并发控制是多个事务在并发运行时,数据库保证事务一致性(Consistency)和隔离性(Isolation)的一种机制。主流商用关系数据库使用的并发控制技术主要有三种:严格两阶段封锁(S2PL)、多版本并发控制(MVCC)和乐观并发控制(OCC)。
本文是PostgreSQL并发控制的第4篇,介绍了在RR(可重复读)隔离级别下MySQL和PostgreSQL的异同。
一、PostgreSQL RR Isolation Level
PostgreSQL的RR隔离级别,其实是SI隔离级别,要求满足以下两个规则:
Rule 1:事务T读取数据对象x,其中x是T启动前已提交事务产生的最新版本
Rule 2:并发事务的写集合之间不相交,否则会出现冲突,其中一个事务必须回滚
PostgreSQL使用的冲突处理协议是FUW(First Updater Wins,先更新者胜):事务Tj已持有数据对象x的锁,同时Ti希望变更x,则Ti必须等待直至Tj提交或回滚;如Tj提交,则Ti回滚,如Tj回滚,则Ti成功获取x的写锁,继续执行。
Rule 1,PostgreSQL RR隔离级别下,在启动事务时获取快照,以后该事务均使用该快照作为元组可读性的判断依据,简单来说就是在此时间点之前已提交的修改,可见,否则(包括未提交或者回滚的),不可见。
Rule 2,参见下面的例子:
时间点 | T1 | T2 |
---|---|---|
t1 | START TRANSACTION ISOLATION LEVEL REPEATABLE READ; | |
t2 | START TRANSACTION ISOLATION LEVEL REPEATABLE READ; | |
t3 | update t1 set id = 1 where id = 5; | |
t4 | update t1 set id = 11 where id = 5; | |
t5 | commit; | |
t6 | 提示出错 |
执行输出如下:
-- T1
[local]:5432 postgres@testdb=# START TRANSACTION ISOLATION LEVEL REPEATABLE READ;
START TRANSACTION
Time: 0.197 ms
-- T2
[local]:5432 postgres@testdb=# START TRANSACTION ISOLATION LEVEL REPEATABLE READ;
START TRANSACTION
Time: 0.181 ms
-- T1
[local]:5432 postgres@testdb=#* update t1 set id = 1 where id = 5;
UPDATE 1
Time: 0.430 ms
-- T2
[local]:5432 postgres@testdb=#* update t1 set id = 11 where id = 5;
---------->wait
-- T1
[local]:5432 postgres@testdb=#* commit;
COMMIT
Time: 3.241 ms
-- T2
[local]:5432 postgres@testdb=#* update t1 set id = 11 where id = 5;
ERROR: could not serialize access due to concurrent update
Time: 3172.768 ms (00:03.173)
二、MySQL RR Isolation Level
MySQL默认的隔离级别是RR
mysql> select version();
+-----------+
| version() |
+-----------+
| 8.0.21 |
+-----------+
1 row in set (0.00 sec)
mysql> show variables like '%isolation%';
+-----------------------+-----------------+
| Variable_name | Value |
+-----------------------+-----------------+
| transaction_isolation | REPEATABLE-READ |
+-----------------------+-----------------+
1 row in set (0.00 sec)
在RR隔离级别下,PostgreSQL可以保证“读不会阻塞写,写不会阻塞读”,但MySQL在RR下会出现阻塞的情况,详见下面的例子。
执行的SQL脚本:
use testdb;
CREATE TABLE tbl1(counter int);
CREATE TABLE tbl2(counter int);
SQL执行顺序:
时间点 | T1 | T2 |
---|---|---|
t1 | begin; | |
t2 | begin; | |
t3 | INSERT INTO tbl1 SELECT count(*) FROM tbl2; | |
t4 | INSERT INTO tbl2 SELECT count(*) FROM tbl1; | |
<Session Hang!> | ||
t5 | commit; | |
t6 | 执行成功 |
在Session Hang的时候使用show engine innodb status;命令查看TRANSACTIONS信息
------------
TRANSACTIONS
------------
Trx id counter 2591
Purge done for trx's n:o < 2587 undo n:o < 0 state: running but idle
History list length 2
LIST OF TRANSACTIONS FOR EACH SESSION:
---TRANSACTION 421821791990000, not started
0 lock struct(s), heap size 1136, 0 row lock(s)
---TRANSACTION 421821791988288, not started
0 lock struct(s), heap size 1136, 0 row lock(s)
---TRANSACTION 421821791987432, not started
0 lock struct(s), heap size 1136, 0 row lock(s)
---TRANSACTION 2590, ACTIVE 63 sec starting index read
mysql tables in use 2, locked 2
LOCK WAIT 2 lock struct(s), heap size 1136, 2 row lock(s)
MySQL thread id 13, OS thread handle 140346785715968, query id 52 localhost root executing
INSERT INTO tbl2 SELECT count(*) FROM tbl1
------- TRX HAS BEEN WAITING 4 SEC FOR THIS LOCK TO BE GRANTED:
RECORD LOCKS space id 9 page no 4 n bits 72 index GEN_CLUST_INDEX of table `testdb`.`tbl1` trx id 2590 lock mode S waiting
Record lock, heap no 2 PHYSICAL RECORD: n_fields 4; compact format; info bits 0
0: len 6; hex 000000020300; asc ;;
1: len 6; hex 000000000a1d; asc ;;
2: len 7; hex 81000001090110; asc ;;
3: len 4; hex 80000000; asc ;;
------------------
---TRANSACTION 2589, ACTIVE 80 sec
4 lock struct(s), heap size 1136, 2 row lock(s), undo log entries 1
MySQL thread id 12, OS thread handle 140346786010880, query id 36 localhost root
--------
注意其中的等待信息:
...
RECORD LOCKS space id 9 page no 4 n bits 72 index GEN_CLUST_INDEX of table `testdb`.`tbl1` trx id 2590 lock mode S waiting
...
可以看到,T2在等待testdb.tbl1索引GEN_CLUST_INDEX(索引组织表内部创建的索引)上的共享锁,无法获取是因为需要等待T1持有tbl1索引GEN_CLUST_INDEX的RECORD X锁。
mysql> SELECT ENGINE_LOCK_ID,ENGINE_TRANSACTION_ID,object_schema,object_name,index_name,LOCK_TYPE,lock_mode,lock_status,lock_data FROM performance_schema.data_locks where object_name='tbl1' and index_name is not null order by ENGINE_TRANSACTION_ID;
+---------------------------------------+-----------------------+---------------+-------------+-----------------+-----------+---------------+-------------+----------------+
| ENGINE_LOCK_ID | ENGINE_TRANSACTION_ID | object_schema | object_name | index_name | LOCK_TYPE | lock_mode | lock_status | lock_data |
+---------------------------------------+-----------------------+---------------+-------------+-----------------+-----------+---------------+-------------+----------------+
| 140346815278488:9:4:2:140346713517384 | 2589 | testdb | tbl1 | GEN_CLUST_INDEX | RECORD | X,REC_NOT_GAP | GRANTED | 0x000000020300 |
| 140346815280200:9:4:2:140346713529984 | 2590 | testdb | tbl1 | GEN_CLUST_INDEX | RECORD | S | WAITING | 0x000000020300 |
+---------------------------------------+-----------------------+---------------+-------------+-----------------+-----------+---------------+-------------+----------------+
2 rows in set (0.00 sec)
PostgreSQL使用heap table,没有“GEN_CLUST_INDEX”这一数据结构,自然也无需对该数据结构进行并发控制,而MySQL使用索引组织表,提升读取性能的同时但需要额外对这一数据结构进行管理和维护。
除了写可能会阻塞读之外,MySQL还有一些让PGer诧异的现象。
测试SQL脚本:
mysql> drop table tbl;
Query OK, 0 rows affected (0.03 sec)
mysql> CREATE TABLE tbl(id int);
Query OK, 0 rows affected (0.03 sec)
mysql> INSERT INTO tbl VALUES (1),(2),(3),(4);
Query OK, 4 rows affected (0.01 sec)
Records: 4 Duplicates: 0 Warnings: 0
执行顺序:
时间点 | T1 | T2 |
---|---|---|
t1 | begin; | |
t2 | begin; | |
t3 | UPDATE tbl SET id=id-1; | |
t4 | SELECT * FROM tbl; | |
t5 | DELETE FROM tbl WHERE id=4; | |
Session Wait! | ||
t6 | commit; | |
t7 | select * from tbl; | |
t8 | DELETE FROM tbl WHERE id=4; |
执行结果是:
...
-- T1
mysql> select * from tbl;
+------+
| id |
+------+
| 0 |
| 1 |
| 2 |
| 3 |
+------+
4 rows in set (0.00 sec)
-- T2
...
mysql> select * from tbl;
+------+
| id |
+------+
| 1 |
| 2 |
| 3 |
| 4 |
+------+
4 rows in set (0.00 sec)
mysql> DELETE FROM tbl WHERE id=4;
Query OK, 0 rows affected (0.00 sec)
查询结果满足RR的要求返回先前的元组版本,这本身没有问题,但对PGer来说不太好理解的地方是id = 4这条记录查询时明明存在,但执行delete时却找不到该记录,结果返回0行。
另外值得一提的是,相对于PostgreSQL,MySQL有很”丰富“的锁类型,从锁本身的语义出发来理解锁虽然直观但看不到背后的原理,如果从并发控制的角度来看MySQL的锁,可以理解这些锁存在的价值或者意义,会有不一样的认识。
三、参考资料
[1] MySQL Document, InnoDB Locking,GEN_CLUST_INDEX...
[2] Daniel Verite,Isolation Repeatable Read in PostgreSQL versus MySQL