一前言

死鎖，其實是一個很有意思也很有挑戰的技術問題，大概每個DBA和部分開發同學都會在工作過程中遇見。關於死鎖我會持續寫一個系列的案例分析，希望能夠對想了解死鎖的朋友有所幫助。

二案例分析

2.1 環境說明

MySQL 5.6.24 事務隔離級別為RR

create table tx (
  id int not null primary key auto_increment ,
  c1 int not null default 0,
  c2 int not null default 0,
  key idx_c1(c1)
) engine=innodb ;
insert into tx values(24,3,4),(25,3,4),
(26,3,4),(30,5,8);

2.2 測試用例

	sess1	sess2
	begin;	begin
T1		select * from tx where id=30 for update;
T2	update tx set c2=8 where c1=5; ERROR 1213 (40001): Deadlock found when trying to get lock; try restarting transaction
T3		delete from tx where id=30;

2.3 死鎖日誌

----------------------------------
LATEST DETECTED DEADLOCK
------------------------
2018-03-27 15:40:40 0x7f75cafce700
*** (1) TRANSACTION:
TRANSACTION 1850, ACTIVE 20 sec starting index read
mysql tables in use 1, locked 1
LOCK WAIT 3 lock struct(s), heap size 1136, 2 row lock(s)
MySQL thread id 379040, OS thread handle 140143994337024, 
query id 1521958 localhost root updating
update tx set c2=8 where c1=5
*** (1) WAITING FOR THIS LOCK TO BE GRANTED:
RECORD LOCKS space id 27 page no 3 n bits 72 index PRIMARY of table 
`test`.`tx` trx id 1850 lock_mode X locks rec but not gap waiting
*** (2) TRANSACTION:
TRANSACTION 1849, ACTIVE 32 sec updating or deleting, 
thread declared inside InnoDB 4999
mysql tables in use 1, locked 1
3 lock struct(s), heap size 1136, 2 row lock(s), undo log entries 1
MySQL thread id 379016, OS thread handle 140143893473024, 
query id 1521976 localhost root updating
delete from tx where id=30
*** (2) HOLDS THE LOCK(S):
RECORD LOCKS space id 27 page no 3 n bits 72 index PRIMARY of 
table `test`.`tx` trx id 1849 lock_mode X locks rec but not gap
*** (2) WAITING FOR THIS LOCK TO BE GRANTED:
RECORD LOCKS space id 27 page no 5 n bits 72 index idx_c1 of 
table `test`.`tx` trx id 1849 lock_mode X locks rec but not 
gap waiting
*** WE ROLL BACK TRANSACTION (1)

2.4 分析死鎖日誌

首先要理解的是對同一個欄位申請加鎖是需要排隊的。

其次表ty中索引idx_c1為非唯一普通索引，我們根據事務執行的時間順序來解釋，這樣比較好理解。

T1: sess2 執行select for update 操作持有記錄id=30的主鍵行鎖:PRIMARY of table test.tx lock_mode X locks rec but not gap

T2: sess1 語句update透過普通索引idx_c1更新c2,先獲取idx_c1 c1=5的X鎖lock_mode X locks rec but not gap,然後去申請對應主鍵id=30的行鎖，但是sess2 已經持有主鍵的行鎖，於是sess1 等待。

T3: sess2 執行根據主鍵id=30刪除記錄，需要申請id=30的行鎖以及c1=5的索引行鎖。但是sess1 以及持有該鎖，故會出現index idx_c1 of table test.tx trx id 1849 lock_mode X locks rec but not gap waiting

sess2(delete)等待sess1(update),sess1(update)等待sess2(select for update) 迴圈等待，造成死鎖。

對於RDBMS系統出現死鎖的根本原因都可以概括為:不同的事務加鎖的順序不一樣導致迴圈等待，進而導致死鎖。