简体中文(ZH-CN) English(EN) 繁体中文(ZH-TW) 日本語(JA) 한국어(KO) Melayu(MS) Français(FR) Deutsch(DE)
首頁 > 資料庫 > mysql教程 > 主體

mysql merge union merge sort_union 的不同

伊谢尔伦
發布： 2016-11-21 15:20:24
原創
1392 人瀏覽過

看到MYSQL手冊的Index Merge Optimization，不禁有一些想法，所以記錄如下文

先來解釋下2種方式不同：
這兩種方式都使用一個table中的不同二級索引進行，注意是單個表。
merge union :在使用or的時候如果二級索引包含了所有的key part,那麼就可以得到排序好的聚集索引的鍵值或者ROWID，那麼簡單的union 去重就可以了，不需要額外的排序
             源碼介面quick_ror_union_select類別
merge sort_union :和上面的不同的是沒有包含二級索引所有的key part，那麼要先獲得排序的聚集索引鍵值或ROWID，才能對WIDROunion操作
                  原始碼介面quick_index_merge_select
參考手冊：9.2.1.4 Index Merge Optimization
作業總的來說只要mysql 無法確定主鍵是排序好的方式就需要額外的排序作業總的來說。


如果我們對merge sort演算法有一定了解，可以看到這樣的處理是必須的，
我們知道在進行歸併的時候所有的需要歸併的子集是需要排序好的，下面是一個簡單的歸併演算法的圖解:

mysql merge union merge sort_union 的不同

如果我們把1 2 5 9 和3 4 7 8看成primary key 那麼他們就是要排序好才能完成最後的歸併，
當然上層的排序操作可以歸併也可以用其他排序方式，只要排序好就可以，另外提一點，歸併
排序熟悉資料結構的朋友應該知道他也是外部磁碟排序的一種好方式。

這裡要理解我們需要對組合索引在INNODB B+樹頁塊的排列有一個了解：
比如：seq int,id1 int,id2 int  seq是主鍵，ID1，DI2是一個組合B+索引
那麼我們插入值

values(1,1,2)
values(2,1,3)
values(3,1,2)
登入後複製

顯然在組合索引的葉節點排列順序如下：

1       2       3
id1:1  id1:1  id1:1
id2:2  id2:2  id2:3
seq:1  seq:3  seq:2
登入後複製

也就是先按照id1進行排序然後按照id2排序最後按照主鍵seq排序.
那麼可以看到最後主鍵的順序為1 3 2並不是有序的,很明顯這樣的
結果集不能作為歸併的結果集，那麼我們就需要進行排序，這也是為什麼
sort_union sort的來源。

那麼下面來示範2種執行計畫的不同
腳本：

create table testmer
(seq int,id1 int,id2 int,id3 int,id4 int,primary key(seq),key(id1,id2),key(id3,id4));
insert into testmer values(1,1,2,4,4);
insert into testmer values(2,1,3,4,5);
insert into testmer values(3,1,2,4,4);
insert into testmer values(4,2,4,5,6);
insert into testmer values(5,2,6,5,8);
insert into testmer values(6,2,10,5,3);
insert into testmer values(7,4,5,8,10);
insert into testmer values(8,0,1,3,4);
登入後複製
mysql> select * from testmer;
+-----+------+------+------+------+
| seq | id1  | id2  | id3  | id4  |
+-----+------+------+------+------+
|   1 |    1 |    2 |    4 |    4 |
|   2 |    1 |    3 |    4 |    5 |
|   3 |    1 |    2 |    4 |    4 |
|   4 |    2 |    4 |    5 |    6 |
|   5 |    2 |    6 |    5 |    8 |
|   6 |    2 |   10 |    5 |    3 |
|   7 |    4 |    5 |    8 |   10 |
|   8 |    0 |    1 |    3 |    4 |
+-----+------+------+------+------+
登入後複製

很明顯這裡只看key(id1,id2) 就需要排序了，因為排列如下：
1      2 id1:1
id2:2  id2:2   id2:3
seq:1  seq:3  seq:2

如果我們把二級索引KEY_PART帶全

Using sort_union：
mysql> explain  select * from testmer force index(id1,id3) where id1=1 or id3=4;
+----+-------------+---------+------------+-------------+---------------+---------+---------+------+------+----------+----------------------------------------+
| id | select_type | table   | partitions | type        | possible_keys | key     | key_len | ref  | rows | filtered | Extra                                  |
+----+-------------+---------+------------+-------------+---------------+---------+---------+------+------+----------+----------------------------------------+
|  1 | SIMPLE      | testmer | NULL       | index_merge | id1,id3       | id1,id3 | 5,5     | NULL |    6 |   100.00 | Using sort_union(id1,id3); Using where |
+----+-------------+---------+------------+-------------+---------------+---------+---------+------+------+----------+----------------------------------------+
1 row in set, 1 warning (5.07 sec)
登入後複製

如果我們把二級索引KEY_PART帶全


mysql> explain  select * from testmer force index(id1,id3) where id1=1 and id2=2 or id3=4 and id4=1;
+----+-------------+---------+------------+-------------+---------------+---------+---------+------+------+----------+-----------------------------------+
| id | select_type | table   | partitions | type        | possible_keys | key     | key_len | ref  | rows | filtered | Extra                             |
+----+-------------+---------+------------+-------------+---------------+---------+---------+------+------+----------+-----------------------------------+
|  1 | SIMPLE      | testmer | NULL       | index_merge | id1,id3       | id1,id3 | 10,10   | NULL |    2 |   100.00 | Using union(id1,id3); Using where |
+----+-------------+---------+------------+-------------+---------------+---------+---------+------+------+----------+-----------------------------------+
登入後複製

這裡當然不需要排序我們。 =2(id3=4 and id4=1 也是一樣)

排列如下：


1         2      
id1:1   id1:1   
id2:2   id2:2 
seq:1  seq:3
登入後複製

也就說如果KEY_PART包含完整那麼主鍵自然排序好的結果，

其實我是在DEBUG環境下跑的，斷點打在了Unique::unique_add


(gdb) info b
Num     Type           Disp Enb Address            What
1       breakpoint     keep y   0x0000000000ebd333 in main(int, char**) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/main.cc:25
        breakpoint already hit 1 time
6       breakpoint     keep y   0x000000000145de13 in Unique::unique_add(void*) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/uniques.h:52
        breakpoint already hit 2 times
登入後複製

在執行select * from testmer force index(id1,id3) where id1=1 and id2=1 or id3=4 and id4=1;
沒有觸發Unique::unique_add，也::unique_add，也::unique_add就是沒有進行排序操作。

最後說明下源碼的merge_sort 排序的介面
QUICK_INDEX_MERGE_SELECT::read_keys_and_merge()
呼叫
Unique::unique_add
(使用balanced binary treetrees,平衡二叉樹非紅豆參考net/7728585/viewspace-2127419/
)

下面是源碼read_keys_and_merge()的註解：

/*
  Perform key scans for all used indexes (except CPK), get rowids and merge 
  them into an ordered non-recurrent sequence of rowids.
  
  The merge/duplicate removal is performed using Unique class. We put all
  rowids into Unique, get the sorted sequence and destroy the Unique.
  
  If table has a clustered primary key that covers all rows (TRUE for bdb
  and innodb currently) and one of the index_merge scans is a scan on PK,
  then rows that will be retrieved by PK scan are not put into Unique and 
  primary key scan is not performed here, it is performed later separately.
  RETURN
    0     OK
    other error
*/
登入後複製

下面是我gdb時候的堆疊情況：

(gdb) bt
#0  tree_insert (tree=0x7fffd801c768, key=0x7fffd801ada0, key_size=0, custom_arg=0x7fffd80103d0) at /root/mysql5.7.14/percona-server-5.7.14-7/mysys/tree.c:207
#1  0x000000000145df19 in Unique::unique_add (this=0x7fffd801c260, ptr=0x7fffd801ada0) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/uniques.h:56
#2  0x000000000178e6a8 in QUICK_INDEX_MERGE_SELECT::read_keys_and_merge (this=0x7fffd89083f0) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/opt_range.cc:10700
#3  0x0000000001778c73 in QUICK_INDEX_MERGE_SELECT::reset (this=0x7fffd89083f0) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/opt_range.cc:1601
#4  0x000000000155e529 in join_init_read_record (tab=0x7fffd8906e20) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/sql_executor.cc:2471
#5  0x000000000155b6a1 in sub_select (join=0x7fffd8905b08, qep_tab=0x7fffd8906e20, end_of_records=false)
    at /root/mysql5.7.14/percona-server-5.7.14-7/sql/sql_executor.cc:1271
#6  0x000000000155b026 in do_select (join=0x7fffd8905b08) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/sql_executor.cc:944
#7  0x0000000001558efc in JOIN::exec (this=0x7fffd8905b08) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/sql_executor.cc:199
#8  0x00000000015f91c6 in handle_query (thd=0x7fffd8000df0, lex=0x7fffd80033d0, result=0x7fffd8007a60, added_options=0, removed_options=0)
    at /root/mysql5.7.14/percona-server-5.7.14-7/sql/sql_select.cc:184
#9  0x00000000015ac025 in execute_sqlcom_select (thd=0x7fffd8000df0, all_tables=0x7fffd8006e98) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/sql_parse.cc:5391
#10 0x00000000015a4640 in mysql_execute_command (thd=0x7fffd8000df0, first_level=true) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/sql_parse.cc:2889
#11 0x00000000015acff6 in mysql_parse (thd=0x7fffd8000df0, parser_state=0x7ffff0fd6600) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/sql_parse.cc:5836
#12 0x00000000015a0eb5 in dispatch_command (thd=0x7fffd8000df0, com_data=0x7ffff0fd6d70, command=COM_QUERY)
    at /root/mysql5.7.14/percona-server-5.7.14-7/sql/sql_parse.cc:1447
#13 0x000000000159fce6 in do_command (thd=0x7fffd8000df0) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/sql_parse.cc:1010
#14 0x00000000016e1c08 in handle_connection (arg=0x3c1c880) at /root/mysql5.7.14/percona-server-5.7.14-7/sql/conn_handler/connection_handler_per_thread.cc:312
#15 0x0000000001d71ed0 in pfs_spawn_thread (arg=0x3bec1b0) at /root/mysql5.7.14/percona-server-5.7.14-7/storage/perfschema/pfs.cc:2188
#16 0x0000003ca62079d1 in start_thread () from /lib64/libpthread.so.0
#17 0x0000003ca5ee8b6d in clone () from /lib64/libc.so.6
登入後複製

下面是我gdb時候的堆疊情況：

merge sort_union：


T@3: | | | | | | | | | | >QUICK_INDEX_MERGE_SELECT::QUICK_INDEX_MERGE_SELECT
T@3: | | | | | | | | | | <QUICK_INDEX_MERGE_SELECT::QUICK_INDEX_MERGE_SELECT 1589
T@3: | | | | | | | | | | >QUICK_INDEX_MERGE_SELECT::init
T@3: | | | | | | | | | | <QUICK_INDEX_MERGE_SELECT::init 1595
T@3: | | | | | | | | >QUICK_INDEX_MERGE_SELECT::reset
T@3: | | | | | | | | | >QUICK_INDEX_MERGE_SELECT::read_keys_and_merge
T@3: | | | | | | | | | <QUICK_INDEX_MERGE_SELECT::read_keys_and_merge 10716
T@3: | | | | | | | | <QUICK_INDEX_MERGE_SELECT::reset 1602
T@3: | | | | | | | | >QUICK_INDEX_MERGE_SELECT::get_next
T@3: | | | | | | | | <QUICK_INDEX_MERGE_SELECT::get_next 10753
T@3: | | | | | | | | >QUICK_INDEX_MERGE_SELECT::get_next
T@3: | | | | | | | | <QUICK_INDEX_MERGE_SELECT::get_next 10753
T@3: | | | | | | | | >QUICK_INDEX_MERGE_SELECT::get_next
T@3: | | | | | | | | <QUICK_INDEX_MERGE_SELECT::get_next 10753
T@3: | | | | | | | | >QUICK_INDEX_MERGE_SELECT::get_next
T@3: | | | | | | | | <QUICK_INDEX_MERGE_SELECT::get_next 10753
T@3: | | | | | | | >QUICK_INDEX_MERGE_SELECT::~QUICK_INDEX_MERGE_SELECT
T@3: | | | | | | | <QUICK_INDEX_MERGE_SELECT::~QUICK_INDEX_MERGE_SELECT 1635
登入後複製

merge union：


T@3: | | | | | | | | | | >QUICK_ROR_UNION_SELECT::init
T@3: | | | | | | | | | | <QUICK_ROR_UNION_SELECT::init 1942
T@3: | | | | | | | | >QUICK_ROR_UNION_SELECT::reset
T@3: | | | | | | | | <QUICK_ROR_UNION_SELECT::reset 2004
T@3: | | | | | | | | >QUICK_ROR_UNION_SELECT::get_next
T@3: | | | | | | | | <QUICK_ROR_UNION_SELECT::get_next 10948
T@3: | | | | | | | | >QUICK_ROR_UNION_SELECT::get_next
T@3: | | | | | | | | <QUICK_ROR_UNION_SELECT::get_next 10948
T@3: | | | | | | | | >QUICK_ROR_UNION_SELECT::get_next
T@3: | | | | | | | | <QUICK_ROR_UNION_SELECT::get_next 10913
T@3: | | | | | | | >QUICK_ROR_UNION_SELECT::~QUICK_ROR_UNION_SELECT
T@3: | | | | | | | <QUICK_ROR_UNION_SELECT::~QUICK_ROR_UNION_SELECT 2021
登入後複製

可以看到調用路徑，查看源碼調用情況我只是想證明確實進行了排序，然後看看使用的什麼方式排序.本文只代表個人觀點，如果有誤起提示。謝謝！


 以上是mysql merge union merge sort_union 的不同 的內容，更多相關內容請關注PHP中文網（www.php.cn）！


🎜🎜
相關標籤：
mysql，sort_union，merge
來源：php.cn
上一篇：讓mysqldump變成並發導出導入的魔法 下一篇：Mac下brew方式安裝mysql
本網站聲明
本文內容由網友自願投稿，版權歸原作者所有。本站不承擔相應的法律責任。如發現涉嫌抄襲或侵權的內容，請聯絡admin@php.cn
作者最新文章
最新問題
function_exists()無法判定自訂函數 function test()    {        return true;    }    if (function_exists('TEST')) {        ech...
來自於 2024-04-29 11:01:01
0
2
1467
google 瀏覽器 手機版顯示的怎麼實現 老師您好，google 瀏覽器怎麼變成手機版樣式的？
來自於 2024-04-23 00:22:19
0
10
1639
子窗口操作父窗口，輸出沒反應 前兩句可執行，最後一句沒辦法應
來自於 2024-04-19 15:37:47
0
1
1397
父視窗沒有輸出 document.onclick = function(){ window.opener.document.write('我是子視窗的輸出');                  ...
來自於 2024-04-18 23:52:34
0
1
1320
關於CSS心智圖的課件在哪？ 課件
來自於 2024-04-16 10:10:18
0
0
1369
相關專題
更多>
熱門推薦
熱門教學
更多>
相關教學
熱門推薦
最新課程
最新下載
更多>
網站特效
網站源碼
網站素材
前端模板
關於我們 免責聲明 Sitemap
PHP中文網：公益線上PHP培訓，幫助PHP學習者快速成長！