OceanBase 升级过程研究(4.2.1.6-4.2.1.8)

模拟业务

使用benchmark加载10仓数据模拟业务场景

升级方法

使用滚动升级方式来进行OB升级。该方法前提是OB集群必须满足官方规定的高可用架构(如果 Zone 个数小于 3，滚动升级时则无法构成多数派), 滚动升级的原理就是轮流完成每个ZONE的升级工作，由于OB属于分布式集群，当一个ZONE处于升级状态中，其余的ZONE提供服务，不影响业务

任务类型

升级时会产生一个主任务及多个子任务。主任务根据每个需要替换 binary 版本的升级节点，来生成数个子任务，来逐步替换掉需要升级的 binary 文件。

• 主任务：将升级路径中每个需要替换 binary 文件的升级节点都生成为一个子任务，用于逐步替换掉需要升级的 binary 文件。 如下图所示，Submit cluster upgrade task 即为生成一个子任务。Wait dag success 为等待子任务执行完成。Submit cluster upgrade task 和 Wait dag success 这个组合可能会执行多次，执行次数取决于升级路径中有多少个替换 binary 的节点。

• 子任务：升级所有 OceanBase 集群版本。主要是执行升级脚本: 预检查脚本、升级 Pre 脚本、替换 OBServer 节点、升级 Post 脚本、版本检查等。

升级任务(子任务)

由于是滚动升级，所以子任务中的升级任务数量取决于ZONE的数量

升级前检查

在 set zone contaxt 任务模块前的所有任务均是升级前对集群的检查，检查租户的凭据以及运行安装目录etc下的升级检查脚本，查看当前集群是否满足升级。由于是检查操作，所以这些任务模块如果运行失败可以进行回滚。这些检查模块运行成功后开始执行ZONE的升级

upgrade_checker.py 脚本进行升级前置检查，脚本执行成功表示可以继续升级

upgrade_pre.py 该脚本主要用于执行升级前的一系列准备工作，确保集群处于适合升级的状态

upgrade_health_checker.py该脚本主要用于执行集群级别的健康检查，确保集群在升级前处于健康状态

upgrade_checker.py脚本内容

[root@sdw1 etc]# cat upgrade_checker.py
#!/usr/bin/env python
# -*- coding: utf-8 -*-import sys
import os
import mysql.connector
from mysql.connector import errorcode
import logging
import getopt
import timeclass UpgradeParams:log_filename = 'upgrade_checker.log'old_version = '4.0.0.0'
#### --------------start : my_error.py --------------
class MyError(Exception):def __init__(self, value):self.value = valuedef __str__(self):return repr(self.value)
#### --------------start : actions.py------------
class Cursor:__cursor = Nonedef __init__(self, cursor):self.__cursor = cursordef exec_sql(self, sql, print_when_succ = True):try:self.__cursor.execute(sql)rowcount = self.__cursor.rowcountif True == print_when_succ:logging.info('succeed to execute sql: %s, rowcount = %d', sql, rowcount)return rowcountexcept mysql.connector.Error, e:logging.exception('mysql connector error, fail to execute sql: %s', sql)raise eexcept Exception, e:logging.exception('normal error, fail to execute sql: %s', sql)raise edef exec_query(self, sql, print_when_succ = True):try:self.__cursor.execute(sql)results = self.__cursor.fetchall()rowcount = self.__cursor.rowcountif True == print_when_succ:logging.info('succeed to execute query: %s, rowcount = %d', sql, rowcount)return (self.__cursor.description, results)except mysql.connector.Error, e:logging.exception('mysql connector error, fail to execute sql: %s', sql)raise eexcept Exception, e:logging.exception('normal error, fail to execute sql: %s', sql)raise edef set_parameter(cur, parameter, value):sql = """alter system set {0} = '{1}'""".format(parameter, value)logging.info(sql)cur.execute(sql)wait_parameter_sync(cur, parameter, value)def wait_parameter_sync(cur, key, value):sql = """select count(*) as cnt from oceanbase.__all_virtual_sys_parameter_statwhere name = '{0}' and value != '{1}'""".format(key, value)times = 10while times > 0:logging.info(sql)cur.execute(sql)result = cur.fetchall()if len(result) != 1 or len(result[0]) != 1:logging.exception('result cnt not match')raise eelif result[0][0] == 0:logging.info("""{0} is sync, value is {1}""".format(key, value))breakelse:logging.info("""{0} is not sync, value should be {1}""".format(key, value))times -= 1if times == 0:logging.exception("""check {0}:{1} sync timeout""".format(key, value))raise etime.sleep(5)#### --------------start :  opt.py --------------
help_str = \
"""
Help:
""" +\
sys.argv[0] + """ [OPTIONS]""" +\
'\n\n' +\
'-I, --help          Display this help and exit.\n' +\
'-V, --version       Output version information and exit.\n' +\
'-h, --host=name     Connect to host.\n' +\
'-P, --port=name     Port number to use for connection.\n' +\
'-u, --user=name     User for login.\n' +\
'-t, --timeout=name  Cmd/Query/Inspection execute timeout(s).\n' +\
'-p, --password=name Password to use when connecting to server. If password is\n' +\
'                    not given it\'s empty string "".\n' +\
'-m, --module=name   Modules to run. Modules should be a string combined by some of\n' +\
'                    the following strings: ddl, normal_dml, each_tenant_dml,\n' +\
'                    system_variable_dml, special_action, all. "all" represents\n' +\
'                    that all modules should be run. They are splitted by ",".\n' +\
'                    For example: -m all, or --module=ddl,normal_dml,special_action\n' +\
'-l, --log-file=name Log file path. If log file path is not given it\'s ' + os.path.splitext(sys.argv[0])[0] + '.log\n' +\
'\n\n' +\
'Maybe you want to run cmd like that:\n' +\
sys.argv[0] + ' -h 127.0.0.1 -P 3306 -u admin -p admin\n'version_str = """version 1.0.0"""class Option:__g_short_name_set = set([])__g_long_name_set = set([])__short_name = None__long_name = None__is_with_param = None__is_local_opt = None__has_value = None__value = Nonedef __init__(self, short_name, long_name, is_with_param, is_local_opt, default_value = None):if short_name in Option.__g_short_name_set:raise MyError('duplicate option short name: {0}'.format(short_name))elif long_name in Option.__g_long_name_set:raise MyError('duplicate option long name: {0}'.format(long_name))Option.__g_short_name_set.add(short_name)Option.__g_long_name_set.add(long_name)self.__short_name = short_nameself.__long_name = long_nameself.__is_with_param = is_with_paramself.__is_local_opt = is_local_optself.__has_value = Falseif None != default_value:self.set_value(default_value)def is_with_param(self):return self.__is_with_paramdef get_short_name(self):return self.__short_namedef get_long_name(self):return self.__long_namedef has_value(self):return self.__has_valuedef get_value(self):return self.__valuedef set_value(self, value):self.__value = valueself.__has_value = Truedef is_local_opt(self):return self.__is_local_optdef is_valid(self):return None != self.__short_name and None != self.__long_name and True == self.__has_value and None != self.__valueg_opts =\
[\
Option('I', 'help', False, True),\
Option('V', 'version', False, True),\
Option('h', 'host', True, False),\
Option('P', 'port', True, False),\
Option('u', 'user', True, False),\
Option('t', 'timeout', True, False, 0),\
Option('p', 'password', True, False, ''),\
# 要跑哪个模块，默认全跑
Option('m', 'module', True, False, 'all'),\
# 日志文件路径，不同脚本的main函数中中会改成不同的默认值
Option('l', 'log-file', True, False)
]\def change_opt_defult_value(opt_long_name, opt_default_val):global g_optsfor opt in g_opts:if opt.get_long_name() == opt_long_name:opt.set_value(opt_default_val)returndef has_no_local_opts():global g_optsno_local_opts = Truefor opt in g_opts:if opt.is_local_opt() and opt.has_value():no_local_opts = Falsereturn no_local_optsdef check_db_client_opts():global g_optsfor opt in g_opts:if not opt.is_local_opt() and not opt.has_value():raise MyError('option "-{0}" has not been specified, maybe you should run "{1} --help" for help'\.format(opt.get_short_name(), sys.argv[0]))def parse_option(opt_name, opt_val):global g_optsfor opt in g_opts:if opt_name in (('-' + opt.get_short_name()), ('--' + opt.get_long_name())):opt.set_value(opt_val)def parse_options(argv):global g_optsshort_opt_str = ''long_opt_list = []for opt in g_opts:if opt.is_with_param():short_opt_str += opt.get_short_name() + ':'else:short_opt_str += opt.get_short_name()for opt in g_opts:if opt.is_with_param():long_opt_list.append(opt.get_long_name() + '=')else:long_opt_list.append(opt.get_long_name())(opts, args) = getopt.getopt(argv, short_opt_str, long_opt_list)for (opt_name, opt_val) in opts:parse_option(opt_name, opt_val)if has_no_local_opts():check_db_client_opts()def deal_with_local_opt(opt):if 'help' == opt.get_long_name():global help_strprint help_strelif 'version' == opt.get_long_name():global version_strprint version_strdef deal_with_local_opts():global g_optsif has_no_local_opts():raise MyError('no local options, can not deal with local options')else:for opt in g_opts:if opt.is_local_opt() and opt.has_value():deal_with_local_opt(opt)# 只处理一个returndef get_opt_host():global g_optsfor opt in g_opts:if 'host' == opt.get_long_name():return opt.get_value()def get_opt_port():global g_optsfor opt in g_opts:if 'port' == opt.get_long_name():return opt.get_value()def get_opt_user():global g_optsfor opt in g_opts:if 'user' == opt.get_long_name():return opt.get_value()def get_opt_password():global g_optsfor opt in g_opts:if 'password' == opt.get_long_name():return opt.get_value()def get_opt_timeout():global g_optsfor opt in g_opts:if 'timeout' == opt.get_long_name():return opt.get_value()def get_opt_module():global g_optsfor opt in g_opts:if 'module' == opt.get_long_name():return opt.get_value()def get_opt_log_file():global g_optsfor opt in g_opts:if 'log-file' == opt.get_long_name():return opt.get_value()
#### ---------------end----------------------#### --------------start :  do_upgrade_pre.py--------------
def config_logging_module(log_filenamme):logging.basicConfig(level=logging.INFO,\format='[%(asctime)s] %(levelname)s %(filename)s:%(lineno)d %(message)s',\datefmt='%Y-%m-%d %H:%M:%S',\filename=log_filenamme,\filemode='w')# 定义日志打印格式formatter = logging.Formatter('[%(asctime)s] %(levelname)s %(filename)s:%(lineno)d %(message)s', '%Y-%m-%d %H:%M:%S')######################################## 定义一个Handler打印INFO及以上级别的日志到sys.stdoutstdout_handler = logging.StreamHandler(sys.stdout)stdout_handler.setLevel(logging.INFO)# 设置日志打印格式stdout_handler.setFormatter(formatter)# 将定义好的stdout_handler日志handler添加到root loggerlogging.getLogger('').addHandler(stdout_handler)
#### ---------------end----------------------fail_list=[]def get_version(version_str):versions = version_str.split(".")if len(versions) != 4:logging.exception("""version:{0} is invalid""".format(version_str))raise emajor = int(versions[0])minor = int(versions[1])major_patch = int(versions[2])minor_patch = int(versions[3])if major > 0xffffffff or minor > 0xffff or major_patch > 0xff or minor_patch > 0xff:logging.exception("""version:{0} is invalid""".format(version_str))raise eversion = (major << 32) | (minor << 16) | (major_patch << 8) | (minor_patch)return version#### START ####
# 1. 检查前置版本
def check_observer_version(query_cur, upgrade_params):(desc, results) = query_cur.exec_query("""select distinct value from GV$OB_PARAMETERS  where name='min_observer_version'""")if len(results) != 1:fail_list.append('min_observer_version is not sync')elif cmp(results[0][0], upgrade_params.old_version) < 0 :fail_list.append('old observer version is expected equal or higher then: {0}, actual version:{1}'.format(upgrade_params.old_version, results[0][0]))logging.info('check observer version success, version = {0}'.format(results[0][0]))def check_data_version(query_cur, input_min_cluster_version):min_cluster_version = 0sql = """select distinct value from GV$OB_PARAMETERS  where name='min_observer_version'"""(desc, results) = query_cur.exec_query(sql)if len(results) != 1:fail_list.append('min_observer_version is not sync')elif len(results[0]) != 1:fail_list.append('column cnt not match')else:min_cluster_version = get_version(results[0][0])input_min_cluster_version[0] = min_cluster_version# check data versionif min_cluster_version < get_version("4.1.0.0"):# last barrier cluster version should be 4.1.0.0fail_list.append('last barrier cluster version is 4.1.0.0. prohibit cluster upgrade from cluster version less than 4.1.0.0')else:data_version_str = ''data_version = 0# check compatible is samesql = """select distinct value from oceanbase.__all_virtual_tenant_parameter_info where name='compatible'"""(desc, results) = query_cur.exec_query(sql)if len(results) != 1:fail_list.append('compatible is not sync')elif len(results[0]) != 1:fail_list.append('column cnt not match')else:data_version_str = results[0][0]data_version = get_version(results[0][0])if data_version < get_version("4.1.0.0"):# last barrier data version should be 4.1.0.0fail_list.append('last barrier data version is 4.1.0.0. prohibit cluster upgrade from data version less than 4.1.0.0')else:# check target_data_version/current_data_versionsql = "select count(*) from oceanbase.__all_tenant"(desc, results) = query_cur.exec_query(sql)if len(results) != 1 or len(results[0]) != 1:fail_list.append('result cnt not match')else:tenant_count = results[0][0]sql = "select count(*) from __all_virtual_core_table where column_name in ('target_data_version', 'current_data_version') and column_value = {0}".format(data_version)(desc, results) = query_cur.exec_query(sql)if len(results) != 1 or len(results[0]) != 1:fail_list.append('result cnt not match')elif 2 * tenant_count != results[0][0]:fail_list.append('target_data_version/current_data_version not match with {0}, tenant_cnt:{1}, result_cnt:{2}'.format(data_version_str, tenant_count, results[0][0]))else:logging.info("check data version success, all tenant's compatible/target_data_version/current_data_version is {0}".format(data_version_str))# 2. 检查paxos副本是否同步, paxos副本是否缺失
def check_paxos_replica(query_cur):# 2.1 检查paxos副本是否同步(desc, results) = query_cur.exec_query("""select count(1) as unsync_cnt from GV$OB_LOG_STAT where in_sync = 'NO'""")if results[0][0] > 0 :fail_list.append('{0} replicas unsync, please check'.format(results[0][0]))# 2.2 检查paxos副本是否有缺失 TODOlogging.info('check paxos replica success')# 3. 检查是否有做balance, locality变更
def check_rebalance_task(query_cur):# 3.1 检查是否有做locality变更(desc, results) = query_cur.exec_query("""select count(1) as cnt from DBA_OB_TENANT_JOBS where job_status='INPROGRESS' and result_code is null""")if results[0][0] > 0 :fail_list.append('{0} locality tasks is doing, please check'.format(results[0][0]))# 3.2 检查是否有做balance(desc, results) = query_cur.exec_query("""select count(1) as rebalance_task_cnt from CDB_OB_LS_REPLICA_TASKS""")if results[0][0] > 0 :fail_list.append('{0} rebalance tasks is doing, please check'.format(results[0][0]))logging.info('check rebalance task success')# 4. 检查集群状态
def check_cluster_status(query_cur, min_cluster_version):# 4.1 检查是否非合并状态(desc, results) = query_cur.exec_query("""select count(1) from CDB_OB_MAJOR_COMPACTION where (GLOBAL_BROADCAST_SCN > LAST_SCN or STATUS != 'IDLE')""")if results[0][0] > 0 :fail_list.append('{0} tenant is merging, please check'.format(results[0][0]))if len(min_cluster_version) > 0 and min_cluster_version[0] < get_version("4.2.0.0"):(desc, results) = query_cur.exec_query("""select /*+ query_timeout(1000000000) */ count(1) from __all_virtual_tablet_compaction_info where max_received_scn > finished_scn and max_received_scn > 0""")if results[0][0] > 0 :fail_list.append('{0} tablet is merging, please check'.format(results[0][0]))logging.info('check cluster tablet major status success, cluster_version={0}'.format(min_cluster_version))else:logging.info('skip check cluster tablet major status, cluster_version={0}'.format(min_cluster_version))# 5. 检查是否有异常租户(creating，延迟删除，恢复中)
def check_tenant_status(query_cur):# check tenant schema(desc, results) = query_cur.exec_query("""select count(*) as count from DBA_OB_TENANTS where status != 'NORMAL'""")if len(results) != 1 or len(results[0]) != 1:fail_list.append('results len not match')elif 0 != results[0][0]:fail_list.append('has abnormal tenant, should stop')else:logging.info('check tenant status success')# check tenant info# don't support restore tenant upgrade(desc, results) = query_cur.exec_query("""select count(*) as count from oceanbase.__all_virtual_tenant_info where tenant_role != 'PRIMARY' and tenant_role != 'STANDBY'""")if len(results) != 1 or len(results[0]) != 1:fail_list.append('results len not match')elif 0 != results[0][0]:fail_list.append('has abnormal tenant info, should stop')else:logging.info('check tenant info success')# 6. 检查无恢复任务
def check_restore_job_exist(query_cur):(desc, results) = query_cur.exec_query("""select count(1) from CDB_OB_RESTORE_PROGRESS""")if len(results) != 1 or len(results[0]) != 1:fail_list.append('failed to restore job cnt')elif results[0][0] != 0:fail_list.append("""still has restore job, upgrade is not allowed temporarily""")logging.info('check restore job success')def check_is_primary_zone_distributed(primary_zone_str):semicolon_pos = len(primary_zone_str)for i in range(len(primary_zone_str)):if primary_zone_str[i] == ';':semicolon_pos = ibreakcomma_pos = len(primary_zone_str)for j in range(len(primary_zone_str)):if primary_zone_str[j] == ',':comma_pos = jbreakif comma_pos < semicolon_pos:return Trueelse:return False# 7. 升级前需要primary zone只有一个
def check_tenant_primary_zone(query_cur):sql = """select distinct value from GV$OB_PARAMETERS  where name='min_observer_version'"""(desc, results) = query_cur.exec_query(sql)if len(results) != 1:fail_list.append('min_observer_version is not sync')elif len(results[0]) != 1:fail_list.append('column cnt not match')else:min_cluster_version = get_version(results[0][0])if min_cluster_version < get_version("4.1.0.0"):(desc, results) = query_cur.exec_query("""select tenant_name,primary_zone from DBA_OB_TENANTS where  tenant_id != 1""");for item in results:if cmp(item[1], "RANDOM") == 0:fail_list.append('{0} tenant primary zone random before update not allowed'.format(item[0]))elif check_is_primary_zone_distributed(item[1]):fail_list.append('{0} tenant primary zone distributed before update not allowed'.format(item[0]))logging.info('check tenant primary zone success')# 8. 修改永久下线的时间，避免升级过程中缺副本
def modify_server_permanent_offline_time(cur):set_parameter(cur, 'server_permanent_offline_time', '72h')# 9. 检查是否有DDL任务在执行
def check_ddl_task_execute(query_cur):(desc, results) = query_cur.exec_query("""select count(1) from __all_virtual_ddl_task_status""")if 0 != results[0][0]:fail_list.append("There are DDL task in progress")logging.info('check ddl task execut status success')# 10. 检查无备份任务
def check_backup_job_exist(query_cur):# Backup jobs cannot be in-progress during upgrade.(desc, results) = query_cur.exec_query("""select count(1) from CDB_OB_BACKUP_JOBS""")if len(results) != 1 or len(results[0]) != 1:fail_list.append('failed to backup job cnt')elif results[0][0] != 0:fail_list.append("""still has backup job, upgrade is not allowed temporarily""")else:logging.info('check backup job success')# 11. 检查无归档任务
def check_archive_job_exist(query_cur):min_cluster_version = 0sql = """select distinct value from GV$OB_PARAMETERS  where name='min_observer_version'"""(desc, results) = query_cur.exec_query(sql)if len(results) != 1:fail_list.append('min_observer_version is not sync')elif len(results[0]) != 1:fail_list.append('column cnt not match')else:min_cluster_version = get_version(results[0][0])# Archive jobs cannot be in-progress before upgrade from 4.0.if min_cluster_version < get_version("4.1.0.0"):(desc, results) = query_cur.exec_query("""select count(1) from CDB_OB_ARCHIVELOG where status!='STOP'""")if len(results) != 1 or len(results[0]) != 1:fail_list.append('failed to archive job cnt')elif results[0][0] != 0:fail_list.append("""still has archive job, upgrade is not allowed temporarily""")else:logging.info('check archive job success')# 12. 检查归档路径是否清空
def check_archive_dest_exist(query_cur):min_cluster_version = 0sql = """select distinct value from GV$OB_PARAMETERS  where name='min_observer_version'"""(desc, results) = query_cur.exec_query(sql)if len(results) != 1:fail_list.append('min_observer_version is not sync')elif len(results[0]) != 1:fail_list.append('column cnt not match')else:min_cluster_version = get_version(results[0][0])# archive dest need to be cleaned before upgrade from 4.0.if min_cluster_version < get_version("4.1.0.0"):(desc, results) = query_cur.exec_query("""select count(1) from CDB_OB_ARCHIVE_DEST""")if len(results) != 1 or len(results[0]) != 1:fail_list.append('failed to archive dest cnt')elif results[0][0] != 0:fail_list.append("""still has archive destination, upgrade is not allowed temporarily""")else:logging.info('check archive destination success')# 13. 检查备份路径是否清空
def check_backup_dest_exist(query_cur):min_cluster_version = 0sql = """select distinct value from GV$OB_PARAMETERS  where name='min_observer_version'"""(desc, results) = query_cur.exec_query(sql)if len(results) != 1:fail_list.append('min_observer_version is not sync')elif len(results[0]) != 1:fail_list.append('column cnt not match')else:min_cluster_version = get_version(results[0][0])# backup dest need to be cleaned before upgrade from 4.0.if min_cluster_version < get_version("4.1.0.0"):(desc, results) = query_cur.exec_query("""select count(1) from CDB_OB_BACKUP_PARAMETER where name='data_backup_dest' and (value!=NULL or value!='')""")if len(results) != 1 or len(results[0]) != 1:fail_list.append('failed to data backup dest cnt')elif results[0][0] != 0:fail_list.append("""still has backup destination, upgrade is not allowed temporarily""")else:logging.info('check backup destination success')def check_server_version(query_cur):sql = """select distinct(substring_index(build_version, '_', 1)) from __all_server""";(desc, results) = query_cur.exec_query(sql);if len(results) != 1:fail_list.append("servers build_version not match")else:logging.info("check server version success")# 14. 检查server是否可服务
def check_observer_status(query_cur):(desc, results) = query_cur.exec_query("""select count(*) from oceanbase.__all_server where (start_service_time <= 0 or status != "active")""")if results[0][0] > 0 :fail_list.append('{0} observer not available , please check'.format(results[0][0]))logging.info('check observer status success')# 15  检查schema是否刷新成功
def check_schema_status(query_cur):(desc, results) = query_cur.exec_query("""select if (a.cnt = b.cnt, 1, 0) as passed from (select count(*) as cnt from oceanbase.__all_virtual_server_schema_info where refreshed_schema_version > 1 and refreshed_schema_version % 8 = 0) as a join (select count(*) as cnt from oceanbase.__all_server join oceanbase.__all_tenant) as b""")if results[0][0] != 1 :fail_list.append('{0} schema not available, please check'.format(results[0][0]))logging.info('check schema status success')# 16. 检查是否存在名为all/all_user/all_meta的租户
def check_not_supported_tenant_name(query_cur):names = ["all", "all_user", "all_meta"](desc, results) = query_cur.exec_query("""select tenant_name from oceanbase.DBA_OB_TENANTS""")for i in range(len(results)):if results[i][0].lower() in names:fail_list.append('a tenant named all/all_user/all_meta (case insensitive) cannot exist in the cluster, please rename the tenant')breaklogging.info('check special tenant name success')# last check of do_check, make sure no function execute after check_fail_list
def check_fail_list():if len(fail_list) != 0 :error_msg ="upgrade checker failed with " + str(len(fail_list)) + " reasons: " + ", ".join(['['+x+"] " for x in fail_list])raise MyError(error_msg)def set_query_timeout(query_cur, timeout):if timeout != 0:sql = """set @@session.ob_query_timeout = {0}""".format(timeout * 1000 * 1000)query_cur.exec_sql(sql)# 开始升级前的检查
def do_check(my_host, my_port, my_user, my_passwd, timeout, upgrade_params):try:conn = mysql.connector.connect(user = my_user,password = my_passwd,host = my_host,port = my_port,database = 'oceanbase',raise_on_warnings = True)conn.autocommit = Truecur = conn.cursor(buffered=True)min_cluster_version = [0]try:query_cur = Cursor(cur)set_query_timeout(query_cur, timeout)check_observer_version(query_cur, upgrade_params)check_data_version(query_cur, min_cluster_version)check_paxos_replica(query_cur)check_rebalance_task(query_cur)check_cluster_status(query_cur, min_cluster_version)check_tenant_status(query_cur)check_restore_job_exist(query_cur)check_tenant_primary_zone(query_cur)check_ddl_task_execute(query_cur)check_backup_job_exist(query_cur)check_archive_job_exist(query_cur)check_archive_dest_exist(query_cur)check_backup_dest_exist(query_cur)check_observer_status(query_cur)check_schema_status(query_cur)check_server_version(query_cur)check_not_supported_tenant_name(query_cur)# all check func should execute before check_fail_listcheck_fail_list()modify_server_permanent_offline_time(cur)except Exception, e:logging.exception('run error')raise efinally:cur.close()conn.close()except mysql.connector.Error, e:logging.exception('connection error')raise eexcept Exception, e:logging.exception('normal error')raise eif __name__ == '__main__':upgrade_params = UpgradeParams()change_opt_defult_value('log-file', upgrade_params.log_filename)parse_options(sys.argv[1:])if not has_no_local_opts():deal_with_local_opts()else:check_db_client_opts()log_filename = get_opt_log_file()upgrade_params.log_filename = log_filename# 日志配置放在这里是为了前面的操作不要覆盖掉日志文件config_logging_module(upgrade_params.log_filename)try:host = get_opt_host()port = int(get_opt_port())user = get_opt_user()password = get_opt_password()timeout = int(get_opt_timeout())logging.info('parameters from cmd: host=\"%s\", port=%s, user=\"%s\", password=\"%s\", timeout=\"%s\", log-file=\"%s\"',\host, port, user, password, timeout, log_filename)do_check(host, port, user, password, timeout, upgrade_params)except mysql.connector.Error, e:logging.exception('mysql connctor error')raise eexcept Exception, e:logging.exception('normal error')raise e
[root@sdw1 etc]#

ZONE升级过程

由于采用的是滚动升级的方法。升级任务为了保障升级接断业务的稳定性，先进行一个ZONE升级，当这个ZONE升级完成后，再进行另外一个ZONE升级。即使一个ZONE正在进行升级或者升级失败，其它ZONE依然可以提供服务，保持业务的可持续性

每个ZONE升级任务阶段均是以下过程:

Set zone context (开始) -> Check zone(结束)

ZONE升级任务

ZONE升级前会关闭该ZONE的运行状态，并且停掉observer的进程，之后下载 升级目标版本的依赖(Install dependencies) 以及 OB安装包(Install OB rpm)

依赖和安装包下载完成后，开启observer进程(Wait observer version refreshed) 以及 observer版本的刷新(Wait observer version refreshed)

observer启动成功之后，会检查 clog的状态(Check clog stat) 和 observer版本更新的情况(Check observer refresh schema) 和 集群的检查(Execute upgrade health checker script)，之后开启ZONE

注意:当前zone开启成功后代表该zone已经升级成功，之后会重复以上步骤升级剩余的zone，所以升级任务的多少取决于zone以及租户的数量

查看业务情况

初始化数据完成，证明滚动升级不会对业务运行产生影响

总结

1.集群升级前会进行相应的集群检查工作，确保当前OB是否满足升级条件

2.升级任务量的多少取决于zone的多少

3.升级过程中会停止zone 与 observer ，该操作不会进行回滚，所以之前的检查任务如果出错必须查看原因，不可以进行跳过