作者：胡呈清
愛可生 DBA 團隊成員，擅長故障分析、效能優化，個人部落格：https://www.jianshu.com/u/a95...，歡迎討論。
本文來源：原創投稿
*愛可生開源社群出品，原創內容未經授權不得隨意使用，轉載請聯絡小編並註明來源。
已知 change buffer 的原理
參考資料：https://juejin.im/post/684490...
對於普通二級索引，當插入、修改、刪除二級索引記錄時，即使資料不在 innodb buffer pool 中，也不需要先把資料從磁碟讀取到記憶體。只需要在 change buffer 中完成 DML 操作，下次讀取時才會從磁碟讀取資料頁到記憶體，並與 change buffer 進行 merge，從而得到正確的資料。這減少了 DML 時的隨機 IO。
疑問
按照上述原理，使用 change buffer 二級索引不需要讀取磁碟，那 delete、update 是如何得到 affected rows 的？
答
不妨先作出假設：
- 如果 delete、update 是以主鍵、唯一索引做為篩選條件，則讀取磁碟或者 innodb buffer pool 中的主鍵、唯一索引來確定 affected rows。對於普通索引頁上記錄的刪除或者修改，還是直接使用 change buffer，不需要單獨將普通索引頁從磁碟上讀取到記憶體。
- 如果 delete、update 是以普通二級索引做為篩選條件，以 delete 為例（update 內部實現是先 delete 再 insert）：delete from t where a=100; 如果索引頁不在記憶體中，則需要先從磁碟讀取 a 索引，找到 a = 100 的記錄對應的 id（主鍵值），再從磁碟掃描主鍵索引（回表）將 id 滿足條件的記錄讀取到記憶體。然後在 innodb buffer pool 中把對應的主鍵索引頁、二級索引頁中的記錄刪除。這裡不使用 change buffer。
驗證
接下來設計兩個實驗來驗證上述假設。
實驗1-以主鍵為篩選條件做 delete
用 sysbench 造一張 100 萬行的表，表中有一個主鍵和一個普通索引：
CREATE TABLE `sbtest1` (
`id` int NOT NULL AUTO_INCREMENT,
`k` int NOT NULL DEFAULT '0',
`c` char(120) COLLATE utf8mb4_bin NOT NULL DEFAULT '',
`pad` char(60) COLLATE utf8mb4_bin NOT NULL DEFAULT '',
PRIMARY KEY (`id`),
KEY `k_2` (`k`)
);重啟 mysqld ，清空 innodb buffer pool，注意引數：
innodb_buffer_pool_size = 64M
innodb_buffer_pool_load_at_startup = 0
innodb_buffer_pool_dump_at_shutdown = 0
innodb_buffer_pool_dump_pct = 0執行 delete，並使用show engine innodb status\G觀察INSERT BUFFER AND ADAPTIVE HASH INDEX 部分資訊，判斷是否使用 change buffer：
mysql> delete from sbtest1 where id=1;
Query OK, 1 row affected (0.00 sec)
mysql> show engine innodb status\G
-------------------------------------
INSERT BUFFER AND ADAPTIVE HASH INDEX
-------------------------------------
Ibuf: size 1, free list len 29, seg size 31, 1 merges
merged operations:
insert 0, delete mark 1, delete 0
discarded operations:
insert 0, delete mark 0, delete 0
mysql> delete from sbtest1 where id=2;
Query OK, 1 row affected (0.00 sec)
-------------------------------------
INSERT BUFFER AND ADAPTIVE HASH INDEX
-------------------------------------
Ibuf: size 1, free list len 29, seg size 31, 2 merges
merged operations:
insert 0, delete mark 2, delete 0
discarded operations:
insert 0, delete mark 0, delete 0
mysql> delete from sbtest1 where id=3;
Query OK, 1 row affected (0.00 sec)
-------------------------------------
INSERT BUFFER AND ADAPTIVE HASH INDEX
-------------------------------------
Ibuf: size 1, free list len 29, seg size 31, 3 merges
merged operations:
insert 0, delete mark 3, delete 0
discarded operations:
insert 0, delete mark 0, delete 0
mysql> select * from sbtest1 where id=4;
mysql> delete from sbtest1 where id=4;
-------------------------------------
INSERT BUFFER AND ADAPTIVE HASH INDEX
-------------------------------------
Ibuf: size 1, free list len 29, seg size 31, 4 merges
merged operations:
insert 0, delete mark 4, delete 0
discarded operations:
insert 0, delete mark 0, delete 0上述實驗說明：如果 delete 是以主鍵做為篩選條件，對於普通索引k，如果索引頁不在記憶體中（select * from sbtest1 where id=4 讀取的只是主鍵索引頁，不會讀取k索引頁），會使用 change buffer（每次 delete 後，delete mark 都增加1）。
實驗2-以普通索引為篩選條件做 delete
重新造資料，重啟 mysqld 清空 buffer pool。下面實驗結果說明：如果 delete 以普通索引做為篩選條件，對於普通索引k，如果索引頁不在記憶體中，不會使用 change buffer。言外之意就是需要讀取磁碟了。
##delete where id=1，delete mark +1，說明使用了change buffer
mysql> delete from sbtest1 where id=1;
Query OK, 1 row affected (0.01 sec)
-------------------------------------
INSERT BUFFER AND ADAPTIVE HASH INDEX
-------------------------------------
Ibuf: size 1, free list len 29, seg size 31, 1 merges
merged operations:
insert 0, delete mark 1, delete 0
discarded operations:
insert 0, delete mark 0, delete 0
##delete where k=367246，delete mark 不變，說明沒有使用change buffer
mysql> select * from sbtest1 where id=2;
+----+--------+-------------------------------------------------------------------------------------------------------------------------+-------------------------------------------------------------+
| id | k | c | pad |
+----+--------+-------------------------------------------------------------------------------------------------------------------------+-------------------------------------------------------------+
| 2 | 367246 | 42909700340-70078987867-62357124096-35495169193-85675377266-14643719347-30417020186-80900182681-50382374444-66260611196 | 74781290517-41121402981-50604677924-34464478849-89102349959 |
+----+--------+-------------------------------------------------------------------------------------------------------------------------+-------------------------------------------------------------+
1 row in set (0.00 sec)
mysql> delete from sbtest1 where k=367246;
Query OK, 1 row affected (0.01 sec)
-------------------------------------
INSERT BUFFER AND ADAPTIVE HASH INDEX
-------------------------------------
Ibuf: size 1, free list len 29, seg size 31, 1 merges
merged operations:
insert 0, delete mark 1, delete 0
discarded operations:
insert 0, delete mark 0, delete 0