MySQL鎖系列 之 死鎖
阿新 • • 發佈:2019-02-06
一、什麼是死鎖
- 1.必須滿足的條件
1. 必須有兩個或者兩個以上的事務
2. 不同事務之間都持有對方需要的鎖資源。 A事務需要B的資源,B事務需要A的資源,這就是典型的AB-BA死鎖
- 2.死鎖相關的引數
* innodb_print_all_deadlocks
1. 如果這個引數開啟,那麼死鎖相關的資訊都會列印輸出到error log
* innodb_lock_wait_timeout
1. 當MySQL獲取row lock的時候,如果wait了innodb_lock_wait_timeout=N的時間,會報以下錯誤
ERROR 1205 (HY000): Lock wait timeout exceeded; try restarting transaction
* innodb_deadlock_detect
1. innodb_deadlock_detect = off 可以關閉掉死鎖檢測,那麼就發生死鎖的時候,用鎖超時來處理。
2. innodb_deadlock_detect = on (預設選項)開啟死鎖檢測,資料庫自動回滾
* innodb_status_lock_output = on
1. 可以看到更加詳細的鎖資訊
二、死鎖有什麼危害
- 死鎖,即表明有多個事務之間需要互相爭奪資源而互相等待。
- 如果沒有死鎖檢測,那麼就會互相卡死,一直hang死
- 如果有死鎖檢測機制,那麼資料庫會自動根據代價來評估出哪些事務可以被回滾掉,用來打破這個僵局
- 所以說:死鎖並沒有啥壞處,官網:www.fhadmin.org 反而可以保護資料庫和應用
- 那麼出現死鎖,而且非常頻繁,我們應該調整業務邏輯,讓其避免產生死鎖方為上策
三、典型的死鎖案例剖析
3.1 死鎖案例一
典型的 官網:www.fhadmin.org AB-BA 死鎖
session 1:
select * from tb_b where id_2 = 1 for update (A)
session 2:
select * from tb_a where id = 2 for update (B)
session 1:
select * from tb_a where id = 2 for update (B)
session 2:
select * from tb_b where id_2 = 1 for update (A)
ERROR 1213 (40001): Deadlock found when trying to get lock; try restarting transaction
1213的死鎖錯誤,mysql會自動回滾
哪個回滾代價最小,回滾哪個(根據undo判斷)
------------------------
LATEST DETECTED DEADLOCK
------------------------
2017 -06-22 16:39:50 0x7f547dd02700
*** (1) TRANSACTION:
TRANSACTION 133601982, ACTIVE 48 sec starting index read
mysql tables in use 1, locked 1
LOCK WAIT 4 lock struct(s), heap size 1136, 2 row lock(s)
MySQL thread id 11900, OS thread handle 140000866637568, query id 25108 localhost dba statistics
select * from tb_a where id = 2 for update -----session1 持有tb_a中記錄為2的鎖
*** (1) WAITING FOR THIS LOCK TO BE GRANTED:
RECORD LOCKS space id 303 page no 3 n bits 72 index PRIMARY of table `lc_5`.`tb_a` trx id 133601982 lock_mode X locks rec but not gap waiting
Record lock, heap no 3 PHYSICAL RECORD: n_fields 3; compact format; info bits 0
0: len 4; hex 80000002; asc ;; --session 1 需要tb_a中記錄為2的鎖( session1 -> session2 )
1: len 6; hex 000007f69ab2; asc ;;
2: len 7; hex dc000027100110; asc ' ;;
*** (2) TRANSACTION:
TRANSACTION 133601983, ACTIVE 28 sec starting index read, thread declared inside InnoDB 5000
mysql tables in use 1, locked 1
4 lock struct(s), heap size 1136, 2 row lock(s)
MySQL thread id 11901, OS thread handle 140000864773888, query id 25109 localhost dba statistics
select * from tb_b where id_2 = 1 for update
*** (2) HOLDS THE LOCK(S):
RECORD LOCKS space id 303 page no 3 n bits 72 index PRIMARY of table `lc_5`.`tb_a` trx id 133601983 lock_mode X locks rec but not gap
Record lock, heap no 3 PHYSICAL RECORD: n_fields 3; compact format; info bits 0
0: len 4; hex 80000002; asc ;; --session 2 持有tb_a中記錄等於2的鎖
1: len 6; hex 000007f69ab2; asc ;;
2: len 7; hex dc000027100110; asc ' ;;
*** (2) WAITING FOR THIS LOCK TO BE GRANTED:
RECORD LOCKS space id 304 page no 3 n bits 72 index PRIMARY of table `lc_5`.`tb_b` trx id 133601983 lock_mode X locks rec but not gap waiting
Record lock, heap no 2 PHYSICAL RECORD: n_fields 3; compact format; info bits 0
0: len 4; hex 80000001; asc ;; --session 2 需要tb_b中記錄為1的鎖 ( session2 -> session1 )
1: len 6; hex 000007f69ab8; asc ;;
2: len 7; hex e0000027120110; asc ' ;;
最終的結果:
死鎖路徑:[session1 -> session2 , session2 -> session1]
ABBA死鎖產生
3.2 死鎖案例二
同一個事務中,官網:www.fhadmin.org S-lock 升級為 X-lock 不能直接繼承
* session 1:
mysql> CREATE TABLE t (i INT) ENGINE = InnoDB;
Query OK, 0 rows affected (1.07 sec)
mysql> INSERT INTO t (i) VALUES(1);
Query OK, 1 row affected (0.09 sec)
mysql> START TRANSACTION;
Query OK, 0 rows affected (0.00 sec)
mysql> SELECT * FROM t WHERE i = 1 LOCK IN SHARE MODE; --獲取S-lock
+------+
| i |
+------+
| 1 |
+------+
* session 2:
mysql> START TRANSACTION;
Query OK, 0 rows affected (0.00 sec)
mysql> DELETE FROM t WHERE i = 1; --想要獲取X-lock,但是被session1的S-lock 卡住,目前處於waiting lock階段
* session 1:
mysql> DELETE FROM t WHERE i = 1; --想要獲取X-lock,session1本身擁有S-lock,但是由於session 2 獲取X-lock再前,所以session1不能夠從S-lock 提升到 X-lock,需要等待session2 釋放才可以獲取,所以造成死鎖
ERROR 1213 (40001): Deadlock found when trying to get lock;
try restarting transaction
死鎖路徑:
session2 -> session1 , session1 -> session2
3.3 死鎖案例三
唯一鍵死鎖 (delete + insert)
關鍵點在於:S-lock
dba:lc_3> show create table uk;
+-------+--------------------------------------------------------------------------------------------------------------+
| Table | Create Table |
+-------+--------------------------------------------------------------------------------------------------------------+
| uk | CREATE TABLE `uk` (
`a` int(11) NOT NULL,
UNIQUE KEY `uniq_a` (`a`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8 |
+-------+--------------------------------------------------------------------------------------------------------------+
1 row in set (0.00 sec)
dba:lc_3> select * from uk;
+---+
| a |
+---+
| 1 |
+---+
1 row in set (0.00 sec)
session 1:
dba:lc_3> begin;
Query OK, 0 rows affected (0.00 sec)
dba:lc_3> delete from uk where a=1;
Query OK, 1 row affected (0.00 sec)
session 2:
dba:(none)> use lc_3;
Database changed
dba:lc_3> insert into uk values(1); --wait lock(想要加S-lock,卻被sesson1的X-lock卡住)
sesson 3:
dba:(none)> use lc_3;
Database changed
dba:lc_3> insert into uk values(1); --wait lock(想要加S-lock,卻被sesson1的X-lock卡住)
session 1:
commit; --session2和session3 都獲得了S-lock,然後都想要去給記錄1 加上X-lock,卻互相被對方的S-lock卡住,死鎖產生
再來看session 2 和 session 3 的結果:
session2:
Query OK, 1 row affected (7.36 sec)
session3:
ERROR 1213 (40001): Deadlock found when trying to get lock; try restarting transaction
總結: 試想想,如果session 1 不是commit,而是rollback會是怎麼樣呢? 大家去測測就會發現,結果肯定是唯一鍵衝突啊
3.4 死鎖案例四
主鍵和二級索引的死鎖
* primary key
1 2 3 4 --primary key col1
10 30 20 40 --idx_key2 col2
100 200 300 400 --idx_key3 col3
* idx_key2 select * from t where col2 > 10: 鎖二級索引順序為:20 =》30 , 對應鎖主鍵的順序為:3 =》2
10 20 30 40
1 3 2 4
* idx_key3 select * from t where col3 > 100:鎖二級索引順序為:200 =》300 , 對應鎖主鍵的順序為:2 =》3
100 200 300 400
1 2 3 4
死鎖路徑:
由於二級索引引起的主鍵加鎖順序: 3 =》2
由於二級索引引起的主鍵加鎖順序: 2 =》3
這個要求併發,且剛好
session 1 加鎖3的時候 session 2 要加鎖2.
session 1 加鎖2的時候 session 3 要加鎖3.
這樣就產生了 AB-BA 死鎖
3.5 死鎖案例五
purge + unique key 引發的死鎖
A表的記錄: id = 1 10 40 100 200 500 800 900
session 1 :
delete from a where id = 10; ???
session 2 :
delete from a where id = 800; ???
session 1 :
insert into a select 800; ???
session 2 :
insert into a select 10; ???
* 如果大家去跑這兩鍾SQL語句的併發測試,是可以導致死鎖的。
* 如何驗證是由於purge導致的問題呢?這個本想用mysqld-debug模式去關閉purge執行緒,但是很遺憾我沒能模擬出來。。。
3.6 死鎖案例六
REPLACE INTO問題
* 這個問題模擬起來非常簡單,原理非常複雜,這裡不過多解釋
* 詳情請看姜老師的文章,據說看懂了年薪都100w了: http://www.innomysql.com/26186-2/
* 解決方案:
* 用insert into ... on duplicate key update 代替 replace into
* 此方案親測有效
四、如何避免死鎖
- 產生死鎖的原因
1. 事務之間互相佔用資源
- 方法和總結
1. 降低隔離級別,修改 RR -> RC , 如果這個調整了,可以避免掉60%的死鎖場景和奇怪的鎖等待
2. 調整業務邏輯和SQL,讓其都按照順序執行操作
3. 減少unique索引,大部分死鎖的場景都是由於unique索引導致
4. 儘量不用replace into,用insert into ... on duplicate key update 代替