1）mvcc多版本控制提高讀寫qps

2) REPEATBLE READ 級別並不能完全避免幻讀，需要加next key locks，可以使顯示鎖(select * where * for update   or lock in share mode)

一些文章寫到InnoDB的可重複讀避免了“幻讀”（phantom read），這個說法並不準確。

做個試驗：(以下所有試驗要注意儲存引擎和隔離級別)

mysql>show create table t_bitfly/G;
CREATE TABLE `t_bitfly` (
`id` bigint(20) NOT NULL default '0',
`value` varchar(32) default NULL,
PRIMARY KEY (`id`)
) ENGINE=InnoDB DEFAULT CHARSET=gbk

mysql>select @@global.tx_isolation, @@tx_isolation;
+-----------------------+-----------------+
| @@global.tx_isolation | @@tx_isolation  |
+-----------------------+-----------------+
| REPEATABLE-READ       | REPEATABLE-READ |
+-----------------------+-----------------+

試驗4-1：

tSessionA                                                     Session B
|
| START TRANSACTION;                           START TRANSACTION;
|
| SELECT * FROM t_bitfly;
| empty set
|                                INSERT INTO t_bitfly VALUES (1, 'a');
|                             
|
| SELECT * FROM t_bitfly;
| empty set
|                                                                      COMMIT;
|
| SELECT * FROM t_bitfly;
| empty set
|
| INSERT INTO t_bitfly VALUES (1, 'a');
| ERROR 1062 (23000):
| Duplicate entry '1' for key 1
v (shit,剛剛明明告訴我沒有這條記錄的)

如此就出現了幻讀，以為表裡沒有資料，其實資料已經存在了，傻乎乎的提交後，才發現數據衝突了。

試驗4-2：

tSessionA               Session B
|
| START TRANSACTION;           START TRANSACTION;
|
| SELECT * FROM t_bitfly;
| +------+-------+
| | id   | value |
| +------+-------+
| |    1 |a     |
| +------+-------+
|                                       INSERT INTO t_bitfly VALUES (2, 'b');
|                            
|
| SELECT * FROM t_bitfly;
| +------+-------+
| | id   | value |
| +------+-------+
| |    1 |a     |
| +------+-------+
|                               COMMIT;
|
| SELECT * FROM t_bitfly;
| +------+-------+
| | id   | value |
| +------+-------+
| |    1 |a     |
| +------+-------+
|
| UPDATE t_bitfly SET value='z';
| Rows matched: 2  Changed:2  Warnings: 0
| (怎麼多出來一行)
|
| SELECT * FROM t_bitfly;
| +------+-------+
| | id   | value |
| +------+-------+
| |    1 |z     |
| |    2 |z     |
| +------+-------+
|
v

本事務中第一次讀取出一行，做了一次更新後，另一個事務裡提交的資料就出現了。也可以看做是一種幻讀。

------

那麼，InnoDB指出的可以避免幻讀是怎麼回事呢？

http://dev.mysql.com/doc/refman/5.0/en/innodb-record-level-locks.html

By default, InnoDB operatesin REPEATABLE READ transaction isolation level and with the innodb_locks_unsafe_for_binlogsystem variable disabled. In this case, InnoDB uses next-key locks for searchesand index scans, which prevents phantom rows (see Section 13.6.8.5, “Avoidingthe Phantom Problem Using Next-Key Locking”).

準備的理解是，當隔離級別是可重複讀，且禁用innodb_locks_unsafe_for_binlog的情況下，在搜尋和掃描index的時候使用的next-keylocks可以避免幻讀。

關鍵點在於，是InnoDB預設對一個普通的查詢也會加next-key locks，還是說需要應用自己來加鎖呢？如果單看這一句，可能會以為InnoDB對普通的查詢也加了鎖，如果是，那和序列化（SERIALIZABLE）的區別又在哪裡呢？

MySQL manual裡還有一段：

13.2.8.5. Avoiding the PhantomProblem Using Next-Key Locking (http://dev.mysql.com/doc/refman/5.0/en/innodb-next-key-locking.html)

Toprevent phantoms, InnoDB usesan algorithm called next-key locking that combinesindex-row locking with gap locking.

Youcan use next-key locking to implement a uniqueness check in your application:If you read your data in share mode and do not see a duplicate for a row youare going to insert, then you can safely insert your row and know that thenext-key lock set on the successor of your row during the read prevents anyonemeanwhile inserting a duplicate for your row. Thus, the next-key lockingenables you to “lock” the nonexistence of something in your table.

我的理解是說，InnoDB提供了next-key locks，但需要應用程式自己去加鎖。manual裡提供一個例子：

SELECT * FROM child WHERE id> 100 FOR UPDATE;

這樣，InnoDB會給id大於100的行（假如child表裡有一行id為102），以及100-102，102+的gap都加上鎖。

可以使用showinnodb status來檢視是否給表加上了鎖。

再看一個實驗，要注意，表t_bitfly裡的id為主鍵欄位。

實驗4-3：

t SessionA             Session B
|
| START TRANSACTION;                START TRANSACTION;
|
| SELECT * FROM t_bitfly
| WHERE id<=1
| FOR UPDATE;
| +------+-------+
| | id   | value |
| +------+-------+
| |    1 |a     |
| +------+-------+
|                                      INSERT INTO t_bitfly   VALUES (2, 'b');
|                            Query OK, 1 row affected
|
| SELECT * FROM t_bitfly;
| +------+-------+
| | id   | value |
| +------+-------+
| |    1 |a     |
| +------+-------+
|                              INSERT INTO t_bitfly
|                              VALUES (0, '0');
|                                  (waiting for lock ...
|                               then timeout)
|                                   ERROR 1205 (HY000):
|                                  Lock wait timeout exceeded;
|                                   try restarting transaction
|
| SELECT * FROM t_bitfly;
| +------+-------+
| | id   | value |
| +------+-------+
| |    1 |a     |
| +------+-------+
|                           COMMIT;
|
| SELECT * FROM t_bitfly;
| +------+-------+
| | id   | value |
| +------+-------+
| |    1 |a     |
| +------+-------+
v

可以看到，用id<=1加的鎖，只鎖住了id<=1的範圍，可以成功新增id為2的記錄，新增id為0的記錄時就會等待鎖的釋放。

MySQL manual裡對可重複讀裡的鎖的詳細解釋：

http://dev.mysql.com/doc/refman/5.0/en/set-transaction.html#isolevel_repeatable-read

Forlocking reads (SELECT with FORUPDATE or LOCK IN SHARE MODE),UPDATE, and DELETE statements, lockingdepends on whether the statement uses a unique index with a unique searchcondition, or a range-type search condition. For a unique index with a uniquesearch condition, InnoDB locksonly the index record found, not the gap before it. For other searchconditions, InnoDB locksthe index range scanned, using gap locks or next-key (gap plus index-record)locks to block insertions by other sessions into the gaps covered by the range.

------

一致性讀和提交讀，先看實驗，

實驗4-4：

tSessionA              Session B
|
| STARTTRANSACTION;             START TRANSACTION;
|
| SELECT * FROM t_bitfly;
| +----+-------+
| | id | value |
| +----+-------+
| |  1 |a     |
| +----+-------+
|                              INSERT INTO t_bitfly   VALUES (2, 'b');
|                              
|                              COMMIT;
|
| SELECT * FROM t_bitfly;
| +----+-------+
| | id | value |
| +----+-------+
| |  1 |a     |
| +----+-------+
|
| SELECT * FROM t_bitfly LOCK IN SHARE MODE;
| +----+-------+
| | id | value |
| +----+-------+
| |  1 |a     |
| |  2 |b     |
| +----+-------+
|
| SELECT * FROM t_bitfly FOR UPDATE;
| +----+-------+
| | id | value |
| +----+-------+
| |  1 |a     |
| |  2 |b     |
| +----+-------+
|
| SELECT * FROM t_bitfly;
| +----+-------+
| | id | value |
| +----+-------+
| |  1 |a     |
| +----+-------+
v

如果使用普通的讀，會得到一致性的結果，如果使用了加鎖的讀，就會讀到“最新的”“提交”讀的結果。

本身，可重複讀和提交讀是矛盾的。在同一個事務裡，如果保證了可重複讀，就會看不到其他事務的提交，違背了提交讀；如果保證了提交讀，就會導致前後兩次讀到的結果不一致，違背了可重複讀。

可以這麼講，InnoDB提供了這樣的機制，在預設的可重複讀的隔離級別裡，可以使用加鎖讀去查詢最新的資料。

http://dev.mysql.com/doc/refman/5.0/en/innodb-consistent-read.html

Ifyou want to see the “freshest” state of the database, you should use either theREAD COMMITTED isolation level or a locking read:
SELECT * FROM t_bitfly LOCK IN SHARE MODE;

------

結論：MySQLInnoDB的可重複讀並不保證避免幻讀，需要應用使用加鎖讀來保證。而這個加鎖度使用到的機制就是next-keylocks。