500000条记录后MySQL变得非常慢

Posted 2023-02-26

【中文标题】500000条记录后MySQL变得非常慢

【英文标题】：After 500000 records MySQL become extremely slow

【发布时间】：2015-05-27 11:41:58

【问题描述】：

我正在使用 phpMyAdmin 4.2.7.1。 mysql 5.6.16。 MS Windows Server 2012 R2 标准版，4GB RAM，Intel Xeon E5-2620 @ 2.00GHz。几天前，我在 MySQL 查询中遇到了这个问题，该问题以前工作得很快。在过去，平均查询会比 5 分钟更快地返回结果，现在即使几个小时后它也无法返回结果。这是我创建的视图：

CREATE ALGORITHM=UNDEFINED DEFINER=root@localhost SQL SECURITY DEFINER 
VIEW okp_view AS 
     select q.MessageId AS MessageId,
q.SenderTimeStamp AS SenderTimeStamp,
r.GsbId AS GsbId,
r.ReceiverTimeStamp AS ReceiverTimeStamp,
r.FinalTimeStamp AS FinalTimeStamp,
k.ErrorCode AS ErrorCodeRes,
t.ErrorType AS ErrorTypeRes,
g.ID AS ID,
g.OIB AS OIB,
g.MssgText AS MssgText,
s.ErrorCode AS ErrorCodeResMssg,
m.ErrorType AS ErrorTypeMssg 
     from ((((((soap_req_env q 
join soap_res_env r) 
join soap_res_err k) 
join res_err_type t) 
join soap_message g) 
join soap_mssg_err s) 
join mssg_err_type m) 
     where ((q.MessageId = g.MessageId) 
and (g.ID = s.ID) 
and (s.ErrorCode = m.ErrorCode) 
and (q.MessageId = r.MessageId) 
and (r.GsbId = k.GsbId) 
and (k.ErrorCode = t.ErrorCode)) 
order by q.SenderTimeStamp desc;

视图包含超过 500000 条记录。 这些是 MySQL 表上的索引：

TABLE_NAME,INDEX_NAME
mssg_err_type,ErrorCode
registar_e_poruka_za_okp,PRIMARY
registar_e_poruka_za_okp,fk_Registar_e_poruka_za_OKP_Sifarnik_posiljatelja_e_poruka1_idx
registar_e_poruka_za_okp,fk_Registar_e_poruka_za_OKP_Sifarnik_zivotnih_situacija1_idx
registar_e_poruka_za_okp,fk_Registar_e_poruka_za_OKP_Sifarnik_tema1_idx
registar_e_poruka_za_okp,fk_Registar_e_poruka_za_OKP_Sifarnik_razine_pouzdanosti_vje_idx
registar_e_poruka_za_okp,fk_Registar_e_poruka_za_OKP_Sifarnik_tipa_privitka1_idx
registar_e_poruka_za_okp,fk_Registar_e_poruka_za_OKP_Sifarnik_frekvencije_slanja_por_idx
registar_e_poruka_za_okp,fk_Registar_e_poruka_za_OKP_Sifarnik_statusa_e_poruke1_idx
res_err_type,ErrorCode
soap_message,PRIMARY
soap_message,MessageId
soap_mssg_err,ID
soap_mssg_err,ErrorCode
soap_req_env,PRIMARY
soap_res_env,PRIMARY
soap_res_env,MessageId
soap_res_err,GsbId
soap_res_err,ErrorCode

现在 MySQL 为我提供了这个查询的数据：

SELECT * FROM okp_view WHERE SenderTimeStamp>="2015-05-25" 
Showing rows 0 - 24 (2132 total, Query took 13.9374 seconds.) 

如果我尝试使用以下方法检索更大的子集：

SELECT * FROM okp_view WHERE SenderTimeStamp>="2015-05-24"    

但这需要很长时间。

如何改进数据库架构以优化数据库并加快数据检索。

编辑： 如果我使用不带视图的查询需要很长时间：

select * from soap_req_env q, soap_res_env r, soap_res_err k, res_err_type t, soap_message g, soap_mssg_err s, mssg_err_type m
where q.messageid=g.messageid
and g.id=s.id
and s.errorcode=m.errorcode
and q.messageid=r.messageid
and r.gsbid=k.gsbid
and k.errorcode=t.errorcode
and q.sendertimestamp>="2015-05-15"
ORDER BY `q`.`SenderTimeStamp` DESC

SHOW VARIABLES LIKE '%buffer%'; 的结果是

Variable_name   Value
bulk_insert_buffer_size     8388608
innodb_buffer_pool_dump_at_shutdown     OFF
innodb_buffer_pool_dump_now     OFF
innodb_buffer_pool_filename     ib_buffer_pool
innodb_buffer_pool_instances    8
innodb_buffer_pool_load_abort   OFF
innodb_buffer_pool_load_at_startup  OFF
innodb_buffer_pool_load_now     OFF
innodb_buffer_pool_size     16777216
innodb_change_buffer_max_size   25
innodb_change_buffering     all
innodb_log_buffer_size  8388608
innodb_sort_buffer_size     1048576
join_buffer_size    262144
key_buffer_size     16777216
myisam_sort_buffer_size     8388608
net_buffer_length   8192
preload_buffer_size     32768
read_buffer_size    262144
read_rnd_buffer_size    524288
sort_buffer_size    524288
sql_buffer_result   OFF

我的表结构是：

CREATE TABLE `soap_req_env` (
 `MessageId` char(36) COLLATE utf8_croatian_ci NOT NULL,
 `SenderTimeStamp` timestamp NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
 PRIMARY KEY (`MessageId`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8 COLLATE=utf8_croatian_ci

    CREATE TABLE `soap_res_env` (
 `MessageId` char(36) COLLATE utf8_croatian_ci NOT NULL,
 `GsbId` char(36) COLLATE utf8_croatian_ci NOT NULL DEFAULT '',
 `ReceiverTimeStamp` timestamp NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
 `FinalTimeStamp` timestamp NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
 PRIMARY KEY (`GsbId`),
 KEY `MessageId` (`MessageId`),
 CONSTRAINT `soap_res_env_ibfk_1` FOREIGN KEY (`MessageId`) REFERENCES `soap_req_env` (`MessageId`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8 COLLATE=utf8_croatian_ci

CREATE TABLE `soap_res_err` (
 `GsbId` char(36) COLLATE utf8_croatian_ci NOT NULL,
 `ErrorCode` char(4) COLLATE utf8_croatian_ci NOT NULL,
 UNIQUE KEY `GsbId` (`GsbId`,`ErrorCode`),
 KEY `ErrorCode` (`ErrorCode`),
 CONSTRAINT `soap_res_err_ibfk_2` FOREIGN KEY (`ErrorCode`) REFERENCES `res_err_type` (`ErrorCode`),
 CONSTRAINT `soap_res_err_ibfk_3` FOREIGN KEY (`GsbId`) REFERENCES `soap_res_env` (`GsbId`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8 COLLATE=utf8_croatian_ci

 CREATE TABLE `res_err_type` (
 `ErrorCode` char(4) COLLATE utf8_croatian_ci NOT NULL,
 `ErrorType` text COLLATE utf8_croatian_ci NOT NULL,
 UNIQUE KEY `ErrorCode` (`ErrorCode`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8 COLLATE=utf8_croatian_ci

CREATE TABLE `soap_message` (
 `ID` bigint(20) unsigned NOT NULL AUTO_INCREMENT,
 `MessageId` char(36) COLLATE utf8_croatian_ci NOT NULL,
 `OIB` char(11) COLLATE utf8_croatian_ci NOT NULL,
 `MssgText` text CHARACTER SET utf8 COLLATE utf8_bin NOT NULL,
 PRIMARY KEY (`ID`),
 UNIQUE KEY `MessageId` (`MessageId`,`OIB`),
 CONSTRAINT `soap_message_ibfk_1` FOREIGN KEY (`MessageId`) REFERENCES `soap_req_env` (`MessageId`)
) ENGINE=InnoDB AUTO_INCREMENT=571197 DEFAULT CHARSET=utf8 COLLATE=utf8_croatian_ci

CREATE TABLE `soap_mssg_err` (
 `ID` bigint(20) unsigned NOT NULL AUTO_INCREMENT,
 `ErrorCode` char(4) COLLATE utf8_croatian_ci NOT NULL,
 KEY `ID` (`ID`),
 KEY `ErrorCode` (`ErrorCode`),
 CONSTRAINT `soap_mssg_err_ibfk_2` FOREIGN KEY (`ErrorCode`) REFERENCES `mssg_err_type` (`ErrorCode`),
 CONSTRAINT `soap_mssg_err_ibfk_3` FOREIGN KEY (`ID`) REFERENCES `soap_message` (`ID`)
) ENGINE=InnoDB AUTO_INCREMENT=571197 DEFAULT CHARSET=utf8 COLLATE=utf8_croatian_ci

CREATE TABLE `mssg_err_type` (
 `ErrorCode` char(4) COLLATE utf8_croatian_ci NOT NULL,
 `ErrorType` text COLLATE utf8_croatian_ci NOT NULL,
 UNIQUE KEY `ErrorCode` (`ErrorCode`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8 COLLATE=utf8_croatian_ci

【问题讨论】：

一件事很明显你正在创建视图，然后从视图中使用 select 语句，不幸的是在视图中你不能使用索引。但是，如果您根据原始选择查询进行选择以创建视图并在表上创建索引将比您现在得到的更快~14 sec

请使用以下语法：... JOIN t ON ... 而不是 JOIN ... WHERE ...

请提供SHOW CREATE TABLE soap_req_env;

@RickJames CREATE TABLE soap_req_env (MessageId char(36) COLLATE utf8_croatian_ci NOT NULL, SenderTimeStamp timestamp NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP, PRIMARY KEY (MessageId) InnoDB DEFAULT CHARSET=utf8 COLLATE=utf8_croatian_ci

请编辑问题为每个表格添加SHOW CREATE TABLE；您提供的索引列表不够详细。另外，让我们看看SHOW VARIABLES LIKE '%buffer%';

【参考方案1】：

UUID 的弊端

（我在这里很犹豫，不确定是否涉及 UUID。）

MessageId CHAR(36) COLLATE utf8... 主键

这闻起来像一个“UUID”的十六进制字符串；我假设它是。

问题 #1

CHAR(36) CHARACTER SET utf8 在任何出现的地方都消耗 108 个字节。它似乎是多个表中的一个键，并且可能出现在辅助键中（InnoDB 隐含地在每个辅助键中包含PRIMARY KEY。）对于 500K 记录，加起来可能是几兆字节，可能是千兆字节。

修复 #1：

CHAR(36) CHARACTER SET ascii 将只有 36 个字节。

修复 #2：

将其转换为二进制并将其存储在 BINARY(16) 中，这将只占用 16 个字节。我的UUID blog 提供了转换代码。

问题 #2

UUID 非常随机。一旦 UUID 索引大于 innodb_buffer_pool_size（或 key_buffer_size，如果是 MyISAM）中的缓存，越来越多的查找必须命中磁盘。例如，当索引是缓存的 20 倍时，95%（或更多）的查找需要磁盘命中。

【讨论】：

我有这个问题。当我尝试测试它时，我得到了一些意想不到的东西。首先我插入了这样的 UUID INSERT INTO test ( UUID ) VALUES (UuidToBin( "6ccd780c-baba-1026-9564-0040f4311e29" )); SELECT UuidFromBin(UUID) FROM test; 结果是 36636364373830632d626162612d313032362d393536342d303034306634333131653239

看起来在您显示的某处有一个HEX() 调用。 UUID 是否被声明为 BINARY(16)？搜索 UuidFromBin——我看到了几个等效的函数。试试他们而不是我的。