无法破译 innodb 状态日志中的死锁

Question

我们从 Microsoft ADO.NET 连接器访问 MySQL。

有时我们会在 innodb 状态中看到以下死锁，并且无法确定问题的原因。看起来事务（2）正在等待并持有相同的锁？

------------------------
LATEST DETECTED DEADLOCK
------------------------
110606  5:35:09
*** (1) TRANSACTION:
TRANSACTION 0 45321452, ACTIVE 0 sec, OS thread id 3804 starting index read
mysql tables in use 1, locked 1
LOCK WAIT 2 lock struct(s), heap size 368, 1 row lock(s)
    MySQL thread id 84, query id 3265713 localhost 127.0.0.1 famdev Updating
    UPDATE people SET company_id = 1610, name = '<name>', password = '<hash>', temp_password = NULL, reset_password_hash = NULL, email = '<redacted>@yahoo.com', phone = NULL, mobile = '<phone>', iphone_device_id = 'android:<id>-<id>', iphone_device_time = '2011-06-06 05:35:09', last_checkin = '2011-06-06 05:24:42', location_lat = <lat>, location_long = -<lng>, gps_strength = 3296, picture_blob_id = 1190, authority = 1, active = 1, date_created = '2011-04-13 20:21:20', last_login = '2011-06-06 05:35:09', panic_mode = 0, battery_level = NULL, battery_state = NULL WHERE people_id = 3125
    *** (1) WAITING FOR THIS LOCK TO BE GRANTED:
    RECORD LOCKS space id 0 page no 11144 n bits 152 index `PRIMARY` of table `family`.`people` trx id 0 45321452 lock_mode X locks rec but not gap waiting
    Record lock, heap no 12 PHYSICAL RECORD: n_fields 25; compact format; info bits 0
    0: len 8; hex 8000000000000c35; asc        5;; 1: len 6; hex 000002b38ce6; asc       ;; 2: len 7; hex 00000002801f89; asc        ;; 3: len 8; hex 800000000000064a; asc        J;; 4: len 4; hex <data>; asc <name>;; 5: len 30; hex <data>; asc <data>;...(truncated); 6: SQL NULL; 7: SQL NULL; 8: len 16; hex <data>; asc <redacted>@yahoo.com;; 9: SQL NULL; 10: len 10; hex <data>; asc <phone>;; 11: len 30; hex <data>; asc android:<id>;...(truncated); 12: len 8; hex <data>; asc    J]  Z;; 13: len 8; hex <data>; asc    J]  Z;; 14: len 8; hex a39410acaa9b4340; asc       C@;; 15: len 8; hex <data>; asc     m S ;; 16: len 2; hex 8ce0; asc   ;; 17: len 8; hex 80000000000004a6; asc         ;; 18: len 4; hex 80000001; asc     ;; 19: len 1; hex 81; asc  ;; 20: len 8; hex <data>; asc    JR   ;; 21: len 8; hex <data>; asc    J]   ;; 22: len 1; hex 80; asc  ;; 23: SQL NULL; 24: SQL NULL;

    *** (2) TRANSACTION:
    TRANSACTION 0 45321448, ACTIVE 0 sec, OS thread id 3176 starting index read, thread declared inside InnoDB 500
    mysql tables in use 1, locked 1
    5 lock struct(s), heap size 1216, 2 row lock(s), undo log entries 1
    MySQL thread id 85, query id 3265714 localhost 127.0.0.1 famdev Updating
    UPDATE people SET company_id = 1610, name = '<name>', password = '<hash>', temp_password = NULL, reset_password_hash = NULL, email = '<redacted>@yahoo.com', phone = NULL, mobile = '<phone>', iphone_device_id = 'android:<id>-<id>-<id>-<id>', iphone_device_time = '2011-06-06 05:24:42', last_checkin = '2011-06-06 05:35:07', location_lat = <lat>, location_long = -<lng>, gps_strength = 3296, picture_blob_id = 1190, authority = 1, active = 1, date_created = '2011-04-13 20:21:20', last_login = '2011-06-06 05:35:09', panic_mode = 0, battery_level = NULL, battery_state = NULL WHERE people_id = 3125
    *** (2) HOLDS THE LOCK(S):
        RECORD LOCKS space id 0 page no 11144 n bits 152 index `PRIMARY` of table `family`.`people` trx id 0 45321448 lock mode S locks rec but not gap
        Record lock, heap no 12 PHYSICAL RECORD: n_fields 25; compact format; info bits 0
        0: len 8; hex 8000000000000c35; asc        5;; 1: len 6; hex 000002b38ce6; asc       ;; 2: len 7; hex 00000002801f89; asc        ;; 3: len 8; hex 800000000000064a; asc        J;; 4: len 4; hex <data>; asc <name>;; 5: len 30; hex <data>; asc <data>;...(truncated); 6: SQL NULL; 7: SQL NULL; 8: len 16; hex <data>; asc <redacted>@yahoo.com;; 9: SQL NULL; 10: len 10; hex <data>; asc <phone>;; 11: len 30; hex <data>; asc android:<id>;...(truncated); 12: len 8; hex <data>; asc    J]  Z;; 13: len 8; hex <data>; asc    J]  Z;; 14: len 8; hex a39410acaa9b4340; asc       C@;; 15: len 8; hex <data>; asc     m S ;; 16: len 2; hex 8ce0; asc   ;; 17: len 8; hex 80000000000004a6; asc         ;; 18: len 4; hex 80000001; asc     ;; 19: len 1; hex 81; asc  ;; 20: len 8; hex <data>; asc    JR   ;; 21: len 8; hex <data>; asc    J]   ;; 22: len 1; hex 80; asc  ;; 23: SQL NULL; 24: SQL NULL;

        *** (2) WAITING FOR THIS LOCK TO BE GRANTED:
        RECORD LOCKS space id 0 page no 11144 n bits 152 index `PRIMARY` of table `family`.`people` trx id 0 45321448 lock_mode X locks rec but not gap waiting
        Record lock, heap no 12 PHYSICAL RECORD: n_fields 25; compact format; info bits 0
        0: len 8; hex 8000000000000c35; asc        5;; 1: len 6; hex 000002b38ce6; asc       ;; 2: len 7; hex 00000002801f89; asc        ;; 3: len 8; hex 800000000000064a; asc        J;; 4: len 4; hex <data>; asc <name>;; 5: len 30; hex <data>; asc <data>;...(truncated); 6: SQL NULL; 7: SQL NULL; 8: len 16; hex <data>; asc <redacted>@yahoo.com;; 9: SQL NULL; 10: len 10; hex <data>; asc <phone>;; 11: len 30; hex <data>; asc android:<id>;...(truncated); 12: len 8; hex <data>; asc    J]  Z;; 13: len 8; hex <data>; asc    J]  Z;; 14: len 8; hex a39410acaa9b4340; asc       C@;; 15: len 8; hex <data>; asc     m S ;; 16: len 2; hex 8ce0; asc   ;; 17: len 8; hex 80000000000004a6; asc         ;; 18: len 4; hex 80000001; asc     ;; 19: len 1; hex 81; asc  ;; 20: len 8; hex <data>; asc    JR   ;; 21: len 8; hex <data>; asc    J]   ;; 22: len 1; hex 80; asc  ;; 23: SQL NULL; 24: SQL NULL;

        *** WE ROLL BACK TRANSACTION (1)

我们阅读了有关 next 键锁定的这篇文章，但它似乎不适用于我们，因为我们没有选择更新。

更新

今天早上我发现了一个稍微不同的死锁签名，这可能是造成这个死锁的根本原因。为了让事情变得简单，我已将该僵局作为一个单独的问题发布。如果我能确认另一个问题是原因，我会在这里更新。

Answer 1

这是我所看到的

我看到三个查询，几乎完全相同。

UPDATE people SET company_id = 1610, name = '<name>', password = '<hash>',
temp_password = NULL, reset_password_hash = NULL, email = '<redacted>@yahoo.com',
phone = NULL, mobile = '<phone>', iphone_device_id = 'android:<id>-<id>',
iphone_device_time = '2011-06-06 05:35:09', last_checkin = '2011-06-06 05:24:42',
location_lat = <lat>, location_long = -<lng>, gps_strength = 3296,
picture_blob_id = 1190,authority = 1, active = 1,
date_created = '2011-04-13 20:21:20',
last_login = '2011-06-06 05:35:09', panic_mode = 0, battery_level = NULL,
battery_state = NULL WHERE people_id = 3125;

差异

交易 1

iphone_device_time = '2011-06-06 05:24:42', last_checkin = '2011-06-06 05:35:07'

交易2

iphone_device_time = '2011-06-06 05:35:09', last_checkin = '2011-06-06 05:24:42'

请注意列值被翻转。通常，当两个不同的事务从两个表中访问两个锁时会发生死锁，其中 TX1（事务 1）获取 A 行然后 B 行，而 TX2 获取 B 行然后 A 行。在这种情况下，TX1 和 TX2 是访问同一行但更改两个不同的列（iphone_device_time，last_checkin）。

这些值没有任何意义。在 5:24:42，您最后一次签到是 5:35:07。10 分 27 秒后 (5:35:07 - 05:24:42)，列值反转。

最大的问题是：为什么 TX1 被阻塞了将近 11 分钟？？？

这不是一个真正的答案。这只是带宽和我的全部内容。我希望这些观察有帮助。

更新 2011-06-06 09:57

请查看有关 innodb_locks_unsafe_for_binlog 的链接：我建议阅读本文的原因是我在您的 INNODB 状态显示中看到的其他内容。短语 lock_mode X（排他锁）和 lock_mode S（共享锁）表示在同一行上施加（或试图施加）两个锁。可能有一些内部序列化正在进行下一行锁定。默认为关闭。阅读本文后，您可能需要考虑启用它。

更新 2011-06-06 10:03

检查这种思路的另一个原因是所有事务都遍历 PRIMARY 键这一事实。由于 PRIMARY 是 InnoDB 中的聚集索引，因此 PRIMARY 键和行本身是在一起的。因此，遍历一行 and 和 PRIMARY KEY 是一回事。因此，PRIMARY KEY 上的任何索引锁也是行级锁。

更新 2011-06-06 19:21

检查您拥有的auocommit值。如果自动提交关闭，我可以看到两（2）个可能的问题

1. 在同一事务中两次更新同一行
2. 在两个不同的事务中更新同一行

实际上，您在问题中显示的 SHOW ENGINE INNODB STATUS 恰好具有两种情况。

Answer 2

罗兰多的回答无疑有助于我们找到正确的解决方案。然而，我们最终没有调整自动提交配置、隔离级别或innodb_locks_unsafe_for_binlog配置。

我们相信我们在第一个问题上发布的死锁日志是我们随后发现并发布在此处的死锁的结果。自从我们解决了这两个查询的问题后，我们就没有看到任何死锁。