_description = None + error_detail = None + if instantiated_ok: + ns_state = "READY" + else: + ns_state = "BROKEN" + error_description = 'Operation: INSTANTIATING.{}, step: {}'.format(nslcmop_id, step) + error_detail = error_text + self._write_ns_status( + nsr_id=nsr_id, + ns_state=ns_state, + current_operation="IDLE", + current_operation_id=None, + error_description=error_description, + error_detail=error_detail + ) + if db_nslcmop_update: self.update_db_2("nslcmops", nslcmop_id, db_nslcmop_update) except DbException as e: @@ -1570,9 +1673,11 @@ class NsLcm(LcmBase): params=desc_params, db_dict=db_dict, timeout=3600) ) else: - task = self.k8sclusterjuju.install(cluster_uuid=cluster_uuid, kdu_model=kdumodel, - atomic=True, params=desc_params, - db_dict=db_dict, timeout=600) + task = asyncio.ensure_future( + self.k8sclusterjuju.install(cluster_uuid=cluster_uuid, kdu_model=kdumodel, + atomic=True, params=desc_params, + db_dict=db_dict, timeout=600) + ) pending_tasks[task] = "_admin.deployed.K8s.{}.".format(index) index += 1 @@ -1873,6 +1978,13 @@ class NsLcm(LcmBase): # or op_index (operationState != 'COMPLETED') return self._reintent_or_skip_suboperation(db_nslcmop, op_index) + # Function to return execution_environment id + + def _get_ee_id(self, vnf_index, vdu_id, vca_deployed_list): + for vca in vca_deployed_list: + if vca["member-vnf-index"] == vnf_index and vca["vdu_id"] == vdu_id: + return vca["ee_id"] + # Helper methods for terminate() async def _terminate_action(self, db_nslcmop, nslcmop_id, nsr_id): @@ -1880,6 +1992,8 @@ class NsLcm(LcmBase): Called from terminate() before deleting instance Calls action() to execute the primitive """ logging_text = "Task ns={} _terminate_action={} ".format(nsr_id, nslcmop_id) + db_nsr = self.db.get_one("nsrs", {"_id": nsr_id}) + vca_deployed_list = db_nsr["_admin"]["deployed"]["VCA"] db_vnfrs_list = self.db.get_list("vnfrs", {"nsr-id-ref": nsr_id}) db_vnfds = {} # Loop over VNFRs @@ -1931,8 +2045,7 @@ class NsLcm(LcmBase): # " primitive={} fails with error {}".format( # vnf_index, seq.get("name"), result_detail)) - # TODO: find ee_id - ee_id = None + ee_id = self._get_ee_id(vnf_index, vdu_id, vca_deployed_list) try: await self.n2vc.exec_primitive( ee_id=ee_id, @@ -1972,6 +2085,13 @@ class NsLcm(LcmBase): 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="TERMINATING", + current_operation="TERMINATING", + current_operation_id=nslcmop_id + ) + step = "Getting nslcmop={} from db".format(nslcmop_id) db_nslcmop = self.db.get_one("nslcmops", {"_id": nslcmop_id}) step = "Getting nsr={} from db".format(nsr_id) @@ -2146,6 +2266,7 @@ class NsLcm(LcmBase): self.logger.error(logging_text + failed_detail[-1]) if failed_detail: + terminate_ok = False self.logger.error(logging_text + " ;".join(failed_detail)) db_nsr_update["operational-status"] = "failed" db_nsr_update["detailed-status"] = "Deletion errors " + "; ".join(failed_detail) @@ -2153,6 +2274,7 @@ class NsLcm(LcmBase): db_nslcmop_update["operationState"] = nslcmop_operation_state = "FAILED" db_nslcmop_update["statusEnteredTime"] = time() else: + terminate_ok = True db_nsr_update["operational-status"] = "terminated" db_nsr_update["detailed-status"] = "Done" db_nsr_update["_admin.nsState"] = "NOT_INSTANTIATED" @@ -2184,6 +2306,25 @@ class NsLcm(LcmBase): db_nsr_update["_admin.current-operation"] = None db_nsr_update["_admin.operation-type"] = None self.update_db_2("nsrs", nsr_id, db_nsr_update) + + if terminate_ok: + ns_state = "IDLE" + error_description = None + error_detail = None + else: + ns_state = "BROKEN" + error_description = 'Operation: TERMINATING.{}, step: {}'.format(nslcmop_id, step) + error_detail = "; ".join(failed_detail) + + self._write_ns_status( + nsr_id=nsr_id, + ns_state=ns_state, + current_operation="IDLE", + current_operation_id=None, + error_description=error_description, + error_detail=error_detail + ) + except DbException as e: self.logger.error(logging_text + "Cannot update database: {}".format(e)) if nslcmop_operation_state: @@ -2330,6 +2471,13 @@ class NsLcm(LcmBase): 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="RUNNING ACTION", + 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}) @@ -2522,6 +2670,12 @@ class NsLcm(LcmBase): db_nsr_update["_admin.nslcmop"] = None db_nsr_update["_admin.current-operation"] = None self.update_db_2("nsrs", nsr_id, db_nsr_update) + self._write_ns_status( + nsr_id=nsr_id, + ns_state=None, + current_operation="IDLE", + current_operation_id=None + ) except DbException as e: self.logger.error(logging_text + "Cannot update database: {}".format(e)) self.logger.debug(logging_text + "Exit") @@ -2565,6 +2719,13 @@ class NsLcm(LcmBase): 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") db_nslcmop = self.db.get_one("nslcmops", {"_id": nslcmop_id}) @@ -2899,7 +3060,7 @@ class NsLcm(LcmBase): # Post-scale reintent 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') - if (op_index == self.SUBOPERATION_STATUS_SKIP): + if op_index == self.SUBOPERATION_STATUS_SKIP: # Skip sub-operation result = 'COMPLETED' result_detail = 'Done' @@ -2907,7 +3068,7 @@ class NsLcm(LcmBase): "vnf_config_primitive={} Skipped sub-operation, result {} {}". format(vnf_config_primitive, result, result_detail)) else: - if (op_index == self.SUBOPERATION_STATUS_NEW): + 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". @@ -2978,6 +3139,14 @@ class NsLcm(LcmBase): db_nsr_update["_admin.operation-type"] = None db_nsr_update["_admin.nslcmop"] = None self.update_db_2("nsrs", nsr_id, db_nsr_update) + + self._write_ns_status( + nsr_id=nsr_id, + ns_state=None, + current_operation="IDLE", + current_operation_id=None + ) + except DbException as e: self.logger.error(logging_text + "Cannot update database: {}".format(e)) if nslcmop_operation_state: