MySQL INSERT 导致的死锁分析

技术分享 7个月前 (08-28) 0 999+

前言

本文选用的 MySQL 版本：8.4.6

使用的数据

表结构：

DROP TABLE IF EXISTS store_snapshot_ext; DROP TABLE IF EXISTS store_snapshot;  create table store_snapshot (     id           varchar(32) not null comment '主键'         primary key,     warehouse_id varchar(32) null comment '仓库主键',     snap_date    datetime    null comment '快照日期',     create_id    varchar(32) null comment '创建人id',     create_time  datetime    null comment '创建日期',     modify_id    varchar(32) null comment '更新人 id',     modify_time  datetime    null comment '更新时间' )     comment '仓库快照';  create table store_snapshot_ext (     id             varchar(32) not null comment '主键'         primary key,     fk_snapshot_id varchar(32) null comment '快照外键',     ext_attr       varchar(32) null comment '扩展属性',     create_id      varchar(32) null comment '创建人id',     create_time    datetime    null comment '创建日期',     modify_id      varchar(32) null comment '更新人 id',     modify_time    datetime    null comment '更新时间',     constraint store_snapshot_ext___fk_snapshot_id         foreign key (fk_snapshot_id) references store_snapshot (id) )     comment '仓库快照扩展属性';

这里使用 Java 语言模拟并发情况下对数据的插入：

import cn.hutool.core.date.DateTime; import cn.hutool.core.util.RandomUtil; import org.junit.jupiter.api.Test; import org.junit.jupiter.api.extension.ExtendWith; import org.mybatis.spring.annotation.MapperScan; import org.springframework.boot.SpringApplication; import org.springframework.boot.autoconfigure.SpringBootApplication; import org.springframework.boot.test.context.SpringBootTest; import org.springframework.test.context.junit.jupiter.SpringExtension; import org.springframework.transaction.PlatformTransactionManager; import org.springframework.transaction.TransactionDefinition; import org.springframework.transaction.TransactionStatus; import org.springframework.transaction.annotation.EnableTransactionManagement; import org.springframework.transaction.support.DefaultTransactionDefinition; import org.tea.common.entity.StoreSnapshot; import org.tea.common.entity.StoreSnapshotExt; import org.tea.common.mapper.StoreSnapshotExtMapper; import org.tea.common.mapper.StoreSnapshotMapper;  import javax.annotation.Resource; import java.util.UUID; import java.util.concurrent.CountDownLatch;  @ExtendWith(SpringExtension.class) @SpringBootTest(classes = StoreSnapshotApplication.class) public class StoreSnapBenchTest {      @Resource     private PlatformTransactionManager txManager;      @Resource     private StoreSnapshotMapper storeSnapshotMapper;      @Resource     private StoreSnapshotExtMapper storeSnapshotExtMapper;      @Test     public void batchTest() throws InterruptedException {         Thread[] ts = new Thread[10];         CountDownLatch startLatch = new CountDownLatch(1);         CountDownLatch endLatch = new CountDownLatch(ts.length);         for (int i = 0; i < ts.length; i++) {             ts[i] = new Thread(() -> {                 DefaultTransactionDefinition definition = new DefaultTransactionDefinition();                 // 设置事务隔离级别为 "可重复读"                 definition.setIsolationLevel(TransactionDefinition.ISOLATION_REPEATABLE_READ);                 TransactionStatus status = txManager.getTransaction(definition);                 try {                     /**                     	提高事务竞争的激烈度                     */                     startLatch.await();                                          /**                     	一般情况下，不会使用如下的循环方式来插入数据，这里这么做的目的是为了                     	提高事务的处理时间，增大锁的竞争激烈度                     */                     for (int j = 0; j < 2000; j++) {                         StoreSnapshot snapshot = new StoreSnapshot();                         // 使用 UUID 的方式来引发 Page 分裂                         snapshot.setId(UUID.randomUUID().toString().replaceAll("-", ""));                         snapshot.setWarehouseId("warehouse_1");                         snapshot.setSnapDate(new DateTime());                         snapshot.init();                          storeSnapshotMapper.insertSelective(snapshot);                          StoreSnapshotExt snapshotExt = new StoreSnapshotExt();                         snapshotExt.setId(UUID.randomUUID().toString().replaceAll("-", ""));                         snapshotExt.setExtAttr(RandomUtil.randomString(32));                         // 注意这里的外键，后文会分析这个外键带来的一些影响                         snapshotExt.setFkSnapshotId(snapshot.getId());                         storeSnapshotExtMapper.insertSelective(snapshotExt);                     }                     txManager.commit(status);                 } catch (InterruptedException e) {                     txManager.rollback(status);                     throw new RuntimeException(e);                 } finally {                     endLatch.countDown();                 }             });         }          for (Thread t : ts) {             t.start();         }          startLatch.countDown();         endLatch.await();     } }  @SpringBootApplication @EnableTransactionManagement @MapperScan("org.tea.*.mapper") class StoreSnapshotApplication {       public static void main(String[] args) {         SpringApplication.run(StoreSnapshotApplication.class, args);     } }

在执行完上面的测试用例后，查看 MySQL InnoDB 的状态信息，发现已经出现了死锁：

------------------------ LATEST DETECTED DEADLOCK ------------------------ 2025-08-26 21:01:55 135637563962944 *** (1) TRANSACTION: TRANSACTION 3866, ACTIVE 2 sec inserting mysql tables in use 1, locked 1 LOCK WAIT 27 lock struct(s), heap size 8312, 220 row lock(s), undo log entries 430 MySQL thread id 17, OS thread handle 135637553456704, query id 8736 localhost ::1 root update INSERT INTO store_snapshot (create_id, warehouse_id, snap_date, modify_id, id, modify_time, create_time)  VALUES  ('system', 'warehouse_1', '2025-08-26 21:01:54', 'system', '402ce43f650a483eb0c9c5138e50d6f0', '2025-08-26 21:01:54', '2025-08-26 21:01:54')  *** (1) HOLDS THE LOCK(S): RECORD LOCKS space id 10 page no 11 n bits 240 index PRIMARY of table `lxh_db`.`store_snapshot` trx id 3866 lock_mode X Record lock, heap no 1 PHYSICAL RECORD: n_fields 1; compact format; info bits 0  0: len 8; hex 73757072656d756d; asc supremum;;   *** (1) WAITING FOR THIS LOCK TO BE GRANTED: RECORD LOCKS space id 10 page no 20 n bits 160 index PRIMARY of table `lxh_db`.`store_snapshot` trx id 3866 lock_mode X locks gap before rec insert intention waiting Record lock, heap no 5 PHYSICAL RECORD: n_fields 9; compact format; info bits 0  0: len 30; hex 343033303963393162373166343731633936323164616565643434666363; asc 40309c91b71f471c9621daeed44fcc; (total 32 bytes);  1: len 6; hex 000000000f14; asc       ;;  2: len 7; hex 82000001070630; asc       0;;  3: len 11; hex 77617265686f7573655f31; asc warehouse_1;;  4: len 5; hex 99b7755074; asc   uPt;;  5: len 6; hex 73797374656d; asc system;;  6: len 5; hex 99b7755074; asc   uPt;;  7: len 6; hex 73797374656d; asc system;;  8: len 5; hex 99b7755074; asc   uPt;;   *** (2) TRANSACTION: TRANSACTION 3860, ACTIVE 3 sec inserting mysql tables in use 1, locked 1 LOCK WAIT 46 lock struct(s), heap size 24696, 1258 row lock(s), undo log entries 2474 MySQL thread id 11, OS thread handle 135637890971200, query id 12330 localhost ::1 root update INSERT INTO store_snapshot (create_id, warehouse_id, snap_date, modify_id, id, modify_time, create_time)  VALUES  ('system', 'warehouse_1', '2025-08-26 21:01:55', 'system', '917f3578682c467384e520fd6c00b86d', '2025-08-26 21:01:55', '2025-08-26 21:01:55')  *** (2) HOLDS THE LOCK(S): RECORD LOCKS space id 10 page no 20 n bits 160 index PRIMARY of table `lxh_db`.`store_snapshot` trx id 3860 lock_mode X locks gap before rec Record lock, heap no 3 PHYSICAL RECORD: n_fields 9; compact format; info bits 0  0: len 30; hex 336661623363323037333232346465666263363762333630363763623666; asc 3fab3c2073224defbc67b36067cb6f; (total 32 bytes);  1: len 6; hex 000000000f14; asc       ;;  2: len 7; hex 82000000932d80; asc      - ;;  3: len 11; hex 77617265686f7573655f31; asc warehouse_1;;  4: len 5; hex 99b7755076; asc   uPv;;  5: len 6; hex 73797374656d; asc system;;  6: len 5; hex 99b7755076; asc   uPv;;  7: len 6; hex 73797374656d; asc system;;  8: len 5; hex 99b7755076; asc   uPv;;  Record lock, heap no 4 PHYSICAL RECORD: n_fields 9; compact format; info bits 0  0: len 30; hex 336664326234626538643463343338326130303764383431366364613536; asc 3fd2b4be8d4c4382a007d8416cda56; (total 32 bytes);  1: len 6; hex 000000000f14; asc       ;;  2: len 7; hex 82000000903800; asc      8 ;;  3: len 11; hex 77617265686f7573655f31; asc warehouse_1;;  4: len 5; hex 99b7755076; asc   uPv;;  5: len 6; hex 73797374656d; asc system;;  6: len 5; hex 99b7755076; asc   uPv;;  7: len 6; hex 73797374656d; asc system;;  8: len 5; hex 99b7755076; asc   uPv;;  Record lock, heap no 5 PHYSICAL RECORD: n_fields 9; compact format; info bits 0  0: len 30; hex 343033303963393162373166343731633936323164616565643434666363; asc 40309c91b71f471c9621daeed44fcc; (total 32 bytes);  1: len 6; hex 000000000f14; asc       ;;  2: len 7; hex 82000001070630; asc       0;;  3: len 11; hex 77617265686f7573655f31; asc warehouse_1;;  4: len 5; hex 99b7755074; asc   uPt;;  5: len 6; hex 73797374656d; asc system;;  6: len 5; hex 99b7755074; asc   uPt;;  7: len 6; hex 73797374656d; asc system;;  8: len 5; hex 99b7755074; asc   uPt;;  Record lock, heap no 93 PHYSICAL RECORD: n_fields 9; compact format; info bits 0  0: len 30; hex 336664396165326435663738346237656161346134323336643031643366; asc 3fd9ae2d5f784b7eaa4a4236d01d3f; (total 32 bytes);  1: len 6; hex 000000000f14; asc       ;;  2: len 7; hex 82000000a81784; asc        ;;  3: len 11; hex 77617265686f7573655f31; asc warehouse_1;;  4: len 5; hex 99b7755077; asc   uPw;;  5: len 6; hex 73797374656d; asc system;;  6: len 5; hex 99b7755077; asc   uPw;;  7: len 6; hex 73797374656d; asc system;;  8: len 5; hex 99b7755077; asc   uPw;;   *** (2) WAITING FOR THIS LOCK TO BE GRANTED: RECORD LOCKS space id 10 page no 11 n bits 240 index PRIMARY of table `lxh_db`.`store_snapshot` trx id 3860 lock_mode X insert intention waiting Record lock, heap no 1 PHYSICAL RECORD: n_fields 1; compact format; info bits 0  0: len 8; hex 73757072656d756d; asc supremum;;  *** WE ROLL BACK TRANSACTION (1) ------------ TRANSACTIONS ------------

可以看到，在日志中，事务 3866 持有一个页号为 11，supremum (相当于双链表的尾部哑节点) 的间隙锁，在等待页号为 20，主键为 40309c91b71f471c9621daeed44fcc 的间隙锁；同时，事务 3860 持有页号为 20，主键为 40309c91b71f471c9621daeed44fcc 的间隙锁，同时在等待页号为 10 的 supremum 的间隙锁。这两个事务构成了循环回路，并且在相互等待，因此形成了死锁，具体的图示如下所示：

具体分析

查询语句的加锁

MySQL 的的一般 SELECT 语句在非串行化隔离级别下是通过一致性读的方式进行读取，本身不会对记录加锁，但是在存在外键约束的情况下，依旧会对关联的外键约束记录上加上 S 型的记录锁，如果关联的外键约束没有被找到，在"可重复读"的隔离级别下，会在外键记录附近加上间隙锁

由于这里插入语句的外键都能被找到，因此这里的外键不是产生间隙锁的原因
INSERT 语句的加锁

单纯的 INSERT 语句在插入时加上的是一种特殊的记录锁，不同事务的插入意向锁不会相互阻塞，但是在插入的记录行的所处位置存在间隙锁的情况下，会为当前的 INSERT 记录加上插入意向锁

在主键重复的情况下，隔离级别为 "可重复读" 或 "串行化" 的情况下，会为插入的记录加上 S 型的 Next-Key 锁。在这种情况下，如果原来待插入行的事务回滚了，由于本身持有的 S 型锁无法再获取到独占锁，就有可能会引发死锁^[2]。

由于这里的主键都是 UUID，不存在重复主键，并且结合相关的日志信息，并不是由于单纯的 INSERT 语句导致的死锁
Page 的分裂

当插入的记录的主键不连续时，MySQL 为了维护聚簇索引的顺序，可能会引发页的分裂。在事务隔离级别为 "可重复读"或"串行化" 的情况下，对分裂的数据进行迁移的过程中，相当于对数据执行了更新的操作，按照 MySQL 对于 UPDATE 语句的加锁情况^[2]，会在记录上加上 Next-Key (记录锁和间隙锁)，因此事务 3860会持有记录40309c91b71f471c9621daeed44fcc的间隙锁；而为了防止在分裂维护过程中重新插入数据，可能不得不为相关页记录的 supremum 加上间隙锁，以维护页分裂的执行过程

这个过程可能如下：

插入记录前：

插入记录导致页分裂后

注意: 这里关于分页而产生的间隙锁为实际实验推断，并无实际文档与之关联。当事务隔离级别为 "读提交" 或插入的记录的主键存在顺序时，都不会出现上文描述的死锁出现