作者:张乾

外星人2号,现兼任六位喵星人的资深铲屎官。

本文来源:原创投稿

*爱可生开源社区出品,原创内容未经授权不得随意使用,转载请联系小编并注明来源。


手滑误删了数据文件,并且没有可替换的节点时,先别急着提桶跑路,可以考虑利用参数 server_permanent_offline_time 来重建受影响的节点。

原理:

server_permanent_offline_time 是 OceanBase 数据库中用于控制节点永久下线时长的参数。当集群中的某个节点宕机后,系统会根据该参数的设置值来进行相应操作。

如果节点宕机时间小于该参数设置的值,系统会暂时不做处理,以避免频繁的数据迁移;如果宕机时间超过该参数设置的值,该节点被标记为永久下线,RootService 会将该 OBServer 上包含的数据副本从 Paxos 成员组中删除,并在同 zone 内其他可用 OBServer 上补充数据,以保证数据副本 Paxos 成员组完整。该参数默认值是 3600 秒,一般设置较大,以避免不必要的副本复制。此外,当永久下线的节点重新被拉起后,其上的全部数据都需要从其他副本重新拉取。

在本场景下,即是通过调低该参数,让故障节点快速永久下线再重新上线,达到数据重建的目的。

请注意,此过程会占用集群一定的资源,可能会影响性能,因此建议在业务低峰期进行。

官方建议

关于 server_permanent_offline_time 的适用场景和建议值,官方提供如下:

  1. OceanBase 数据库版本升级场景:建议将该配置项的值设置为72h。
  2. OBServer 硬件更换场景:建议将该配置项的值设置为4h。
  3. OBServer 清空上线场景:建议将该配置项的值设置为10m,使集群快速上线。

准备过程

预备一套环境

使用OBD工具快速部署一套3节点OB以及一个OBProxy,再创建好一个租户sysbench_tenant,primary_zone为RANDOM。

注:本文基于OB 3.1.2版本,其他版本需注意另作验证。

准备些数据

使用 sysbench 创建一个表 sbtest1 并插入1W数据。

sysbench ./oltp_insert.lua --mysql-host=10.186.60.3 --mysql-port=2883 --mysql-db=sysbenchdb --mysql-user="sysbench@sysbench_tenant"  --mysql-password=sysbench --tables=1 --table_size=10000 --threads=1 --time=600 --report-interval=10 --db-driver=mysql --db-ps-mode=disable --skip-trx=on --mysql-ignore-errors=6002,6004,4012,2013,4016,1062,5157,4038 prepare

这里改写了 sysbench 的建表语句,分了3个区,查询 sbtest1 表分区副本分布如下

MySQL [oceanbase]> select tenant.tenant_name, zone, svr_ip,svr_port, case when role=1 then 'leader' when role=2 then 'follower' else NULL end as role, count(1) as partition_cnt from __all_virtual_meta_table meta  inner join __all_tenant tenant  on meta.tenant_id=tenant.tenant_id inner join __all_virtual_table tab  on meta.tenant_id=tab.tenant_id and meta.table_id=tab.table_id where tenant.tenant_id=1001 and tab.table_name='sbtest1' group by  tenant.tenant_name,zone, svr_ip,svr_port, 5 order by  tenant.tenant_name, zone, svr_ip, role desc;
+-----------------+-------+--------------+----------+----------+---------------+
| tenant_name     | zone  | svr_ip       | svr_port | role     | partition_cnt |
+-----------------+-------+--------------+----------+----------+---------------+
| sysbench_tenant | zone1 | 10.186.64.74 |     2882 | leader   |             1 |
| sysbench_tenant | zone1 | 10.186.64.74 |     2882 | follower |             2 |
| sysbench_tenant | zone2 | 10.186.64.75 |     2882 | leader   |             1 |
| sysbench_tenant | zone2 | 10.186.64.75 |     2882 | follower |             2 |
| sysbench_tenant | zone3 | 10.186.64.79 |     2882 | leader   |             1 |
| sysbench_tenant | zone3 | 10.186.64.79 |     2882 | follower |             2 |
+-----------------+-------+--------------+----------+----------+---------------+

开始实验

使用 sysbench 持续写入数据,维持一定的流量,便于在节点重建后对比各节点数据是否一致。

sysbench ./oltp_insert.lua --mysql-host=10.186.60.3 --mysql-port=2883 --mysql-db=sysbenchdb --mysql-user="sysbench@sysbench_tenant"  --mysql-password=sysbench --tables=1 --table_size=10000 --threads=1 --time=300 --report-interval=10 --db-driver=mysql --db-ps-mode=disable --skip-trx=on --mysql-ignore-errors=6002,6004,4012,2013,4016,1062,5157,4038 run
删除某节点的数据文件

选择zone3下的10.186.64.79节点,将数据文件删除。

[root@localhost data]# rm -rf 1/sstable/block_file
[root@localhost data]# cd 1/sstable/
[root@localhost sstable]# ll
total 0
永久下线故障节点

1.调小参数 server_permanent_offline_time ,缩短节点永久下线时间

server_permanent_offline_time 默认值为3600s

MySQL [oceanbase]> alter system set server_permanent_offline_time='60s';
Query OK, 0 rows affected (0.030 sec)
 
MySQL [oceanbase]> SHOW PARAMETERS LIKE "%server_permanent_offline_time%";
+-------+----------+--------------+----------+-------------------------------+-----------+-------+-----------------------------------------------------------------------------------------------------------------------------------+--------------+---------+---------+-------------------+
| zone  | svr_type | svr_ip       | svr_port | name                          | data_type | value | info                                                                                                                              | section      | scope   | source  | edit_level        |
+-------+----------+--------------+----------+-------------------------------+-----------+-------+-----------------------------------------------------------------------------------------------------------------------------------+--------------+---------+---------+-------------------+
| zone3 | observer | 10.186.64.79 |     2882 | server_permanent_offline_time | NULL      | 60s   | the time interval between any two heartbeats beyond which a server is considered to be \'permanently\' offline. Range: [20s,+∞)   | ROOT_SERVICE | CLUSTER | DEFAULT | DYNAMIC_EFFECTIVE |
| zone1 | observer | 10.186.64.74 |     2882 | server_permanent_offline_time | NULL      | 60s   | the time interval between any two heartbeats beyond which a server is considered to be \'permanently\' offline. Range: [20s,+∞)   | ROOT_SERVICE | CLUSTER | DEFAULT | DYNAMIC_EFFECTIVE |
| zone2 | observer | 10.186.64.75 |     2882 | server_permanent_offline_time | NULL      | 60s   | the time interval between any two heartbeats beyond which a server is considered to be \'permanently\' offline. Range: [20s,+∞)   | ROOT_SERVICE | CLUSTER | DEFAULT | DYNAMIC_EFFECTIVE |
+-------+----------+--------------+----------+-------------------------------+-----------+-------+-----------------------------------------------------------------------------------------------------------------------------------+--------------+---------+---------+-------------------+

2.停止故障节点对外提供服务

在kill ob进程前,建议使用隔离(ISOLATE SERVER)或者停止(STOP SERVER)节点的命令,停掉发往该节点的请求,转移副本leader角色。在节点重建恢复后,再开启流量。

# 停掉79节点服务
MySQL [oceanbase]> ALTER SYSTEM STOP SERVER '10.186.64.79:2882' ZONE='zone3';
  
# 或者隔离
ALTER SYSTEM ISOLATE SERVER '10.186.64.79:2882' ZONE='zone3';

3.kill observer进程

执行kill -9 $observer_pid,等待 server_permanent_offline_time 的时间,该ob进入”永久下线”状态。判断ob是否已经永久下线,可以查询表 __all_rootservice_event_history,存在名为 “permanent_offline “的event记录,确认时间和ip都一致后,即可认为ob已经永久下线。

MySQL [oceanbase]> select * from __all_rootservice_event_history where event='permanent_offline' ;                       
+----------------------------+--------+-------------------+--------+---------------------+-------+--------+-------+--------+-------+--------+-------+--------+-------+--------+------------+--------------+-------------+
| gmt_create                 | module | event             | name1  | value1              | name2 | value2 | name3 | value3 | name4 | value4 | name5 | value5 | name6 | value6 | extra_info | rs_svr_ip    | rs_svr_port |
+----------------------------+--------+-------------------+--------+---------------------+-------+--------+-------+--------+-------+--------+-------+--------+-------+--------+------------+--------------+-------------+
| 2023-03-29 17:34:09.596035 | server | permanent_offline | server | "10.186.64.79:2882" |       |        |       |        |       |        |       |        |       |        |            | 10.186.64.74 |        2882 |
+----------------------------+--------+-------------------+--------+---------------------+-------+--------+-------+--------+-------+--------+-------+--------+-------+--------+------------+--------------+-------------+

查询分区副本分布如下,已不存在79节点的分区副本信息,进一步确认了79节点已永久下线。

zone2下的75节点有一个从副本升级为leader角色,此时集群仍然可以继续对外服务。

MySQL [oceanbase]> select tenant.tenant_name, zone, svr_ip,svr_port, case when role=1 then 'leader' when role=2 then 'follower' else NULL end as role, count(1) as partition_cnt from __all_virtual_meta_table meta  inner join __all_tenant tenant  on meta.tenant_id=tenant.tenant_id inner join __all_virtual_table tab  on meta.tenant_id=tab.tenant_id and meta.table_id=tab.table_id where tenant.tenant_id=1001 and tab.table_name='sbtest1' group by  tenant.tenant_name,zone, svr_ip,svr_port, 5 order by  tenant.tenant_name, zone, svr_ip, role desc;
+-----------------+-------+--------------+----------+----------+---------------+
| tenant_name     | zone  | svr_ip       | svr_port | role     | partition_cnt |
+-----------------+-------+--------------+----------+----------+---------------+
| sysbench_tenant | zone1 | 10.186.64.74 |     2882 | leader   |             1 |
| sysbench_tenant | zone1 | 10.186.64.74 |     2882 | follower |             2 |
| sysbench_tenant | zone2 | 10.186.64.75 |     2882 | leader   |             2 |
| sysbench_tenant | zone2 | 10.186.64.75 |     2882 | follower |             1 |
+-----------------+-------+--------------+----------+----------+---------------+
4 rows in set (0.005 sec)
拉起故障节点,触发数据自动重建

1.启动79节点的ob进程,进程启动后会自动触发重建。

注:防止ob启动失败或存在其他问题,建议启动前将数据文件和事务日志均清空。

[root@localhost data]# rm -rf log1/clog/*
[root@localhost data]# rm -rf log1/ilog/*
[root@localhost data]# rm -rf log1/slog/*
[root@localhost data]# rm -rf 1/sstable/block_file
[root@localhost data]# cd 1/sstable/
[root@localhost sstable]# ll
total 0
[root@localhost sstable]# su admin
bash-4.2$ cd /home/admin/ && ./bin/observer
./bin/observer

进程启动后,确认ob心跳恢复状态为active,然后查看分区正在不断补足中

MySQL [oceanbase]> select svr_ip,zone,with_rootserver,status,stop_time,start_service_time,build_version from __all_server;
+--------------+-------+-----------------+--------+-----------+--------------------+----------------------------------------------------------------------------------------+
| svr_ip       | zone  | with_rootserver | status | stop_time        | start_service_time | build_version                                                                          |
+--------------+-------+-----------------+--------+-----------+--------------------+----------------------------------------------------------------------------------------+
| 10.186.64.74 | zone1 |               1 | active |                0 |   1679984798650860 | 3.1.2_10000392021123010-d4ace121deae5b81d8f0b40afbc4c02705b7fc1d(Dec 30 2021 02:47:29) |
| 10.186.64.75 | zone2 |               0 | active |                0 |   1679984801289281 | 3.1.2_10000392021123010-d4ace121deae5b81d8f0b40afbc4c02705b7fc1d(Dec 30 2021 02:47:29) |
| 10.186.64.79 | zone3 |               0 | active | 1680082329964975 |   1680082511964975 | 3.1.2_10000392021123010-d4ace121deae5b81d8f0b40afbc4c02705b7fc1d(Dec 30 2021 02:47:29) |
+--------------+-------+-----------------+--------+-----------+--------------------+----------------------------------------------------------------------------------------+
3 rows in set (0.002 sec)
 
MySQL [oceanbase]> select count(*),zone from gv$partition group by zone;
+----------+-------+
| count(*) | zone  |
+----------+-------+
|     1322 | zone1 |
|     1322 | zone2 |
|      152 | zone3 |
+----------+-------+
3 rows in set (0.228 sec)
  
MySQL [oceanbase]> select count(*),zone from gv$partition group by zone;
+----------+-------+
| count(*) | zone  |
+----------+-------+
|     1322 | zone1 |
|     1322 | zone2 |
|      664 | zone3 |
+----------+-------+
3 rows in set (0.113 sec)
MySQL [oceanbase]> select count(*),zone from gv$partition group by zone;                                                
+----------+-------+
| count(*) | zone  |
+----------+-------+
|     1322 | zone1 |
|     1322 | zone2 |
|     1179 | zone3 |
+----------+-------+
3 rows in set (0.112 sec)
  
MySQL [oceanbase]> select count(*),zone from gv$partition group by zone;
+----------+-------+
| count(*) | zone  |
+----------+-------+
|     1322 | zone1 |
|     1322 | zone2 |
|     1322 | zone3 |
+----------+-------+
3 rows in set (0.116 sec)

当3个zone内的分区个数一致后,同时查看zone3已存在副本信息,认为重建完毕。

由于79节点处于隔离状态,所以还没有leader副本。

MySQL [oceanbase]> select tenant.tenant_name, zone, svr_ip,svr_port, case when role=1 then 'leader' when role=2 then 'follower' else NULL end as role, count(1) as partition_cnt from __all_virtual_meta_table meta  inner join __all_tenant tenant  on meta.tenant_id=tenant.tenant_id inner join __all_virtual_table tab  on meta.tenant_id=tab.tenant_id and meta.table_id=tab.table_id where tenant.tenant_id=1001 and tab.table_name='sbtest1' group by  tenant.tenant_name,zone, svr_ip,svr_port, 5 order by  tenant.tenant_name, zone, svr_ip, role desc;
+-----------------+-------+--------------+----------+----------+---------------+
| tenant_name     | zone  | svr_ip       | svr_port | role     | partition_cnt |
+-----------------+-------+--------------+----------+----------+---------------+
| sysbench_tenant | zone1 | 10.186.64.74 |     2882 | leader   |             1 |
| sysbench_tenant | zone1 | 10.186.64.74 |     2882 | follower |             2 |
| sysbench_tenant | zone2 | 10.186.64.75 |     2882 | leader   |             2 |
| sysbench_tenant | zone2 | 10.186.64.75 |     2882 | follower |             1 |
| sysbench_tenant | zone3 | 10.186.64.79 |     2882 | follower |             3 |
+-----------------+-------+--------------+----------+----------+---------------+
6 rows in set (0.005 sec)

2.开启故障节点服务

执行命令解除79节点的隔离状态。

ALTER SYSTEM START SERVER '10.186.64.79:2882' ZONE='zone3';

查询分区副本分布如下,leader角色已迁回79节点。

MySQL [oceanbase]> select tenant.tenant_name, zone, svr_ip,svr_port, case when role=1 then 'leader' when role=2 then 'follower' else NULL end as role, count(1) as partition_cnt from __all_virtual_meta_table meta  inner join __all_tenant tenant  on meta.tenant_id=tenant.tenant_id inner join __all_virtual_table tab  on meta.tenant_id=tab.tenant_id and meta.table_id=tab.table_id where tenant.tenant_id=1001 and tab.table_name='sbtest1' group by  tenant.tenant_name,zone, svr_ip,svr_port, 5 order by  tenant.tenant_name, zone, svr_ip, role desc;
+-----------------+-------+--------------+----------+----------+---------------+
| tenant_name     | zone  | svr_ip       | svr_port | role     | partition_cnt |
+-----------------+-------+--------------+----------+----------+---------------+
| sysbench_tenant | zone1 | 10.186.64.74 |     2882 | leader   |             1 |
| sysbench_tenant | zone1 | 10.186.64.74 |     2882 | follower |             2 |
| sysbench_tenant | zone2 | 10.186.64.75 |     2882 | leader   |             1 |
| sysbench_tenant | zone2 | 10.186.64.75 |     2882 | follower |             2 |
| sysbench_tenant | zone3 | 10.186.64.79 |     2882 | leader   |             1 |
| sysbench_tenant | zone3 | 10.186.64.79 |     2882 | follower |             2 |
+-----------------+-------+--------------+----------+----------+---------------+

3.把server_permanent_offline_time参数的预知重新设置为默认的3600s

MySQL [oceanbase]> alter system set server_permanent_offline_time='3600s';
Query OK, 0 rows affected (0.028 sec)
 
MySQL [oceanbase]> SHOW PARAMETERS LIKE "%server_permanent_offline_time%";
+-------+----------+--------------+----------+-------------------------------+-----------+-------+-----------------------------------------------------------------------------------------------------------------------------------+--------------+---------+---------+-------------------+
| zone  | svr_type | svr_ip       | svr_port | name                          | data_type | value | info                                                                                                                              | section      | scope   | source  | edit_level        |
+-------+----------+--------------+----------+-------------------------------+-----------+-------+-----------------------------------------------------------------------------------------------------------------------------------+--------------+---------+---------+-------------------+
| zone2 | observer | 10.186.64.75 |     2882 | server_permanent_offline_time | NULL      | 3600s | the time interval between any two heartbeats beyond which a server is considered to be \'permanently\' offline. Range: [20s,+∞)   | ROOT_SERVICE | CLUSTER | DEFAULT | DYNAMIC_EFFECTIVE |
| zone1 | observer | 10.186.64.74 |     2882 | server_permanent_offline_time | NULL      | 3600s | the time interval between any two heartbeats beyond which a server is considered to be \'permanently\' offline. Range: [20s,+∞)   | ROOT_SERVICE | CLUSTER | DEFAULT | DYNAMIC_EFFECTIVE |
| zone3 | observer | 10.186.64.79 |     2882 | server_permanent_offline_time | NULL      | 3600s | the time interval between any two heartbeats beyond which a server is considered to be \'permanently\' offline. Range: [20s,+∞)   | ROOT_SERVICE | CLUSTER | DEFAULT | DYNAMIC_EFFECTIVE |
+-------+----------+--------------+----------+-------------------------------+-----------+-------+-----------------------------------------------------------------------------------------------------------------------------------+--------------+---------+---------+-------------------+
3 rows in set (0.007 sec)
校验各ob节点数据量

sysbench已运行结束,直连各observer,校验数据量是一致的。

[root@localhost ~]#  obclient -h10.186.64.74 -P2881 -usysbench@sysbench_tenant -Dsysbenchdb -A -psysbench
Welcome to the OceanBase.  Commands end with ; or \g.
Your MySQL connection id is 3221545401
Server version: 5.7.25 OceanBase 3.1.2 (r10000392021123010-d4ace121deae5b81d8f0b40afbc4c02705b7fc1d) (Built Dec 30 2021 02:47:29)
 
Copyright (c) 2000, 2018, Oracle, MariaDB Corporation Ab and others.
 
Type 'help;' or '\h' for help. Type '\c' to clear the current input statement.
 
MySQL [sysbenchdb]> select count(*) from sbtest1;
+----------+
| count(*) |
+----------+
|    53195 |
+----------+
1 row in set (0.036 sec)
 
MySQL [sysbenchdb]> exit
Bye
[root@localhost ~]#  obclient -h10.186.64.75 -P2881 -usysbench@sysbench_tenant -Dsysbenchdb -A -psysbench
Welcome to the OceanBase.  Commands end with ; or \g.
Your MySQL connection id is 3221823448
Server version: 5.7.25 OceanBase 3.1.2 (r10000392021123010-d4ace121deae5b81d8f0b40afbc4c02705b7fc1d) (Built Dec 30 2021 02:47:29)
 
Copyright (c) 2000, 2018, Oracle, MariaDB Corporation Ab and others.
 
Type 'help;' or '\h' for help. Type '\c' to clear the current input statement.
 
MySQL [sysbenchdb]> select count(*) from sbtest1;
+----------+
| count(*) |
+----------+
|    53195 |
+----------+
1 row in set (0.040 sec)
 
MySQL [sysbenchdb]> exit
Bye
[root@localhost ~]#  obclient -h10.186.64.79 -P2881 -usysbench@sysbench_tenant -Dsysbenchdb -A -psysbench
Welcome to the OceanBase.  Commands end with ; or \g.
Your MySQL connection id is 3222011907
Server version: 5.7.25 OceanBase 3.1.2 (r10000392021123010-d4ace121deae5b81d8f0b40afbc4c02705b7fc1d) (Built Dec 30 2021 02:47:29)
 
Copyright (c) 2000, 2018, Oracle, MariaDB Corporation Ab and others.
 
Type 'help;' or '\h' for help. Type '\c' to clear the current input statement.
 
MySQL [sysbenchdb]> select count(*) from sbtest1;
+----------+
| count(*) |
+----------+
|    53195 |
+----------+
1 row in set (0.037 sec)
 
MySQL [sysbenchdb]>

总结

数据文件损坏或者丢失时,可通过调整参数 server_permanent_offline_time 来重建受影响的节点。

  1. 设小 server_permanent_offline_time 阈值
  2. 停止故障节点对外服务
  3. 终止该节点进程。
  4. 超过阈值后,节点将被标记为永久下线,系统会自动清空副本以及向同zone内其他节点迁移数据。
  5. 启动 OB 进程,自动触发重建节点数据。
  6. 开启故障节点服务。
  7. 把server_permanent_offline_time参数改回原来的值

avatar
100
  Subscribe  
提醒