查看原文
其他

故障解析丨Clone节点导致主从故障

王权富贵 老叶茶馆 2024-07-08
* GreatSQL使

故障解析丨Clone节点导致主从故障

1.背景概述

在一次主从复制架构中,由于主节点binlog损坏,导致从节点无法正常同步数据,只能重做从节点;因此使用MySQL 8.0.17开始提供的clone技术进行恢复,恢复后的2天都发生了主从报错数据冲突。

通过解析binlog发现,同一时刻主从节点都在执行同一条语句,因此询问业务是否在主从节点都执行了定时任务,业务回复定时任务只在主节点执行。

最后排查发现,克隆后的从节点的定时任务也会是开启的状态,因此同一时刻,主从节点同时执行定时任务,导致主从报错,最终将从节点的定时任务关闭后解决此问题。

2.问题复现

本次测试基于 GreatSQL 8.0.32-24

greatsql> SELECT VERSION();
+-----------+
| VERSION() |
+-----------+
| 8.0.32-24 |
+-----------+
1 row in set (0.00 sec)

1.搭建一套主从架构

2.创建event

greatsql> create database test;
greatsql> use test;
greatsql> CREATE TABLE `test` (
 `id` int(11unsigned NOT NULL AUTO_INCREMENT COMMENT 'ID',
 `now` datetime DEFAULT NULL COMMENT '时间',
 PRIMARY KEY (`id`)
);

greatsql> CREATE EVENT event_test 
ON SCHEDULE EVERY 1 MINUTE
ON COMPLETION PRESERVE 
ENABLE 
COMMENT '每隔1分钟向test表插入记录'
DO INSERT INTO test VALUES(NULLnow());

3.查看event状态

主节点,默认情况下event状态为 ENABLED

greatsql> show events;
+------+------------+---------+-----------+-----------+------------+----------------+----------------+---------------------+------+---------+------------+----------------------+----------------------+--------------------+
| Db  | Name    | Definer | Time zone | Type    | Execute at | Interval value | Interval field | Starts        | Ends | Status  | Originator | character_set_client | collation_connection | Database Collation |
+------+------------+---------+-----------+-----------+------------+----------------+----------------+---------------------+------+---------+------------+----------------------+----------------------+--------------------+
test | event_test | root@%  | SYSTEM   | RECURRING | NULL    | 1        | MINUTE     | 2023-10-12 17:11:14 | NULL | ENABLED |      1 | utf8mb4        | utf8mb4_unicode_ci  | utf8mb4_unicode_ci |
+------+------------+---------+-----------+-----------+------------+----------------+----------------+---------------------+------+---------+------------+----------------------+----------------------+--------------------+
1 row in set (0.00 sec)

从节点,默认情况下event状态为 SLAVESIDE_DISABLED

greatsql> show events;
+------+------------+---------+-----------+-----------+------------+----------------+----------------+---------------------+------+--------------------+------------+----------------------+----------------------+--------------------+
| Db  | Name    | Definer | Time zone | Type    | Execute at | Interval value | Interval field | Starts        | Ends | Status       | Originator | character_set_client | collation_connection | Database Collation |
+------+------------+---------+-----------+-----------+------------+----------------+----------------+---------------------+------+--------------------+------------+----------------------+----------------------+--------------------+
test | event_test | root@%  | SYSTEM   | RECURRING | NULL    | 1        | MINUTE     | 2023-10-12 17:11:14 | NULL | SLAVESIDE_DISABLED |      1 | utf8mb4        | utf8mb4_unicode_ci  | utf8mb4_unicode_ci |
+------+------------+---------+-----------+-----------+------------+----------------+----------------+---------------------+------+--------------------+------------+----------------------+----------------------+--------------------+
1 row in set (0.00 sec)

4.查看数据

greatsql> select * from test.test;
+----+---------------------+
| id | now         |
+----+---------------------+
|  1 | 2023-08-08 16:00:39 |
|  2 | 2023-08-08 16:01:39 |
|  3 | 2023-08-08 16:02:39 |
+----+---------------------+
3 rows in set (0.00 sec)

5.从节点进行克隆

# 安装克隆插件,主从节点都需要

greatsql> install plugin clone soname 'mysql_clone.so';

# 从节点进行clone

greatsql> set global clone_valid_donor_list='172.17.137.162:6001';

greatsql> clone instance from root@'172.17.137.162':6001 identified by 'greatsql';

6.重新建立主从复制

greatsql> change master to master_user='root',master_password='greatsql',master_host='172.17.137.162',master_port=6001,master_auto_position=1;
Query OK, 0 rows affected, 7 warnings (0.04 sec)

greatsql> start slave;
Query OK, 0 rows affected, 1 warning (0.04 sec)

7.查看主从状态

greatsql> show slave status\G
*************************** 1. row ***************************
               Slave_IO_State: Waiting for source to send event
                  Master_Host: 172.17.137.162
                  Master_User: root
                  Master_Port: 
6001
                Connect_Retry: 
60
              Master_Log_File: 
binlog.000001
          Read_Master_Log_Pos: 
2959
               Relay_Log_File: 
relaylog.000002
                Relay_Log_Pos: 
395
        Relay_Master_Log_File: 
binlog.000001
             Slave_IO_Running: Yes
            Slave_SQL_Running: 
No
              Replicate_Do_DB: 
          Replicate_Ignore_DB: 
           Replicate_Do_Table: 
       Replicate_Ignore_Table: 
      Replicate_Wild_Do_Table: 
  Replicate_Wild_Ignore_Table: 
                   Last_Errno: 
1062
                   Last_Error: Coordinator stopped because there were 
error(s) in the worker(s). The most recent failure being: Worker 1 failed executing transaction 'e8bf88f9-2acd-11ee-a98a-00163e605c74:8' at master log binlog.000001, end_log_pos 2606. See error log and/or performance_schema.replication_applier_status_by_worker table for more details about this failure or others, if any.
                 Skip_Counter: 
0
          Exec_Master_Log_Pos: 
2307
              Relay_Log_Space: 
1242
              Until_Condition: 
None
               Until_Log_File: 
greatsql> 
select * from performance_schema.replication_applier_status_by_worker limit 1\G
*************************** 
1. row ***************************
                                           CHANNEL_NAME: 
                                              WORKER_ID: 
1
                                              THREAD_ID: 
NULL
                                          SERVICE_STATE: 
OFF
                                      LAST_ERROR_NUMBER: 
1062
                                     LAST_ERROR_MESSAGE: Worker 
1 failed executing transaction 'e8bf88f9-2acd-11ee-a98a-00163e605c74:8' at master log binlog.000001, end_log_pos 2606; Could not execute Write_rows event on table test.test; Duplicate entry '5' for key 'test.PRIMARY', Error_code: 1062; handler error HA_ERR_FOUND_DUPP_KEY; the event's master log FIRST, end_log_pos 2606
                                   LAST_ERROR_TIMESTAMP: 2023-08-08 16:03:39.033240
                               LAST_APPLIED_TRANSACTION: 
     LAST_APPLIED_TRANSACTION_ORIGINAL_COMMIT_TIMESTAMP: 0000-00-00 00:00:00.000000
    LAST_APPLIED_TRANSACTION_IMMEDIATE_COMMIT_TIMESTAMP: 0000-00-00 00:00:00.000000
         LAST_APPLIED_TRANSACTION_START_APPLY_TIMESTAMP: 0000-00-00 00:00:00.000000
           LAST_APPLIED_TRANSACTION_END_APPLY_TIMESTAMP: 0000-00-00 00:00:00.000000
                                   APPLYING_TRANSACTION: e8bf88f9-2acd-11ee-a98a-00163e605c74:8
         APPLYING_TRANSACTION_ORIGINAL_COMMIT_TIMESTAMP: 2023-08-08 16:02:45.795753
        APPLYING_TRANSACTION_IMMEDIATE_COMMIT_TIMESTAMP: 2023-08-08 16:02:45.795753
             APPLYING_TRANSACTION_START_APPLY_TIMESTAMP: 2023-08-08 16:03:39.032510
                 LAST_APPLIED_TRANSACTION_RETRIES_COUNT: 0
   LAST_APPLIED_TRANSACTION_LAST_TRANSIENT_ERROR_NUMBER: 0
  LAST_APPLIED_TRANSACTION_LAST_TRANSIENT_ERROR_MESSAGE: 
LAST_APPLIED_TRANSACTION_LAST_TRANSIENT_ERROR_TIMESTAMP: 0000-00-00 00:00:00.000000
                     APPLYING_TRANSACTION_RETRIES_COUNT: 0
       APPLYING_TRANSACTION_LAST_TRANSIENT_ERROR_NUMBER: 0
      APPLYING_TRANSACTION_LAST_TRANSIENT_ERROR_MESSAGE: 
    APPLYING_TRANSACTION_LAST_TRANSIENT_ERROR_TIMESTAMP: 0000-00-00 00:00:00.000000
1 row in 
set (0.00 sec)

可以看到从节点报错发生了主键冲突。

8.查看从节点定时任务状态

当前从节点定时任务状态为 ENABLED

greatsql> show events;
+------+------------+---------+-----------+-----------+------------+----------------+----------------+---------------------+------+---------+------------+----------------------+----------------------+--------------------+
| Db  | Name    | Definer | Time zone | Type    | Execute at | Interval value | Interval field | Starts        | Ends | Status  | Originator | character_set_client | collation_connection | Database Collation |
+------+------------+---------+-----------+-----------+------------+----------------+----------------+---------------------+------+---------+------------+----------------------+----------------------+--------------------+
test | event_test | root@%  | SYSTEM   | RECURRING | NULL    | 1        | MINUTE     | 2023-08-08 15:58:45 | NULL | ENABLED |      1 | utf8mb4        | utf8mb4_unicode_ci  | utf8mb4_unicode_ci |
+------+------------+---------+-----------+-----------+------------+----------------+----------------+---------------------+------+---------+------------+---------------------+----------------------+--------------------+
1 row in set (0.00 sec)

可以看到由于从节点的定时任务也执行了,从节点写入数据,导致主键冲突。

9.故障解决

greatsql> alter event event_test DISABLE;
Query OK, 0 rows affected (0.01 sec)

关闭从节点的定时任务event,然后跳过主键冲突的报错,最后重新启动主从复制。

3.总结

1.如果主库有定时任务,通过clone的方式搭建从库,在从库恢复之后需要关闭定时任务,避免主从同时执行定时任务导致主从故障。

2.克隆时,如果捐赠节点有主从复制信息,则克隆后的接收节点也会克隆此复制信息,并在克隆完成自动重启实例后,自动启动复制;避免此问题可以在接收节点的配置文件中增加 skip-slave-start,避免节点重启后自动启动复制。


Enjoy GreatSQL :)



《深入浅出MGR》视频课程

戳此小程序即可直达B站






https://www.bilibili.com/medialist/play/1363850082?business=space_collection&business_id=343928&desc=0



文章推荐:





由阿里通义万相生成
提示词:MySQL Clone

想看更多技术好文,点在看”吧!

继续滑动看下一个
向上滑动看下一个

您可能也对以下帖子感兴趣

文章有问题?点此查看未经处理的缓存