zoukankan      html  css  js  c++  java
  • Mysql之force index

    Mysql的查询优化一般场景下好用,但在一些特殊场景下还是需要直接指定查询索引,这里复盘一个慢查询问题的修复。

    表结构:

    show create table wK_qy_messageG
    *************************** 1. row ***************************
    Table: wK_qy_message
    Create Table: CREATE TABLE `wK_qy_message` (
    `id` bigint(20) NOT NULL AUTO_INCREMENT,
    `wxid` varchar(200) CHARACTER SET utf8 NOT NULL COMMENT 'x号',
    `qyid` varchar(100) NOT NULL COMMENT '企业号 id',
    `chatServerId` varchar(100) NOT NULL COMMENT '联系人服务端 id',
    `msgSvrId` bigint(20) NOT NULL COMMENT '消息服务端 id',
    `msgId` bigint(20) NOT NULL COMMENT '消息客户端 id',
    PRIMARY KEY (`id`),
    UNIQUE KEY `wk_qy_message_wxid_qyid_chatServerId_msgSvrId_uindex`
    (`wxid`,`qyid`,`chatServerId`,`msgSvrId`),
    KEY `idx_wxid` (`wxid`)
    ) ENGINE=InnoDB AUTO_INCREMENT=2 DEFAULT CHARSET=utf8mb4
    1 row in set (0.00 sec)

    表状态:

    mysql> show table status like 'wk_qy_message'G
    *************************** 1. row ***************************
    Name: wk_qy_message
    Engine: InnoDB
    Version: 10
    Row_format: Compact
    Rows: 96841188
    Avg_row_length: 1007Data_length: 97555316736
    Max_data_length: 0
    Index_length: 36063674368
    Data_free: 4194304
    Auto_increment: 651285142
    Create_time: 2020-10-12 17:52:01
    Update_time: NULL
    Check_time: NULL
    Collation: utf8mb4_general_ci
    Checksum: NULL
    Create_options:
    Comment:
    Block_format: Original
    1 row in set (0.00 sec)

    不使用强制索引查询:

    desc select `msgId`, `msgSvrId`, `createTime`, `insert_time` from wk_qy_message
    where (`wxid` = 'wxid_1') order by `id` desc limit 1G
    *************************** 1. row ***************************
    id: 1
    select_type: SIMPLE
    table: wk_qy_message
    type: index
    possible_keys: wk_qy_message_wxid_qyid_chatServerId_msgSvrId_uindex,idx_wxid
    key: PRIMARY
    key_len: 8
    ref: NULL
    rows: 8782
    Extra: Using where
    1 row in set (0.00 sec)
    mysql> desc select `msgId`, `msgSvrId`, `createTime`, `insert_time` from
    wk_qy_message where (`wxid` = 'wxid_2') order by `id` desc limit 1
    G
    *************************** 1. row ***************************
    id: 1
    select_type: SIMPLE
    table: wk_qy_message
    type: ref
    possible_keys: wk_qy_message_wxid_qyid_chatServerId_msgSvrId_uindex,idx_wxid
    key: idx_wxid
    key_len: 602
    ref: const
    rows: 682248
    Extra: Using where
    1 row in set (0.00 sec)

    强制使用索引 wk_qy_message_wxid_qyid_chatServerId_msgSvrId_uindex:

    DESC select `msgId`, `msgSvrId`, `createTime`, `insert_time` from wk_qy_message
    force index(wk_qy_message_wxid_qyid_chatServerId_msgSvrId_uindex) where (`wxid`
    = 'wxid_1') order by `id` desc limit 1G
    *************************** 1. row ***************************
    id: 1
    select_type: SIMPLE
    table: wk_qy_message
    type: ref
    possible_keys: wk_qy_message_wxid_qyid_chatServerId_msgSvrId_uindexkey: wk_qy_message_wxid_qyid_chatServerId_msgSvrId_uindex
    key_len: 602
    ref: const
    rows: 11026
    Extra: Using index condition; Using where; Using filesort
    1 row in set (0.00 sec)
    desc select `msgId`, `msgSvrId`, `createTime`, `insert_time` from wk_qy_message
    force index(wk_qy_message_wxid_qyid_chatServerId_msgSvrId_uindex) where (`wxid`
    = 'wxid_2') order by `id` desc limit 1G
    *************************** 1. row ***************************
    id: 1
    select_type: SIMPLE
    table: wk_qy_message
    type: ref
    possible_keys: wk_qy_message_wxid_qyid_chatServerId_msgSvrId_uindex
    key: wk_qy_message_wxid_qyid_chatServerId_msgSvrId_uindex
    key_len: 602
    ref: const
    rows: 764316
    Extra: Using index condition; Using where; Using filesort
    1 row in set (0.00 sec)

    这样看无论 where wxid 取值怎样,使用索引
    wk_qy_message_wxid_qyid_chatServerId_msgSvrId_uindex 检索数据,Ertxa 都是 Using
    index condition; Using where; Using filesort

    接下来换一个索引:

    强制使用索引 wxid:

    DESC select `msgId`, `msgSvrId`, `createTime`, `insert_time` from wk_qy_message
    force index(idx_wxid) where (`wxid` = 'wxid_1') order by `id` desc
    limit 1G
    *************************** 1. row ***************************
    id: 1
    select_type: SIMPLE
    table: wk_qy_message
    type: ref
    possible_keys: idx_wxid
    key: idx_wxid
    key_len: 602
    ref: const
    rows: 11096
    Extra: Using where
    1 row in set (0.00 sec)
    DESC select `msgId`, `msgSvrId`, `createTime`, `insert_time` from wk_qy_message
    force index(idx_wxid) where (`wxid` = 'wxid_2') order by `id` desc
    limit 1G
    *************************** 1. row ***************************
    id: 1
    select_type: SIMPLE
    table: wk_qy_message
    type: ref
    possible_keys: idx_wxid
    key: idx_wxid
    key_len: 602
    ref: const
    rows: 682248
    Extra: Using where
    1 row in set (0.00 sec)

    这样看来

    无论 where wxid 取值怎样,使用索引 wxid 检索数据,Ertxa 都是 Using where;
    对比使用索引 wk_qy_message_wxid_qyid_chatServerId_msgSvrId_uindex 执行时间为 1s 到 10s
    不等
    ,而使用索引 wxid 后,该查询执行的时间为 10ms 到 30ms 不等;
    如果不使用 force index,让查询自己选择索引,则如果 wxid 的数据量较小时,会选择主键索引检索数据,
    此时查询时间非常大(1/8-21 点出现问题的时候就是这种情况)

    如果不提示使用那个索引呢:

    DESC select `msgId`, `msgSvrId`, `createTime`, `insert_time` from wk_qy_message
    force index(idx_wxid) where (`wxid` = 'wxid_1') order by `id` desc
    limit 1G
    *************************** 1. row ***************************
    id: 1
    select_type: SIMPLE
    table: wk_qy_message
    type: ref
    possible_keys: idx_wxid
    key: idx_wxid
    key_len: 602
    ref: const
    rows: 11096
    Extra: Using where
    1 row in set (0.00 sec)
    DESC select `msgId`, `msgSvrId`, `createTime`, `insert_time` from wk_qy_message
    where (`wxid` = 'wxid_2') order by `id` desc limit 1G
    *************************** 1. row ***************************
    id: 1
    select_type: SIMPLE
    table: wk_qy_message
    type: index
    possible_keys: wk_qy_message_wxid_qyid_chatServerId_msgSvrId_uindex,idx_wxid
    key: PRIMARY
    key_len: 8
    ref: NULL
    rows: 9084
    Extra: Using where
    1 row in set (0.00 sec)
    select count(*) from wk_qy_message where (`wxid` = 'wxid_1') order
    by `id` desc limit 1G
    *************************** 1. row ***************************
    count(*): 442888
    1 row in set (0.56 sec)
    mysql> select count(*) from wk_qy_message where (`wxid` = 'wxid_2')
    order by `id` desc limitt 1G
    *************************** 1. row ***************************
    count(*): 6077
    1 row in set (0.01 sec)

    结论

    固定 idx_wxid 索引,不要让数据库自己选择索引,遇到很大的索引列的数目很小时优化器会选择主键,此时
    导致执行时间巨大
    结论:选择 idx_wxid 索引,减少查询时间
    结论无论 where 条件中对应的 wxid 取值怎样,使用索引 wxid 检索数据,Ertxa 都是 Using where;
    对比使用索引 wk_qy_message_wxid_qyid_chatServerId_msgSvrId_uindex 执行时间为 1s 到 10s
    不等,而使用索引 wxid 后,该查询执行的时间为 10ms 到 30ms 不等;
    如果不使用 force index,让查询自己选择索引,遇到 wxid 的数据量很大时,很选择 wx_id 索引检索数据,
    此时很快 0.2s 返回查询结果;如果 wxid 的数据量较小时,会选择主键索引检索数据,此时查询时间非常大
    (1/8-21 点出现问题的时候就是这种情况)
    订正的语法如下

    select `msgId`, `msgSvrId`, `createTime`, `insert_time` from wk_qy_message force
    index(idx_wxid) where (`wxid` = 'wxid_1') order by `id` desc limit
    1;
  • 相关阅读:
    Windows下Yarn安装与使用
    论文阅读Graph Convolutional Matrix Completion
    如何快速查询中科院JCR分区和汤森路透JCR分区
    Implement GAN from scratch
    PyCharm将main.py解析成text文件的解决方法
    理解PyTorch的自动微分机制
    mcast_set_if函数
    mcast_get_if函数
    mcast_unblock_source函数
    mcast_block_source函数
  • 原文地址:https://www.cnblogs.com/timseng/p/14657433.html
Copyright © 2011-2022 走看看