事务库事务隔离级别
為了快速同步數(shù)據(jù)的需要,我分段執(zhí)行了兩次python腳本,即開啟了兩個進程同步數(shù)據(jù),結果服務器不時報出數(shù)據(jù)庫死鎖異常,通過排查代碼和數(shù)據(jù)庫日志發(fā)現(xiàn),是由長事務并發(fā)引起的。代碼中有入賬和出賬兩個方法,里面涉及操作較多,都為其加了事務,拋出異常時可自動回滾,采用數(shù)據(jù)庫(mysql)默認的隔離級別(Repeatable read)。提到并發(fā),一般就會想到用同步代碼塊的方法的處理,但是由于項目是分布式的,共用一個主庫,單單在代碼加鎖是不能保證數(shù)據(jù)的準確的,那就只能在數(shù)據(jù)庫層面去考慮加鎖了。由于數(shù)據(jù)量暫時不大,一開始我的解決方法是將隔離級別調整為最高(Serializable),這樣雖然代價較大,但是能保證數(shù)據(jù)的準確性,數(shù)據(jù)庫的鎖也會相互等待,但當我再重新同步數(shù)據(jù)時,還是報出了大量死鎖異常(com.mysql.jdbc.exceptions.jdbc4.MySQLTransactionRollbackException: Deadlock found when trying to get lock; try restarting transaction),仔細追蹤數(shù)據(jù)庫日志,發(fā)現(xiàn)有兩個鎖在兩條update語句上互相等待(操作的是一張統(tǒng)計表),但這兩條update語句不是落到同一行上的,因為查詢條件并不一樣。那就奇怪了,一般來說是行鎖的問題,最后排查發(fā)現(xiàn),是該表未加索引的原因,造成了事務鎖住了這張表,而由于多個事務是同時執(zhí)行到這里,造成了鎖的互相等待,最終數(shù)據(jù)庫選擇回滾了其中一個事務。發(fā)現(xiàn)問題后,我將表加了索引,再運行時,已經(jīng)沒報這個錯誤了(子查詢也是會導致鎖表的)。為了保證代碼的運行效率,我將數(shù)據(jù)庫級別設置為Repeatable read,但隨之而來發(fā)現(xiàn)了一些數(shù)據(jù)出現(xiàn)了幻讀,最后我是在update語句那里加了一個beforeMoney的過濾(update wallet_stat set money = afterMoney where id = xxx and money = beforeMoney),這樣就能保證數(shù)據(jù)的準確性了,不過這里不足的是,雖保證了數(shù)據(jù)的正確,但并發(fā)的那個事務卻必須回滾。這里也可以考慮借助隊列或者樂觀鎖等方式來解決這個問題,就不一一詳述了。
以上就是我處理這個問題的全過程,為了查找解決方案,我在查找了很多文章,獲益良多。以下是我認為比較好的資料:
什么是事務:
數(shù)據(jù)庫事務的隔離級別有4個,由低到高依次為Read uncommitted、Read committed、Repeatable read、Serializable,這四個級別可以逐個解決臟讀、不可重復讀、幻讀這幾類問題。
???????????????????????????????????????????????????????????????????????
注意:我們討論隔離級別的場景,主要是在多個事務并發(fā)的情況下,因此,接下來的講解都圍繞事務并發(fā)。
Read uncommitted 讀未提交
公司發(fā)工資了,領導把5000元打到singo的賬號上,但是該事務并未提交,而singo正好去查看賬戶,發(fā)現(xiàn)工資已經(jīng)到賬,是5000元整,非常高興。可是不幸的是,領導發(fā)現(xiàn)發(fā)給singo的工資金額不對,是2000元,于是迅速回滾了事務,修改金額后,將事務提交,最后singo實際的工資只有2000元,singo空歡喜一場。
出現(xiàn)上述情況,即我們所說的臟讀,兩個并發(fā)的事務,“事務A:領導給singo發(fā)工資”、“事務B:singo查詢工資賬戶”,事務B讀取了事務A尚未提交的數(shù)據(jù)。
當隔離級別設置為Read uncommitted時,就可能出現(xiàn)臟讀,如何避免臟讀,請看下一個隔離級別。
Read committed 讀提交
singo拿著工資卡去消費,系統(tǒng)讀取到卡里確實有2000元,而此時她的老婆也正好在網(wǎng)上轉賬,把singo工資卡的2000元轉到另一賬戶,并在singo之前提交了事務,當singo扣款時,系統(tǒng)檢查到singo的工資卡已經(jīng)沒有錢,扣款失敗,singo十分納悶,明明卡里有錢,為何......
出現(xiàn)上述情況,即我們所說的不可重復讀,兩個并發(fā)的事務,“事務A:singo消費”、“事務B:singo的老婆網(wǎng)上轉賬”,事務A事先讀取了數(shù)據(jù),事務B緊接了更新了數(shù)據(jù),并提交了事務,而事務A再次讀取該數(shù)據(jù)時,數(shù)據(jù)已經(jīng)發(fā)生了改變。
當隔離級別設置為Read committed時,避免了臟讀,但是可能會造成不可重復讀。
大多數(shù)數(shù)據(jù)庫的默認級別就是Read committed,比如Sql Server , Oracle。如何解決不可重復讀這一問題,請看下一個隔離級別。
Repeatable read 重復讀
當隔離級別設置為Repeatable read時,可以避免不可重復讀。當singo拿著工資卡去消費時,一旦系統(tǒng)開始讀取工資卡信息(即事務開始),singo的老婆就不可能對該記錄進行修改,也就是singo的老婆不能在此時轉賬。
雖然Repeatable read避免了不可重復讀,但還有可能出現(xiàn)幻讀。
singo的老婆工作在銀行部門,她時常通過銀行內部系統(tǒng)查看singo的信用卡消費記錄。有一天,她正在查詢到singo當月信用卡的總消費金額(select sum(amount) from transaction where month = 本月)為80元,而singo此時正好在外面胡吃海塞后在收銀臺買單,消費1000元,即新增了一條1000元的消費記錄(insert transaction ... ),并提交了事務,隨后singo的老婆將singo當月信用卡消費的明細打印到A4紙上,卻發(fā)現(xiàn)消費總額為1080元,singo的老婆很詫異,以為出現(xiàn)了幻覺,幻讀就這樣產(chǎn)生了。
注:mysql的默認隔離級別就是Repeatable read。
Serializable 序列化
Serializable是最高的事務隔離級別,同時代價也花費最高,性能很低,一般很少使用,在該級別下,事務順序執(zhí)行,不僅可以避免臟讀、不可重復讀,還避免了幻像讀。
?
mysql鎖阻塞分析:
查看鎖阻塞線程信息
這里用幾中方法進行分析:
使用show processlist查看
MySQL [(none)]> show processlist;
+----+------+-----------+------+---------+------+--------------+------------------------------------------+
| Id | User | Host | db | Command | Time | State | Info |
+----+------+-----------+------+---------+------+--------------+------------------------------------------+
| 2 | root | localhost | NULL | Query | 0 | init | show processlist |
| 3 | root | localhost | test | Query | 70 | Sending data | select count(*) from t3 a,t3 b |
| 4 | root | localhost | test | Query | 65 | updating | delete from emp where empno=7788 |
| 7 | root | localhost | test | Query | 68 | updating | update emp set sal=3500 where empno=7788 |
+----+------+-----------+------+---------+------+--------------+------------------------------------------+
4 rows in set (0.00 sec)
如果數(shù)據(jù)庫存在較多線程的話,這種方法確實不太好確認的。
直接使用show engine innodb status查看
------------
TRANSACTIONS
------------
Trx id counter 4131
Purge done for trx's n:o < 4119 undo n:o < 0 state: running but idle
History list length 126
LIST OF TRANSACTIONS FOR EACH SESSION:
---TRANSACTION 0, not started
MySQL thread id 2, OS thread handle 0x7f953ffff700, query id 115 localhost root init
show engine innodb status
---TRANSACTION 4130, ACTIVE 41 sec starting index read
mysql tables in use 1, locked 1
LOCK WAIT 2 lock struct(s), heap size 360, 1 row lock(s)
MySQL thread id 4, OS thread handle 0x7f953ff9d700, query id 112 localhost root updating
delete from emp where empno=7788
------- TRX HAS BEEN WAITING 41 SEC FOR THIS LOCK TO BE GRANTED: ## 等待了41s
RECORD LOCKS space id 16 page no 3 n bits 88 index `PRIMARY` of table `test`.`emp` trx id 4130 lock_mode X locks rec but not gap waiting
Record lock, heap no 9 PHYSICAL RECORD: n_fields 10; compact format; info bits 0 ## 線程4在等待往test.emp中的主鍵上加X鎖,page num=3
0: len 4; hex 80001e6c; asc l;;
1: len 6; hex 000000001018; asc ;;
2: len 7; hex 91000001420084; asc B ;;
3: len 5; hex 53434f5454; asc SCOTT;;
4: len 7; hex 414e414c595354; asc ANALYST;;
5: len 4; hex 80001d8e; asc ;;
6: len 4; hex 208794f0; asc ;;
7: len 4; hex 80000bb8; asc ;;
8: SQL NULL;
9: len 4; hex 80000014; asc ;;
------------------
---TRANSACTION 4129, ACTIVE 45 sec starting index read
mysql tables in use 1, locked 1
LOCK WAIT 2 lock struct(s), heap size 360, 1 row lock(s)
MySQL thread id 7, OS thread handle 0x7f953ff6c700, query id 111 localhost root updating
update emp set sal=3500 where empno=7788
------- TRX HAS BEEN WAITING 45 SEC FOR THIS LOCK TO BE GRANTED: ## 等待了45s
RECORD LOCKS space id 16 page no 3 n bits 88 index `PRIMARY` of table `test`.`emp` trx id 4129 lock_mode X locks rec but not gap waiting
Record lock, heap no 9 PHYSICAL RECORD: n_fields 10; compact format; info bits 0 ## 線程7在等待往test.emp中的主鍵上加X鎖,page num=3
0: len 4; hex 80001e6c; asc l;;
1: len 6; hex 000000001018; asc ;;
2: len 7; hex 91000001420084; asc B ;;
3: len 5; hex 53434f5454; asc SCOTT;;
4: len 7; hex 414e414c595354; asc ANALYST;;
5: len 4; hex 80001d8e; asc ;;
6: len 4; hex 208794f0; asc ;;
7: len 4; hex 80000bb8; asc ;;
8: SQL NULL;
9: len 4; hex 80000014; asc ;;
------------------
---TRANSACTION 4128, ACTIVE 51 sec
2 lock struct(s), heap size 360, 1 row lock(s)
MySQL thread id 3, OS thread handle 0x7f953ffce700, query id 110 localhost root cleaning up
我們知道,主要根因還是thread=3引起的,但從innodb status中卻無法分析得到這個結果。
從上面來看,線程4和線程7都在等待往test.emp中的主鍵上加X鎖,page num=3,但是線程7等待的時間為45s,而線程4等待的時間為41s,是較線程7之后申請的鎖,所以可以判斷是線程7阻塞了線程4。至于線程7為什么出現(xiàn)等待,這里分析不到根因。
使用mysqladmin debug查看
# mysqladmin -S /tmp/mysql3306.sock debug
然后在error日志中,會看到:
Thread database.table_name Locked/Waiting Lock_type
3 test.t3 Locked - read Low priority read lock
7 test.emp Locked - write High priority write lock
這種方法中,能找到線程ID=3和7是阻塞者,但還是不太準確,判斷不出來線程7也是被線程ID=3阻塞的。
使用innodb_lock_monitor來獲取阻塞鎖線程
MySQL [test]> CREATE TABLE innodb_lock_monitor (a INT) ENGINE=INNODB; ## 隨便在一個數(shù)據(jù)庫中創(chuàng)建這個表,就會打開lock monitor
Query OK, 0 rows affected, 1 warning (0.07 sec)
MySQL [test]> show warnings\G
*************************** 1. row ***************************
Level: Warning
Code: 131
Message: Using the table name innodb_lock_monitor to enable diagnostic output is deprecated and may be removed in future releases. Use INFORMATION_SCHEMA or PERFORMANCE_SCHEMA tables or SET GLOBAL innodb_status_output=ON.
1 row in set (0.00 sec)
說明:這個在5.6中有一個warning,但不影響使用。
然后再使用show engine innodb status查看:
------------
TRANSACTIONS
------------
Trx id counter 4667
Purge done for trx's n:o < 4659 undo n:o < 0 state: running but idle
History list length 138
LIST OF TRANSACTIONS FOR EACH SESSION:
---TRANSACTION 0, not started
MySQL thread id 9, OS thread handle 0x7f813c5f7700, query id 152 localhost root init
show engine innodb status
---TRANSACTION 4663, ACTIVE 78 sec starting index read
mysql tables in use 1, locked 1
LOCK WAIT 2 lock struct(s), heap size 360, 1 row lock(s)
MySQL thread id 4, OS thread handle 0x7f813c628700, query id 149 localhost root updating
delete from emp where empno=7788
------- TRX HAS BEEN WAITING 78 SEC FOR THIS LOCK TO BE GRANTED: ## 等待了78s
RECORD LOCKS space id 16 page no 3 n bits 88 index `PRIMARY` of table `test`.`emp` trx id 4663 lock_mode X locks rec but not gap waiting
Record lock, heap no 9 PHYSICAL RECORD: n_fields 10; compact format; info bits 0 ## 線程4在等待往test.emp中的主鍵上加X鎖,page num=3
0: len 4; hex 80001e6c; asc l;;
1: len 6; hex 000000001018; asc ;;
2: len 7; hex 91000001420084; asc B ;;
3: len 5; hex 53434f5454; asc SCOTT;;
4: len 7; hex 414e414c595354; asc ANALYST;;
5: len 4; hex 80001d8e; asc ;;
6: len 4; hex 208794f0; asc ;;
7: len 4; hex 80000bb8; asc ;;
8: SQL NULL;
9: len 4; hex 80000014; asc ;;
------------------
TABLE LOCK table `test`.`emp` trx id 4663 lock mode IX ## 在給主鍵行上加X鎖之前,先要在表上加意向鎖IX
RECORD LOCKS space id 16 page no 3 n bits 88 index `PRIMARY` of table `test`.`emp` trx id 4663 lock_mode X locks rec but not gap waiting
Record lock, heap no 9 PHYSICAL RECORD: n_fields 10; compact format; info bits 0
0: len 4; hex 80001e6c; asc l;;
1: len 6; hex 000000001018; asc ;;
2: len 7; hex 91000001420084; asc B ;;
3: len 5; hex 53434f5454; asc SCOTT;;
4: len 7; hex 414e414c595354; asc ANALYST;;
5: len 4; hex 80001d8e; asc ;;
6: len 4; hex 208794f0; asc ;;
7: len 4; hex 80000bb8; asc ;;
8: SQL NULL;
9: len 4; hex 80000014; asc ;;
---TRANSACTION 4662, ACTIVE 81 sec starting index read
mysql tables in use 1, locked 1
LOCK WAIT 2 lock struct(s), heap size 360, 1 row lock(s)
MySQL thread id 7, OS thread handle 0x7f813c5c6700, query id 148 localhost root updating
update emp set sal=3500 where empno=7788
------- TRX HAS BEEN WAITING 81 SEC FOR THIS LOCK TO BE GRANTED: ## 等待了81s
RECORD LOCKS space id 16 page no 3 n bits 88 index `PRIMARY` of table `test`.`emp` trx id 4662 lock_mode X locks rec but not gap waiting
Record lock, heap no 9 PHYSICAL RECORD: n_fields 10; compact format; info bits 0 ## 線程7在等待往test.emp中的主鍵上加X鎖,page num=3
0: len 4; hex 80001e6c; asc l;;
1: len 6; hex 000000001018; asc ;;
2: len 7; hex 91000001420084; asc B ;;
3: len 5; hex 53434f5454; asc SCOTT;;
4: len 7; hex 414e414c595354; asc ANALYST;;
5: len 4; hex 80001d8e; asc ;;
6: len 4; hex 208794f0; asc ;;
7: len 4; hex 80000bb8; asc ;;
8: SQL NULL;
9: len 4; hex 80000014; asc ;;
------------------
TABLE LOCK table `test`.`emp` trx id 4662 lock mode IX ## 在給主鍵行上加X鎖之前,先要在表上加意向鎖IX
RECORD LOCKS space id 16 page no 3 n bits 88 index `PRIMARY` of table `test`.`emp` trx id 4662 lock_mode X locks rec but not gap waiting
Record lock, heap no 9 PHYSICAL RECORD: n_fields 10; compact format; info bits 0
0: len 4; hex 80001e6c; asc l;;
1: len 6; hex 000000001018; asc ;;
2: len 7; hex 91000001420084; asc B ;;
3: len 5; hex 53434f5454; asc SCOTT;;
4: len 7; hex 414e414c595354; asc ANALYST;;
5: len 4; hex 80001d8e; asc ;;
6: len 4; hex 208794f0; asc ;;
7: len 4; hex 80000bb8; asc ;;
8: SQL NULL;
9: len 4; hex 80000014; asc ;;
---TRANSACTION 4615, ACTIVE 1579 sec, thread declared inside InnoDB 1222
mysql tables in use 2, locked 0
2 lock struct(s), heap size 360, 1 row lock(s)
MySQL thread id 3, OS thread handle 0x7f813c659700, query id 147 localhost root Sending data
select count(*) from t3 a,t3 b ## 這是線程3當前正在執(zhí)行的SQL
Trx read view will not see trx with id >= 4662, sees < 4659
TABLE LOCK table `test`.`emp` trx id 4615 lock mode IX ## 線程3中正在擁有表上的意向IX鎖,并且有test.emp表上主鍵的行級X鎖,page num=3
RECORD LOCKS space id 16 page no 3 n bits 88 index `PRIMARY` of table `test`.`emp` trx id 4615 lock_mode X locks rec but not gap
Record lock, heap no 9 PHYSICAL RECORD: n_fields 10; compact format; info bits 0
0: len 4; hex 80001e6c; asc l;;
1: len 6; hex 000000001018; asc ;;
2: len 7; hex 91000001420084; asc B ;;
3: len 5; hex 53434f5454; asc SCOTT;;
4: len 7; hex 414e414c595354; asc ANALYST;;
5: len 4; hex 80001d8e; asc ;;
6: len 4; hex 208794f0; asc ;;
7: len 4; hex 80000bb8; asc ;;
8: SQL NULL;
9: len 4; hex 80000014; asc ;;
為什么線程3當前執(zhí)行的是一個select t3表操作,但卻鎖住了test.emp表上page num=3?
有可能是線程3之前對test.emp表的操作事務沒有及時提交導致。
所以得出:線程3阻塞了線程7,而線程7又阻塞了線程4,所以根因就是線程3,讓線程3盡快提交或是kill掉即可。
結論
在分析innodb中鎖阻塞時,幾種方法的對比情況:
(1)使用show processlist查看不靠譜;
(2)直接使用show engine innodb status查看,無法判斷到問題的根因;
(3)使用mysqladmin debug查看,能看到所有產(chǎn)生鎖的線程,但無法判斷哪個才是根因;
(4)開啟innodb_lock_monitor后,再使用show engine innodb status查看,能夠找到鎖阻塞的根因。
?
注:文章出自
?http://blog.csdn.net/fg2006/article/details/6937413
?http://blog.csdn.net/hw_libo/article/details/39080809
轉載于:https://www.cnblogs.com/zhuanghuang/p/5516041.html
總結
- 上一篇: Fibonacci数列的java实现
- 下一篇: oracle操作