MySQL死锁排查与预防实战

Deadlock found when trying to get lock; try restarting transaction

SHOW ENGINE INNODB STATUS\G

*** (1) TRANSACTION:
UPDATE orders SET status = 'paid' WHERE id = 1001
*** (1) HOLDS THE LOCK(S):  -- 持有orders表的锁
*** (1) WAITING FOR THIS LOCK:  -- 等inventory表的锁

*** (2) TRANSACTION:
UPDATE inventory SET quantity = quantity - 1 WHERE product_id = 2001
*** (2) HOLDS THE LOCK(S):  -- 持有inventory表的锁
*** (2) WAITING FOR THIS LOCK:  -- 等orders表的锁

*** WE ROLL BACK TRANSACTION (2)

// 路径A：先改订单再扣库存
updateOrderStatus(orderId, "paid");
decreaseInventory(productId, 1);

// 路径B：先扣库存再改订单（另一个接口）
decreaseInventory(productId, 1);
updateOrderStatus(orderId, "paid");

// 统一顺序：先orders后inventory
@Transactional
public void processOrder(long orderId, long productId) {
    updateOrderStatus(orderId, "paid");  // 永远先锁orders
    decreaseInventory(productId, 1);     // 再锁inventory
}

List<Long> ids = Arrays.asList(id1, id2, id3);
Collections.sort(ids);
for (Long id : ids) {
    lockAndProcess(id);
}

-- 事务A
SELECT * FROM orders WHERE user_id = 3 FOR UPDATE;
-- 没有user_id=3的数据，但会锁住(1,5)这个间隙

-- 事务B
SELECT * FROM orders WHERE user_id = 7 FOR UPDATE;
-- 锁住(5,10)这个间隙

-- 然后两边各自INSERT
-- 事务A想插入user_id=6，要等(5,10)的间隙锁
-- 事务B想插入user_id=4，要等(1,5)的间隙锁
-- 死锁

SET SESSION TRANSACTION ISOLATION LEVEL READ COMMITTED;

// 这种写法不好
@Transactional
public void process() {
    queryData();        // 查数据
    callExternalApi();  // 调外部接口，可能很慢
    updateDatabase();   // 更新数据库
}

// 改成这样
public void process() {
    queryData();
    callExternalApi();  // 外部调用放事务外面
    updateInTransaction();
}

@Transactional
public void updateInTransaction() {
    updateDatabase();   // 只有真正需要事务的操作
}

# 简单脚本，每分钟检查死锁次数
DEADLOCKS=$(mysql -e "SHOW GLOBAL STATUS LIKE 'Innodb_deadlocks'" | awk 'NR==2{print $2}')
echo "$(date) deadlocks: $DEADLOCKS" >> /var/log/deadlock.log

- alert: MySQLDeadlock
  expr: increase(mysql_global_status_innodb_deadlocks[5m]) > 0
  for: 1m

int retry = 3;
while (retry-- > 0) {
    try {
        doTransaction();
        break;
    } catch (DeadlockException e) {
        if (retry == 0) throw e;
        Thread.sleep(100);  // 等一下再试
    }
}