暂无图片
暂无图片
暂无图片
暂无图片
暂无图片
首页 / OCEANBASE分区切主对事务的影响

OCEANBASE分区切主对事务的影响

原创 范计杰 云和恩墨技术服务团队 2024-07-03
308

概述

OCEANBASE高可用可以做到分区级，相当多的操作都是分区级的，在某些情况主分区切主不影响事务的继续运行，下面测试验证什么情况下切主事务可以继续，无感知。

相关参数

trx_force_kill_threshold 用于设置因冻结或切主需要杀事务时的最长等待时间。

属性 描述
参数类型 时间类型
默认值 100ms
取值范围 [1ms, 10s]
是否重启 OBServer 生效 否

OceanBase 数据库 V3.2 版本对于分布式事务中的 Leader 切换做了优化，尽量避免 Leader 切换导致的事务回滚。优化内容为：

  • 在进行有主改选时，如果旧的 Leader 上正在执行的事务日志能在规定的时间内（trx_force_kill_threshold, default=100ms）把同步到新 Leader 上，则新 Leader 上任后可以继续推进事务的执行。
  • 如果不能在规定的时间内完成事务日志的同步，则需要回滚正在执行的事务。

该机制是否只能用在原主副本可用时，计划内切主，或内部操作预期内切主的场景？

测试切主是否对小事务无影响

  1. 创建测试表

    ---用户租户(oracle模式)
create table tabha(id number,c varchar2(100));
insert into tabha(id,c) select rownum,'test'||rownum from dual connect by rownum<=10000;
commit;
create index idx_tabha on tab1(id);
--sys租户
MySQL [oceanbase]> alter system major freeze;
Query OK, 0 rows affected (0.016 sec)

  2. 修改查询，事务超时时间为7200,避免操作过程中超时

    set ob_query_timeout=7200000000;
set ob_trx_idle_timeout=7200000000;
set ob_trx_lock_timeout=7200000000;
set ob_trx_timeout=7200000000;

obclient [SYS]> show variables like '%timeout%';
+---------------------+------------------+
| VARIABLE_NAME       | VALUE            |
+---------------------+------------------+
| connect_timeout     | 10               |
| interactive_timeout | 28800            |
| net_read_timeout    | 30               |
| net_write_timeout   | 60               |
| ob_pl_block_timeout | 3216672000000000 |
| ob_query_timeout    | 7200000000       |
| ob_trx_idle_timeout | 7200000000       |
| ob_trx_lock_timeout | 7200000000       |
| ob_trx_timeout      | 7200000000       |
| wait_timeout        | 259200           |
+---------------------+------------------+
10 rows in set (0.004 sec)

  3. 执行一条UPDATE，不提交

    update tabha set c=c where id=100;


obclient [SYS]> update tabha set c=c where id=100;
Query OK, 1 row affected (0.005 sec)
Rows matched: 1  Changed: 1  Warnings: 0

  4. 查询tabha 主副本位置

    select * from __all_virtual_meta_table where table_id in (select table_id from __all_virtual_table where table_name='tabha') and role=1\G

MySQL [oceanbase]> select * from __all_virtual_meta_table where table_id in (select table_id from __all_virtual_table where table_name='tabha') and role=1\G
*************************** 1. row ***************************
         tenant_id: 1001
          table_id: 1100611139454075
      partition_id: 0
            svr_ip: 192.168.56.36
          svr_port: 2882
        gmt_create: 2024-06-05 10:45:56.786010
      gmt_modified: 2024-06-05 10:47:14.871100
          sql_port: 2881
           unit_id: 1002
     partition_cnt: 0
              zone: zone1
              role: 1
       member_list: 192.168.56.36:2882:1717555556783731,192.168.56.37:2882:1717555556783731,192.168.56.38:2882:1717555556783731
         row_count: 10000
         data_size: 52933
      data_version: 212
     data_checksum: 263436233
      row_checksum: 0
   column_checksum:
is_original_leader: 0
is_previous_leader: 1717555556815486
       create_time: 0
           rebuild: 0
      replica_type: 0
     required_size: 2097152
            status: REPLICA_STATUS_NORMAL
        is_restore: 0
partition_checksum: 0
            quorum: 3
         fail_list:
recovery_timestamp: 0
  memstore_percent: 100
      data_file_id: 1700968058986177
1 row in set (0.037 sec)

  5. 对表tabha进行手工切主操作(SWITCH REPLICA LEADER)

    MySQL [oceanbase]> select *from __all_virtual_election_info where table_id=1100611139454075 and role=1\G
*************************** 1. row ***************************
                     svr_ip: 192.168.56.36
                   svr_port: 2882
                   table_id: 1100611139454075
              partition_idx: 0
                 is_running: 1
         is_changing_leader: 0
             current_leader: "192.168.56.36:2882"
            previous_leader: 0.0.0.0
            proposal_leader: 0.0.0.0
                member_list: 3{server:"192.168.56.36:2882", timestamp:1717555556783731, flag:0, need_encrypt:false}{server:"192.168.56.37:2882", timestamp:1717555556783731, flag:0, need_encrypt:false}{server:"192.168.56.38:2882", timestamp:1717555556783731, flag:0, need_encrypt:false}
                replica_num: 3
                lease_start: 1717555561250000
                  lease_end: 1717555571450000
                time_offset: 4050000
           active_timestamp: 1717555556791050
               T1_timestamp: 1717555557200000
               leader_epoch: 1717555551600000
                      state: 1
                       role: 1
                      stage: 1
                      eg_id: 6565001008975067605
remaining_time_in_blacklist: 0
1 row in set (0.048 sec)


ALTER SYSTEM SWITCH REPLICA LEADER PARTITION_ID  '0%0@1100611139454075' SERVER  '192.168.56.36:2882' ;



MySQL [oceanbase]> ALTER SYSTEM SWITCH REPLICA LEADER PARTITION_ID  '0%0@1100611139454075' SERVER  '192.168.56.36:2882' ;
Query OK, 0 rows affected (0.006 sec)


MySQL [oceanbase]> select *from __all_virtual_election_info where table_id=1100611139454075 and role=1\G
*************************** 1. row ***************************
                     svr_ip: 192.168.56.36
                   svr_port: 2882
                   table_id: 1100611139454075
              partition_idx: 0
                 is_running: 1
         is_changing_leader: 0
             current_leader: "192.168.56.36:2882"
            previous_leader: 0.0.0.0
            proposal_leader: 0.0.0.0
                member_list: 3{server:"192.168.56.36:2882", timestamp:1717555556783731, flag:0, need_encrypt:false}{server:"192.168.56.37:2882", timestamp:1717555556783731, flag:0, need_encrypt:false}{server:"192.168.56.38:2882", timestamp:1717555556783731, flag:0, need_encrypt:false}
                replica_num: 3
                lease_start: 1717555561250000
                  lease_end: 1717555571450000
                time_offset: 4050000
           active_timestamp: 1717555556791050
               T1_timestamp: 1717555557200000
               leader_epoch: 1717555551600000
                      state: 1
                       role: 1
                      stage: 1
                      eg_id: 6565001008975067605
remaining_time_in_blacklist: 0
1 row in set (0.047 sec)

  6. 事务继续update仍然不会中断

    obclient [SYS]> update tabha set c=c where id=101;
Query OK, 1 row affected (0.005 sec)
Rows matched: 1  Changed: 1  Warnings: 0

  7. primary_zone从’zone1;zone2;zone3’改为’zone2;zone1;zone3’，发生主副本所在节点变化

    alter table tabha set primary_zone='zone2;zone1;zone3';


obclient [SYS]> alter table tabha set primary_zone='zone2;zone1;zone3';
Query OK, 0 rows affected (0.111 sec)

MySQL [oceanbase]> select *from __all_virtual_election_info where table_id=1100611139454075 and role=1\G
*************************** 1. row ***************************
                     svr_ip: 192.168.56.37
                   svr_port: 2882
                   table_id: 1100611139454075
              partition_idx: 0
                 is_running: 1
         is_changing_leader: 0
             current_leader: "192.168.56.37:2882" <<<
            previous_leader: 192.168.56.36 <<<
            proposal_leader: 0.0.0.0
                member_list: 3{server:"192.168.56.36:2882", timestamp:1717555556783731, flag:0, need_encrypt:false}{server:"192.168.56.37:2882", timestamp:1717555556783731, flag:0, need_encrypt:false}{server:"192.168.56.38:2882", timestamp:1717555556783731, flag:0, need_encrypt:false}
                replica_num: 3
                lease_start: 1717556607050000
                  lease_end: 1717556617250000
                time_offset: 4050000
           active_timestamp: 1717555556790992
               T1_timestamp: 1717556603000000
               leader_epoch: 1717556597400000
                      state: 1
                       role: 1
                      stage: 1
                      eg_id: 11361271041241770193
remaining_time_in_blacklist: 0
1 row in set (0.047 sec)

  8. 事务继续update仍然不会中断

    obclient [SYS]> update tabha set c=c where id=102;
Query OK, 1 row affected (0.018 sec)
Rows matched: 1  Changed: 1  Warnings: 0

  9. 进行大事务测试，大事务切主后不能继续

    insert 300万行数据
obclient [SYS]> insert into tabha select rownum,'test'||rownum from dual  connect by rownum<=3000000;
Query OK, 3000000 rows affected (49.525 sec)
Records: 3000000  Duplicates: 0  Warnings: 0

切主
MySQL [oceanbase]> ALTER SYSTEM SWITCH REPLICA LEADER PARTITION_ID  '0%0@1100611139454075' SERVER  '192.168.56.36:2882' ;
Query OK, 0 rows affected (0.004 sec)

继续insert会提示事务需要rollback;
obclient [SYS]> insert into tabha select rownum,'test'||rownum from dual  connect by rownum<=3000000;
ORA-24761: transaction rolled back: transaction needs rollback

obclient [SYS]> rollback;
Query OK, 0 rows affected (0.001 sec)

  10. 正在执行中的DML会失败

    obclient [SYS]> update tabha set c=c where id=1 and (select  avg(dbms_random.random()) from dual connect by rownum<100000)<100;



MySQL [oceanbase]> ALTER SYSTEM SWITCH REPLICA LEADER PARTITION_ID  '0%0@1100611139454075' SERVER  '192.168.56.36:2882' ;
Query OK, 0 rows affected (0.004 sec)

obclient [SYS]> update tabha set c=c where id=1 and (select  avg(dbms_random.random()) from dual connect by rownum<100000)<100;
ORA-24761: transaction rolled back: transaction context does not exist

测试结论

1、小事务未提交(初步定义为日志传输小于100ms)，涉及的分区发生切主，事务可以继续

2、大事务，涉及的分区发生切主，事务需要回滚

3、正在进行中的DML，涉及的分区发生切主，会中断，事务需要回滚

墨力计划oceanbase
「喜欢这篇文章，您的关注和赞赏是给作者最好的鼓励」
关注作者
【版权声明】本文为墨天轮用户原创内容，转载时必须标注文章的来源（墨天轮），文章链接，文章作者等基本信息，否则作者和墨天轮有权追究责任。如果您发现墨天轮中有涉嫌抄袭或者侵权的内容，欢迎发送邮件至：contact@modb.pro进行举报，并提供相关证据，一经查实，墨天轮将立刻删除相关内容。

评论