X-Git-Url: https://osm.etsi.org/gitweb/?a=blobdiff_plain;f=osm_lcm%2Fvim_sdn.py;h=4d8d83bef0be8e3355fca50c8d1a7aeb12420db8;hb=refs%2Fheads%2Ffeature7928;hp=7c0bc0b81a20ffb058f7b6ca31129028538448d5;hpb=73d8bd00fd53c8864fa0803f2ae48e93fb06c791;p=osm%2FLCM.git diff --git a/osm_lcm/vim_sdn.py b/osm_lcm/vim_sdn.py index 7c0bc0b..4d8d83b 100644 --- a/osm_lcm/vim_sdn.py +++ b/osm_lcm/vim_sdn.py @@ -16,10 +16,13 @@ # under the License. ## +import asyncio +import yaml import logging import logging.handlers from osm_lcm import ROclient from osm_lcm.lcm_utils import LcmException, LcmBase +from n2vc.k8s_helm_conn import K8sHelmConnector from osm_common.dbbase import DbException from copy import deepcopy @@ -912,3 +915,349 @@ class SdnLcm(LcmBase): except DbException as e: self.logger.error(logging_text + "Cannot update database: {}".format(e)) self.lcm_tasks.remove("sdn", sdn_id, order_id) + + +class K8sClusterLcm(LcmBase): + + def __init__(self, db, msg, fs, lcm_tasks, vca_config, loop): + """ + Init, Connect to database, filesystem storage, and messaging + :param config: two level dictionary with configuration. Top level should contain 'database', 'storage', + :return: None + """ + + self.logger = logging.getLogger('lcm.k8scluster') + self.loop = loop + self.lcm_tasks = lcm_tasks + self.vca_config = vca_config + self.fs = fs + self.db = db + + self.k8scluster = K8sHelmConnector( + kubectl_command=self.vca_config.get("kubectlpath"), + helm_command=self.vca_config.get("helmpath"), + fs=self.fs, + log=self.logger, + db=self.db, + on_update_db=None + ) + + super().__init__(db, msg, fs, self.logger) + + async def create(self, k8scluster_content, order_id): + + # HA tasks and backward compatibility: + # If 'vim_content' does not include 'op_id', we a running a legacy NBI version. + # In such a case, HA is not supported by NBI, 'op_id' is None, and lock_HA() will do nothing. + # Register 'create' task here for related future HA operations + op_id = k8scluster_content.pop('op_id', None) + if not self.lcm_tasks.lock_HA('k8scluster', 'create', op_id): + return + + k8scluster_id = k8scluster_content["_id"] + k8scluster_content.pop("op_id", None) + logging_text = "Task k8scluster_create={} ".format(k8scluster_id) + self.logger.debug(logging_text + "Enter") + + db_k8scluster = None + db_k8scluster_update = {} + + exc = None + operationState_HA = '' + detailed_status_HA = '' + try: + step = "Getting k8scluster-id='{}' from db".format(k8scluster_id) + self.logger.debug(logging_text + step) + db_k8scluster = self.db.get_one("k8sclusters", {"_id": k8scluster_id}) + self.db.encrypt_decrypt_fields(db_k8scluster.get("credentials"), 'decrypt', ['password', 'secret'], + schema_version=db_k8scluster["schema_version"], salt=db_k8scluster["_id"]) + print(db_k8scluster.get("credentials")) + print("\n\n\n FIN CREDENTIALS") + print(yaml.safe_dump(db_k8scluster.get("credentials"))) + print("\n\n\n FIN OUTPUT") + cluster_uuid, uninstall_sw = await self.k8scluster.init_env(yaml.safe_dump(db_k8scluster. + get("credentials"))) + db_k8scluster_update["cluster-uuid"] = cluster_uuid + if uninstall_sw: + db_k8scluster_update["uninstall-sw"] = uninstall_sw + step = "Getting the list of repos" + self.logger.debug(logging_text + step) + task_list = [] + db_k8srepo_list = self.db.get_list("k8srepos", {}) + for repo in db_k8srepo_list: + step = "Adding repo {} to cluster: {}".format(repo["name"], cluster_uuid) + self.logger.debug(logging_text + step) + task = asyncio.ensure_future(self.k8scluster.repo_add(cluster_uuid=cluster_uuid, + name=repo["name"], url=repo["url"], + repo_type="chart")) + task_list.append(task) + if not repo["_admin"].get("cluster-inserted"): + repo["_admin"]["cluster-inserted"] = [] + repo["_admin"]["cluster-inserted"].append(cluster_uuid) + self.update_db_2("k8srepos", repo["_id"], repo) + + done = None + pending = None + if len(task_list) > 0: + self.logger.debug('Waiting for terminate pending tasks...') + done, pending = await asyncio.wait(task_list, timeout=3600) + if not pending: + self.logger.debug('All tasks finished...') + else: + self.logger.info('There are pending tasks: {}'.format(pending)) + db_k8scluster_update["_admin.operationalState"] = "ENABLED" + + except Exception as e: + self.logger.critical(logging_text + "Exit Exception {}".format(e), exc_info=True) + exc = e + finally: + if exc and db_k8scluster: + db_k8scluster_update["_admin.operationalState"] = "ERROR" + db_k8scluster_update["_admin.detailed-status"] = "ERROR {}: {}".format(step, exc) + # Mark the k8scluster 'create' HA task as erroneous + operationState_HA = 'FAILED' + detailed_status_HA = "ERROR {}: {}".format(step, exc) + try: + if db_k8scluster_update: + self.update_db_2("k8sclusters", k8scluster_id, db_k8scluster_update) + # Register the K8scluster 'create' HA task either + # succesful or erroneous, or do nothing (if legacy NBI) + self.lcm_tasks.register_HA('k8scluster', 'create', op_id, + operationState=operationState_HA, + detailed_status=detailed_status_HA) + except DbException as e: + self.logger.error(logging_text + "Cannot update database: {}".format(e)) + self.lcm_tasks.remove("k8sclusters", k8scluster_id, order_id) + + async def delete(self, k8scluster_content, order_id): + + # HA tasks and backward compatibility: + # If 'vim_content' does not include 'op_id', we a running a legacy NBI version. + # In such a case, HA is not supported by NBI, 'op_id' is None, and lock_HA() will do nothing. + # Register 'delete' task here for related future HA operations + op_id = k8scluster_content.pop('op_id', None) + if not self.lcm_tasks.lock_HA('k8scluster', 'delete', op_id): + return + + k8scluster_id = k8scluster_content["_id"] + k8scluster_content.pop("op_id", None) + logging_text = "Task k8scluster_delete={} ".format(k8scluster_id) + self.logger.debug(logging_text + "Enter") + + db_k8scluster = None + db_k8scluster_update = {} + exc = None + operationState_HA = '' + detailed_status_HA = '' + try: + step = "Getting k8scluster='{}' from db".format(k8scluster_id) + self.logger.debug(logging_text + step) + db_k8scluster = self.db.get_one("k8sclusters", {"_id": k8scluster_id}) + uninstall_sw = db_k8scluster.get("uninstall-sw") + if uninstall_sw is False or uninstall_sw is None: + uninstall_sw = False + cluster_removed = await self.k8scluster.reset(cluster_uuid=db_k8scluster.get("cluster-uuid"), + uninstall_sw=uninstall_sw) + + if cluster_removed: + step = "Removing k8scluster='{}' from db".format(k8scluster_id) + self.logger.debug(logging_text + step) + db_k8srepo_list = self.db.get_list("k8srepos", {}) + for k8srepo in db_k8srepo_list: + index = 0 + for cluster in k8srepo["_admin"]["cluster-inserted"]: + if db_k8scluster.get("cluster-uuid") == cluster: + del(k8srepo["_admin"]["cluster-inserted"][index]) + break + index += 1 + self.update_db_2("k8srepos", k8srepo["_id"], k8srepo) + self.db.del_one("k8sclusters", {"_id": k8scluster_id}) + else: + raise LcmException("An error happened during the reset of the k8s cluster '{}'".format(k8scluster_id)) + # if not cluster_removed: + # raise Exception("K8scluster was not properly removed") + + except Exception as e: + self.logger.critical(logging_text + "Exit Exception {}".format(e), exc_info=True) + exc = e + finally: + if exc and db_k8scluster: + db_k8scluster_update["_admin.operationalState"] = "ERROR" + db_k8scluster_update["_admin.detailed-status"] = "ERROR {}: {}".format(step, exc) + # Mark the WIM 'create' HA task as erroneous + operationState_HA = 'FAILED' + detailed_status_HA = "ERROR {}: {}".format(step, exc) + try: + if db_k8scluster_update: + self.update_db_2("k8sclusters", k8scluster_id, db_k8scluster_update) + # Register the K8scluster 'delete' HA task either + # succesful or erroneous, or do nothing (if legacy NBI) + self.lcm_tasks.register_HA('k8scluster', 'delete', op_id, + operationState=operationState_HA, + detailed_status=detailed_status_HA) + except DbException as e: + self.logger.error(logging_text + "Cannot update database: {}".format(e)) + self.lcm_tasks.remove("k8sclusters", k8scluster_id, order_id) + + +class K8sRepoLcm(LcmBase): + + def __init__(self, db, msg, fs, lcm_tasks, vca_config, loop): + """ + Init, Connect to database, filesystem storage, and messaging + :param config: two level dictionary with configuration. Top level should contain 'database', 'storage', + :return: None + """ + + self.logger = logging.getLogger('lcm.k8srepo') + self.loop = loop + self.lcm_tasks = lcm_tasks + self.vca_config = vca_config + self.fs = fs + self.db = db + + self.k8srepo = K8sHelmConnector( + kubectl_command=self.vca_config.get("kubectlpath"), + helm_command=self.vca_config.get("helmpath"), + fs=self.fs, + log=self.logger, + db=self.db, + on_update_db=None + ) + + super().__init__(db, msg, fs, self.logger) + + async def create(self, k8srepo_content, order_id): + + # HA tasks and backward compatibility: + # If 'vim_content' does not include 'op_id', we a running a legacy NBI version. + # In such a case, HA is not supported by NBI, 'op_id' is None, and lock_HA() will do nothing. + # Register 'create' task here for related future HA operations + + op_id = k8srepo_content.pop('op_id', None) + if not self.lcm_tasks.lock_HA('k8srepo', 'create', op_id): + return + + k8srepo_id = k8srepo_content.get("_id") + logging_text = "Task k8srepo_create={} ".format(k8srepo_id) + self.logger.debug(logging_text + "Enter") + + db_k8srepo = None + db_k8srepo_update = {} + exc = None + operationState_HA = '' + detailed_status_HA = '' + try: + step = "Getting k8srepo-id='{}' from db".format(k8srepo_id) + self.logger.debug(logging_text + step) + db_k8srepo = self.db.get_one("k8srepos", {"_id": k8srepo_id}) + step = "Getting k8scluster_list from db" + self.logger.debug(logging_text + step) + db_k8scluster_list = self.db.get_list("k8sclusters", {}) + db_k8srepo_update["_admin.cluster-inserted"] = [] + task_list = [] + for k8scluster in db_k8scluster_list: + step = "Adding repo to cluster: {}".format(k8scluster["cluster-uuid"]) + self.logger.debug(logging_text + step) + task = asyncio.ensure_future(self.k8srepo.repo_add(cluster_uuid=k8scluster["cluster-uuid"], + name=db_k8srepo["name"], url=db_k8srepo["url"], + repo_type="chart")) + task_list.append(task) + db_k8srepo_update["_admin.cluster-inserted"].append(k8scluster["cluster-uuid"]) + + done = None + pending = None + if len(task_list) > 0: + self.logger.debug('Waiting for terminate pending tasks...') + done, pending = await asyncio.wait(task_list, timeout=3600) + if not pending: + self.logger.debug('All tasks finished...') + else: + self.logger.info('There are pending tasks: {}'.format(pending)) + db_k8srepo_update["_admin.operationalState"] = "ENABLED" + except Exception as e: + self.logger.critical(logging_text + "Exit Exception {}".format(e), exc_info=True) + exc = e + finally: + if exc and db_k8srepo: + db_k8srepo_update["_admin.operationalState"] = "ERROR" + db_k8srepo_update["_admin.detailed-status"] = "ERROR {}: {}".format(step, exc) + # Mark the WIM 'create' HA task as erroneous + operationState_HA = 'FAILED' + detailed_status_HA = "ERROR {}: {}".format(step, exc) + try: + if db_k8srepo_update: + self.update_db_2("k8srepos", k8srepo_id, db_k8srepo_update) + # Register the K8srepo 'create' HA task either + # succesful or erroneous, or do nothing (if legacy NBI) + self.lcm_tasks.register_HA('k8srepo', 'create', op_id, + operationState=operationState_HA, + detailed_status=detailed_status_HA) + except DbException as e: + self.logger.error(logging_text + "Cannot update database: {}".format(e)) + self.lcm_tasks.remove("k8srepo", k8srepo_id, order_id) + + async def delete(self, k8srepo_content, order_id): + + # HA tasks and backward compatibility: + # If 'vim_content' does not include 'op_id', we a running a legacy NBI version. + # In such a case, HA is not supported by NBI, 'op_id' is None, and lock_HA() will do nothing. + # Register 'delete' task here for related future HA operations + op_id = k8srepo_content.pop('op_id', None) + if not self.lcm_tasks.lock_HA('k8srepo', 'delete', op_id): + return + + k8srepo_id = k8srepo_content.get("_id") + logging_text = "Task k8srepo_delete={} ".format(k8srepo_id) + self.logger.debug(logging_text + "Enter") + + db_k8srepo = None + db_k8srepo_update = {} + + operationState_HA = '' + detailed_status_HA = '' + try: + step = "Getting k8srepo-id='{}' from db".format(k8srepo_id) + self.logger.debug(logging_text + step) + db_k8srepo = self.db.get_one("k8srepos", {"_id": k8srepo_id}) + step = "Getting k8scluster_list from db" + self.logger.debug(logging_text + step) + db_k8scluster_list = self.db.get_list("k8sclusters", {}) + + task_list = [] + for k8scluster in db_k8scluster_list: + task = asyncio.ensure_future(self.k8srepo.repo_remove(cluster_uuid=k8scluster["cluster-uuid"], + name=db_k8srepo["name"])) + task_list.append(task) + done = None + pending = None + if len(task_list) > 0: + self.logger.debug('Waiting for terminate pending tasks...') + done, pending = await asyncio.wait(task_list, timeout=3600) + if not pending: + self.logger.debug('All tasks finished...') + else: + self.logger.info('There are pending tasks: {}'.format(pending)) + self.db.del_one("k8srepos", {"_id": k8srepo_id}) + + except Exception as e: + self.logger.critical(logging_text + "Exit Exception {}".format(e), exc_info=True) + exc = e + finally: + if exc and db_k8srepo: + db_k8srepo_update["_admin.operationalState"] = "ERROR" + db_k8srepo_update["_admin.detailed-status"] = "ERROR {}: {}".format(step, exc) + # Mark the WIM 'create' HA task as erroneous + operationState_HA = 'FAILED' + detailed_status_HA = "ERROR {}: {}".format(step, exc) + try: + if db_k8srepo_update: + self.update_db_2("k8srepos", k8srepo_id, db_k8srepo_update) + # Register the K8srepo 'delete' HA task either + # succesful or erroneous, or do nothing (if legacy NBI) + self.lcm_tasks.register_HA('k8srepo', 'delete', op_id, + operationState=operationState_HA, + detailed_status=detailed_status_HA) + except DbException as e: + self.logger.error(logging_text + "Cannot update database: {}".format(e)) + self.lcm_tasks.remove("k8srepo", k8srepo_id, order_id)