关于数据库:不同于OracleMySQL的insert会阻塞update

67次阅读

共计 9143 个字符，预计需要花费 23 分钟才能阅读完成。

  • GreatSQL 社区原创内容未经受权不得随便应用，转载请分割小编并注明起源。
  • GreatSQL 是 MySQL 的国产分支版本，应用上与 MySQL 统一。
  • 作者：王庆勋
  • 文章起源：GreatSQL 社区原创

某银行客户在从 Oracle 迁徙到 MySQL 的开发中，MySQL 在 READ-COMMITTED 隔离级别下，呈现了 insert 阻塞 update 的状况，但同样的状况下，Oracle 的 insert 则不会阻塞 update。本文通过复现该问题，剖析 MySQL 的锁信息，确认是 MySQL 与 Oracle 在并发事务处理上的差别，在进行数据库迁徙革新的程序开发应予以关注。

1. 问题复现

1.1. 环境筹备

MySQL 版本的 8.0.26，隔离级别是READ-COMMITTED，测试表 t 的字段 a 为主键。

mysql> select version();
+-----------+
| version() |
+-----------+
| 8.0.26    |
+-----------+
1 row in set (0.02 sec)

mysql> show variables like 'transaction_isolation';
+-----------------------+----------------+
| Variable_name         | Value          |
+-----------------------+----------------+
| transaction_isolation | READ-COMMITTED |
+-----------------------+----------------+
1 row in set (0.00 sec)

mysql> desc t;
+-------+------+------+-----+---------+-------+
| Field | Type | Null | Key | Default | Extra |
+-------+------+------+-----+---------+-------+
| a     | int  | NO   | PRI | NULL    |       |
| b     | int  | YES  |     | NULL    |       |
+-------+------+------+-----+---------+-------+
2 rows in set (0.01 sec)

mysql> select * from t;
+---+------+
| a | b    |
+---+------+
| 7 |    7 |
+---+------+
1 row in set (0.00 sec)

1.2. insert 阻塞 update 的操作步骤

insert 语句未提交时，update 同样主键的数据会被阻塞。

session1 session2
插入一条数据 (a=8) 后未提交。mysql> begin; Query OK, 0 rows affected (0.00 sec) mysql> insert into t values(8,8); Query OK, 1 row affected (0.01 sec)
更改数据，条件是 a =8, 将会被阻塞 mysql> update t set b=0 where a=8; << 挂起，期待 innodb_lock_wait_timeout 超时

2. 剖析起因

2.1. 查看事务锁信息

mysql> select * from information_schema.innodb_trx\G
*************************** 1. row ***************************
                    trx_id: 3795
                 trx_state: LOCK WAIT
               trx_started: 2022-10-11 16:03:38
     trx_requested_lock_id: 139727275779216:52:4:3:139724882995456
          trx_wait_started: 2022-10-11 16:03:38
                trx_weight: 2
       trx_mysql_thread_id: 9346
                 trx_query: update t set b=0 where a=8
       trx_operation_state: starting index read
         trx_tables_in_use: 1
         trx_tables_locked: 1
          trx_lock_structs: 2
     trx_lock_memory_bytes: 1128
           trx_rows_locked: 1
         trx_rows_modified: 0
   trx_concurrency_tickets: 0
       trx_isolation_level: READ COMMITTED
         trx_unique_checks: 1
    trx_foreign_key_checks: 1
trx_last_foreign_key_error: NULL
 trx_adaptive_hash_latched: 0
 trx_adaptive_hash_timeout: 0
          trx_is_read_only: 0
trx_autocommit_non_locking: 0
       trx_schedule_weight: 1
*************************** 2. row ***************************
                    trx_id: 3790
                 trx_state: RUNNING
               trx_started: 2022-10-11 16:03:29
     trx_requested_lock_id: NULL
          trx_wait_started: NULL
                trx_weight: 3
       trx_mysql_thread_id: 9320
                 trx_query: NULL
       trx_operation_state: NULL
         trx_tables_in_use: 0
         trx_tables_locked: 1
          trx_lock_structs: 2
     trx_lock_memory_bytes: 1128
           trx_rows_locked: 1
         trx_rows_modified: 1
   trx_concurrency_tickets: 0
       trx_isolation_level: READ COMMITTED
         trx_unique_checks: 1
    trx_foreign_key_checks: 1
trx_last_foreign_key_error: NULL
 trx_adaptive_hash_latched: 0
 trx_adaptive_hash_timeout: 0
          trx_is_read_only: 0
trx_autocommit_non_locking: 0
       trx_schedule_weight: NULL
2 rows in set (0.00 sec)

阐明：通过 InnoDB 的事务表 innodb_trx 查问到 thread_id=9346 的事务 3795 正在期待锁（trx_state: LOCK WAIT）,
thread_id=9320 的事务 3790 正在执行（trx_state: RUNNING）。mysql> select * from performance_schema.data_locks\G
*************************** 1. row ***************************
               ENGINE: INNODB
       ENGINE_LOCK_ID: 139727275779216:1113:139724882998560
ENGINE_TRANSACTION_ID: 3795
            THREAD_ID: 9441
             EVENT_ID: 5000
        OBJECT_SCHEMA: testdb
          OBJECT_NAME: t
       PARTITION_NAME: NULL
    SUBPARTITION_NAME: NULL
           INDEX_NAME: NULL
OBJECT_INSTANCE_BEGIN: 139724882998560
            LOCK_TYPE: TABLE
            LOCK_MODE: IX
          LOCK_STATUS: GRANTED
            LOCK_DATA: NULL
*************************** 2. row ***************************
               ENGINE: INNODB
       ENGINE_LOCK_ID: 139727275779216:52:4:3:139724882995456
ENGINE_TRANSACTION_ID: 3795
            THREAD_ID: 9441
             EVENT_ID: 5012
        OBJECT_SCHEMA: testdb
          OBJECT_NAME: t
       PARTITION_NAME: NULL
    SUBPARTITION_NAME: NULL
           INDEX_NAME: PRIMARY
OBJECT_INSTANCE_BEGIN: 139724882995456
            LOCK_TYPE: RECORD
            LOCK_MODE: X,REC_NOT_GAP
          LOCK_STATUS: WAITING
            LOCK_DATA: 8
*************************** 3. row ***************************
               ENGINE: INNODB
       ENGINE_LOCK_ID: 139727275781640:1113:139724883017072
ENGINE_TRANSACTION_ID: 3790
            THREAD_ID: 9415
             EVENT_ID: 15467
        OBJECT_SCHEMA: testdb
          OBJECT_NAME: t
       PARTITION_NAME: NULL
    SUBPARTITION_NAME: NULL
           INDEX_NAME: NULL
OBJECT_INSTANCE_BEGIN: 139724883017072
            LOCK_TYPE: TABLE
            LOCK_MODE: IX
          LOCK_STATUS: GRANTED
            LOCK_DATA: NULL
*************************** 4. row ***************************
               ENGINE: INNODB
       ENGINE_LOCK_ID: 139727275781640:52:4:3:139724883013968
ENGINE_TRANSACTION_ID: 3790
            THREAD_ID: 9441
             EVENT_ID: 5007
        OBJECT_SCHEMA: testdb
          OBJECT_NAME: t
       PARTITION_NAME: NULL
    SUBPARTITION_NAME: NULL
           INDEX_NAME: PRIMARY
OBJECT_INSTANCE_BEGIN: 139724883013968
            LOCK_TYPE: RECORD
            LOCK_MODE: X,REC_NOT_GAP
          LOCK_STATUS: GRANTED
            LOCK_DATA: 8
4 rows in set (0.00 sec)
阐明：事务 3795 正在期待 LOCK_TYPE: RECORD，LOCK_MODE:X,REC_NOT_GAP，期待的主键值为 8；事务 3790 已获取主键值为 8 的 LOCK_TYPE: RECORD，LOCK_MODE:X,REC_NOT_GAP 锁。mysql> select * from sys.innodb_lock_waits\G
*************************** 1. row ***************************
                wait_started: 2022-10-11 16:03:38
                    wait_age: 00:02:50
               wait_age_secs: 170
                locked_table: `testdb`.`t`
         locked_table_schema: testdb
           locked_table_name: t
      locked_table_partition: NULL
   locked_table_subpartition: NULL
                locked_index: PRIMARY
                 locked_type: RECORD
              waiting_trx_id: 3795
         waiting_trx_started: 2022-10-11 16:03:38
             waiting_trx_age: 00:02:50
     waiting_trx_rows_locked: 1
   waiting_trx_rows_modified: 0
                 waiting_pid: 9346
               waiting_query: update t set b=0 where a=8
             waiting_lock_id: 139727275779216:52:4:3:139724882995456
           waiting_lock_mode: X,REC_NOT_GAP
             blocking_trx_id: 3790
                blocking_pid: 9320
              blocking_query: NULL
            blocking_lock_id: 139727275781640:52:4:3:139724883013968
          blocking_lock_mode: X,REC_NOT_GAP
        blocking_trx_started: 2022-10-11 16:03:29
            blocking_trx_age: 00:02:59
    blocking_trx_rows_locked: 1
  blocking_trx_rows_modified: 1
     sql_kill_blocking_query: KILL QUERY 9320
sql_kill_blocking_connection: KILL 9320
1 row in set (0.01 sec)
阐明：事务 3795 期待 testdb.t 上的 rec_not_gap 独占锁，事务 3790 持有该独占锁。mysql> select distinct,* from sys.processlist where conn_id in (select trx_mysql_thread_id from information_schema.innodb_trx)\G
*************************** 1. row ***************************
                thd_id: 9441
               conn_id: 9346
                  user: admin@172.17.128.73
                    db: testdb
               command: Query
                 state: updating
                  time: 141
     current_statement: update t set b=0 where a=8
     statement_latency: 2.37 min
              progress: NULL
          lock_latency: 431.00 us
         rows_examined: 0
             rows_sent: 0
         rows_affected: 0
            tmp_tables: 0
       tmp_disk_tables: 0
             full_scan: NO
        last_statement: NULL
last_statement_latency: NULL
        current_memory: 140.15 KiB
             last_wait: wait/io/table/sql/handler
     last_wait_latency: Still Waiting
                source: handler.cc:3250
           trx_latency: 13.30 min
             trx_state: ACTIVE
        trx_autocommit: NO
                   pid: 9632
          program_name: mysql
*************************** 2. row ***************************
                thd_id: 9415
               conn_id: 9320
                  user: admin@172.17.128.73
                    db: testdb
               command: Sleep
                 state: NULL
                  time: 801
     current_statement: NULL
     statement_latency: NULL
              progress: NULL
          lock_latency: 288.00 us
         rows_examined: 0
             rows_sent: 0
         rows_affected: 1
            tmp_tables: 0
       tmp_disk_tables: 0
             full_scan: NO
        last_statement: insert into t values(8,8)
last_statement_latency: 765.23 us
        current_memory: 218.19 KiB
             last_wait: wait/io/socket/sql/client_connection
     last_wait_latency: Still Waiting
                source: viosocket.cc:146
           trx_latency: 13.52 min
             trx_state: ACTIVE
        trx_autocommit: NO
                   pid: 9600
          program_name: mysql
阐明：被阻塞事务执行的 sql 语句 update t set b=0 where a=8，阻塞事务执行的 sql 语句是 insert into t values(8,8)。

阐明：

MySQL 的隔离级别是通过索引上的锁实现并发事务管制的。在 READ-COMMITTED 隔离级别下，session1 在执行 insert 语句时，在主键索引上获取了 a = 8 的行记录独占锁，以禁止插入雷同主键的数据；session2 如果同时插入雷同的主键数据被阻塞，容易了解（Oracle 也同样阻塞）。出于同样的起因 session2 执行 update 时，因为无奈获取 a = 8 的行记录独占锁，同样也会被阻塞。

2.2. 验证 MySQL 事务未提交时已写入数据文件

验证事务未提交时，insert 语句已将数据写入数据文件，索引数据也已生成。

测试表 test1
mysql> CREATE TABLE `test1` (
    ->   `id` int NOT NULL AUTO_INCREMENT,
    ->   `k` int NOT NULL DEFAULT '0',
    ->   `c` char(120) NOT NULL DEFAULT '',
    ->   `pad` char(60) NOT NULL DEFAULT '',
    ->   PRIMARY KEY (`id`),
    ->   KEY `k_1` (`k`)
    -> ) ;
Query OK, 0 rows affected (0.07 sec)

开启一个事务，插入 10 万条数据。mysql> begin;
Query OK, 0 rows affected (0.00 sec)

mysql> insert into test1 select * from sbtest1;
Query OK, 100000 rows affected (1.44 sec)
Records: 100000  Duplicates: 0  Warnings: 0

检查表的 data_length 和 index_length
mysql> show table status where name like 'test1'\G
*************************** 1. row ***************************
           Name: test1
         Engine: InnoDB
        Version: 10
     Row_format: Dynamic
           Rows: 98712
 Avg_row_length: 228
    Data_length: 22593536
Max_data_length: 0
   Index_length: 2637824
      Data_free: 4194304
 Auto_increment: 100001
    Create_time: 2022-10-11 22:14:50
    Update_time: NULL
     Check_time: NULL
      Collation: utf8mb4_0900_ai_ci
       Checksum: NULL
 Create_options: 
        Comment: 
1 rows in set (0.01 sec)

回滚 insert 操作
mysql> rollback;
Query OK, 0 rows affected (1.35 sec)

更新统计信息
mysql> analyze table test1;

再次检查表的 data_length 和 index_length
mysql> show table status where name like 'test1'\G 
*************************** 1. row ***************************
           Name: test1
         Engine: InnoDB
        Version: 10
     Row_format: Dynamic
           Rows: 0
 Avg_row_length: 0
    Data_length: 16384
Max_data_length: 0
   Index_length: 16384
      Data_free: 29360128
 Auto_increment: 100001
    Create_time: 2022-10-11 22:22:36
    Update_time: NULL
     Check_time: NULL
      Collation: utf8mb4_0900_ai_ci
       Checksum: NULL
 Create_options: 
        Comment: 
1 row in set (0.00 sec)

查看数据文件的大小
[root@host73 testdb]# ll *test1.ibd
-rw-r----- 1 greatdb greatdb 30408704 Oct 11 15:12 sbtest1.ibd
-rw-r----- 1 greatdb greatdb 33554432 Oct 11 22:24 test1.ibd

阐明：

MySQL 在执行 insert 语句进行数据插入，未提交时，数据也已写入表的汇集索引，辅助索引也已生成。MySQL 能够应用未提交数据的索引，通过锁机制实现事务的并发管制。

3.Oracle 中 insert 没有阻塞 update

在 Oracle 中，创立同样的测试表 t，执行同样的 insert 和 update，但 insert 不会阻塞 update。

CREATE TABLE t (
  a int NOT NULL PRIMARY KEY ,
  b int DEFAULT NULL
);
insert into t values(7,7);
commit;

执行雷同的 insert 和 update 语句。

session1 session2
SQL> insert into t values(8,8);
1 row created. SQL> update t set b=0 where a=8;0 rows updated.

Enjoy GreatSQL :)

## 对于 GreatSQL

GreatSQL 是由万里数据库保护的 MySQL 分支，专一于晋升 MGR 可靠性及性能，反对 InnoDB 并行查问个性，是实用于金融级利用的 MySQL 分支版本。

相干链接：GreatSQL 社区 Gitee GitHub Bilibili

GreatSQL 社区：

社区博客有奖征稿详情：https://greatsql.cn/thread-100-1-1.html

技术交换群：

微信：扫码增加 GreatSQL 社区助手 微信好友，发送验证信息 加群。

正文完
数据库
发表至： 数据库
2023-03-28
 0
关于数据库:实践教程之如何在ARM平台部署PolarDBX
关于数据库:TDengine在黑格智造的落地应用
关于数据库:ElasticJob后续设计规划
关于数据库:TDSQL水平扩容实践案例