Due to the limitations of mysql, many sites adopt the mysql+memcached architecture.

Some other sites gave up mysql and adopted NoSQL, such as TokyoCabinet/Tyrant and so on.
It is undeniable that NoSQL is much faster than mysql when doing some simple queries (especially primary key queries). And the vast majority of queries on the website are simple queries like this.

But DeNA has been using mysql and memcached, and has created a record of 750,000 simple queries per second on a single ordinary server.

Maybe you won’t believe that a single mysql can achieve 750,000 qps, but this is a fact, let’s talk about it in detail below.

Is SQL really suitable for primary key queries?

DeNA company's applications often require primary key (PK) query. For example, userinfo is retrieved based on user id, and log content is retrieved based on diary id. Memcached and NoSQL are very suitable for this kind of thing.
If you test memcached, it is very likely that 400,000 get operations can be performed per second, even if memcached and client are located on different servers.
On a server with a 2.5GHz 8-core Nehalem 3-kiloM network card, libmemcached and memcached can perform 420,000 get operations per second.

What about mysql5? How many PK queries can be done per second?
innodb:

[matsunobu@host ~]$ mysqlslap --query="select user_name,..  from test.user where user_id=1" /
--number-of-queries=10000000 --concurrency=30 --host=xxx -uroot  -p

Copy after login

##In addition, you can also use tools such as sysbench or super-smack for benchmarking.

Open a new shell and take a look:

[matsunobu@host ~]$ mysqladmin extended-status -i 1 -r -uroot /
| grep -e "Com_select"
 
 
| Com_select                        | 107069     |
| Com_select                        | 108873     |
| Com_select                        | 108921     |
| Com_select                        | 109511     |
| Com_select                        | 108084     |
| Com_select                        | 108115     |

Copy after login

100,000 qps is about 1/4 of memcached. Why is it so slow?
The server has enough memory, and all these data should be in memory.
It is also a memory operation. Why is mysql so much slower than memcached?

vmstat data is as follows:

[matsunobu@host ~]$ vmstat 1
r  b  swpd   free   buff  cache      in     cs us sy id wa st
23  0     0 963004 224216 29937708 58242 163470 59 28 12  0  0
24  0     0 963312 224216 29937708 57725 164855 59 28 13  0  0
19  0     0 963232 224216 29937708 58127 164196 60 28 12  0  0
16  0     0 963260 224216 29937708 58021 165275 60 28 12  0  0
20  0     0 963308 224216 29937708 57865 165041 60 28 12  0  0

Copy after login

##%user and The CPU occupied by %system is quite high.

Let’s take a look at the statistics from oprofile:

ps: This tool is good, at the kernel level.

samples  %        app name                 symbol name
259130    4.5199  mysqld                   MYSQLparse(void*)
196841    3.4334  mysqld                   my_pthread_fastmutex_lock
106439    1.8566  libc-2.5.so              _int_malloc
94583     1.6498  bnx2                     /bnx2
84550     1.4748  ha_innodb_plugin.so.0.0.0 ut_delay
67945     1.1851  mysqld                   _ZL20make_join_statistics
P4JOINP10TABLE_LISTP4ItemP16st_dynamic_array
63435     1.1065  mysqld                   JOIN::optimize()
55825     0.9737  vmlinux                  wakeup_stack_begin
55054     0.9603  mysqld                   MYSQLlex(void*, void*)
50833     0.8867  libpthread-2.5.so        pthread_mutex_trylock
49602     0.8652  ha_innodb_plugin.so.0.0.0 row_search_for_mysql
47518     0.8288  libc-2.5.so              memcpy
46957     0.8190  vmlinux                  .text.elf_core_dump
46499     0.8111  libc-2.5.so              malloc

Copy after login

MYSQLparse is in version 5.x, and YYparse

in 4.x

MYSQLparse() and MYSQLlex() are called when mysql parses sql statements.

make_join_statistics() and JOIN::optimize() are called during the query optimization phase.

It is precisely because of the use of SQL statements that there are these additional burdens.

The following conclusions can be drawn from the output of oprofile:

The SQL layer seriously affects the performance of mysql queries.

Compared with memcached and SQL, mysql needs to do some extra work:

* Parsing SQL statements Parsing sql statements * Opening, locking tables Opening and locking tables
* Making SQL execution plans ???
* Unlocking, closing tables Unlocking and closing tables

Hua Rong's Note: Using the prepared statement API in mysqli can avoid parsing sql statements.

Mysql must also do a lot of concurrency control. For example, when sending/receiving network data packets, fcntl() will be called many, many times.

Global mutexes: LOCK_open LOCK_thread_count is also called frequently. So in the output of oprofile, the second one is my_pthread_fastmutex_lock(). And the CPU occupied by %system is quite high (28%).
In fact, the mysql development team and some peripheral development groups understand the impact of a large number of concurrency controls on performance.
They have solved certain problems in mysql 5.5. In the future mysql, %system will occupy less and less CPU.

But, how to deal with the 60% cpu occupied by %user?

Mutex contentions result in %system increase, not %user increase

Copy after login

Even if all concurrency issues are dealt with, it is estimated that it will be difficult to achieve 300,000 qps.

Perhaps you have heard that the performance of HANDLER statement is also good.

But the HANDLER statement requires query parsing and open/close table. Won't be of much help.

In the case of full memory operations, CPU efficiency is very important

If only a small part of the data can enter the memory, then the SQL statement The extra burden brought by the analysis is nothing.

Because disk IO operations will take longer.

In our mysql server, the memory is large, and almost all data can be put into the memory.

The SQL layer becomes an additional burden and takes up a lot of CPU resources.

在线上的应用中，我们要进行大量的PK查询。即使70-80%的查询都是在同一张表上进行的， mysql还是每次都要parse/open/lock/unlock/close，看起来就感觉效率低下。

We needed to execute lots of primary key lookups(i.e. SELECT x FROM t WHERE id=?) or limited range scans. Even though 70-80% of queries were simple PK lookups
 from the same table (difference was just values in WHERE), every time
 MySQL had to parse/open/lock/unlock/close, which seemed not efficient for us.

Copy after login

花荣注：难道说mysql中的table_open_cache不是用来减少table open的次数的么。。

NDBAPI

有没有办法在sql层进行优化呢？

http://dev.mysql.com/doc/ndbapi/en/index.html

如果你使用mysql cluster, NDBAPI会是最佳解决方案。

It’s recommended using NDBAPI for frequent access patterns,
and using SQL + MySQL + NDB for ad-hoc or infrequent query patterns.

Copy after login

这就是我们想要的：
1 faster access API.
2 sql语句仍然要可用,以处理一些特定的或者复杂的查询。

但是，把innodb转化成ndb可不是一件轻松的事情。

HandlerSocket Plugin

最好的办法是在mysql内部实现一个NoSQL的网络服务。daemon plugin。
它监听在某个端口，接受NoSQL 协议/API的数据包，使用Mysql internal storage engine API直接在innodb数据表上进行操作，并且返回相应的数据。
关于mysql internal storage engine API可以看这个文档：

//m.sbmmt.com/

这个概念首先被Cybozu Labs 的Kazuho Oku 提出，然后他写了一个MyCached UDF，用的是memcached的协议。

http://developer.cybozu.co.jp/kazuho/2009/08/mycached-memcac.html

随后，Akira Higuchi 写了另外一个plugin: HandlerSocket。

http://github.com/ahiguti/HandlerSocket-Plugin-for-MySQL

从图中可以看到，客户端既可以使用普通的mysql api来操作mysql(3306端口)，
也可以使用HandlerSocket API对数据库进行PK查询，以及INSERT/UPDATE/DELETE操作(9998与9999端口)。

在使用HandlerSocket操作的时候，省去了SQL parsing, Opening table, Making Query Plans, Closing table等步骤。

而且数据都是保存在原来的INNODB表中的。

在HandlerSocket操作innodb数据表的时候，显然也需要open/close table。
但它只打开一次，然后会reuse。
由于open/close table非常耗时，并且会带来严重的mutex竞争，所以这种改进，极大地提升了性能。

Of course HandlerSocket closes tables when traffics become small etc so that it won’t block administrative commands (DDL) forever.

memcached主要用来缓存数据集(database records)。
因为memcached的get操作比mysql的PK查询要快得多。
如果HandlerSocket获取数据的速度比memcached还要快，那么我们就没必要再使用memcached来缓存数据集了。做做其它缓存还是可以的。比如生成出来的HTML代码，还有一些统计数据等等。

花荣注：貌似我从来没有直接把mysql的数据集扔到memcached中。一直都是把中间结果保存到里面。难道走偏了？

使用 HandlerSocket

看下面这张user表：

CREATE TABLE user (
user_id INT UNSIGNED PRIMARY KEY,
user_name VARCHAR(50),
user_email VARCHAR(255),
created DATETIME
) ENGINE=InnoDB;

Copy after login

在mysql中取出数据的方法如下：

mysql> SELECT user_name, user_email, created FROM user WHERE user_id=101;
+---------------+-----------------------+---------------------+
| user_name     | user_email            | created             |
+---------------+-----------------------+---------------------+
| Yukari Takeba | yukari.takeba@dena.jp | 2010-02-03 11:22:33 |
+---------------+-----------------------+---------------------+

Copy after login

在HandlerSocket中怎样操作呢？

安装HandlerSocket

详细文档看这里：

http://github.com/ahiguti/HandlerSocket-Plugin-for-MySQL/blob/master/docs-en/installation.en.txt

大概安装步骤如下：
1 下载

http://github.com/ahiguti/HandlerSocket-Plugin-for-MySQL

2 编译 HandlerSocket客户端和服务器端程序：

./configure --with-mysql-source=... --with-mysql-bindir=... ; make; make install

Copy after login

3 安装插件

mysql> INSTALL PLUGIN &#39;HandlerSocket&#39; SONAME &#39;HandlerSocket.so&#39;;

Copy after login

安装完毕。不需要修改mysql的源代码。
mysql需要5.1或者以后版本。

编写HandlerSocket 客户端代码

目前HandlerSocket客户端只有C++ 和 Perl的库。还没有php和C的。

Copy after login

 
#!/usr/bin/perl
 
use strict;
use warnings;
use Net::HandlerSocket;
 
#1. establishing a connection
my $args = { host => &#39;ip_to_remote_host&#39;, port => 9998 };
my $hs = new Net::HandlerSocket($args);
 
#2. initializing an index so that we can use in main logics.
 # MySQL tables will be opened here (if not opened)
my $res = $hs->open_index(0, &#39;test&#39;, &#39;user&#39;, &#39;PRIMARY&#39;,
    &#39;user_name,user_email,created&#39;);
die $hs->get_error() if $res != 0;
 
#3. main logic
 #fetching rows by id
 #execute_single (index id, cond, cond value, max rows, offset)
$res = $hs->execute_single(0, &#39;=&#39;, [ &#39;101&#39; ], 1, 0);
die $hs->get_error() if $res->[0] != 0;
shift(@$res);
for (my $row = 0; $row < 1; ++$row) {
  my $user_name= $res->[$row + 0];
  my $user_email= $res->[$row + 1];
  my $created= $res->[$row + 2];
  print "$user_name/t$user_email/t$created/n";
}
 
#4. closing the connection
$hs->close();

Copy after login

这段脚本从user表中取出来了 user_name, user_email 和created字段。

[matsunobu@host ~]$ perl sample.pl
Yukari Takeba   yukari.takeba@dena.jp   2010-02-03 11:22:33

Copy after login

对于HandlerSocket，推荐使用persistent connection。以减少数据库连接的次数。

HandlerSocket协议 is a small-sized text based protocol。
与memcached类似，可以使用telnet来获取数据。

[matsunobu@host ~]$ telnet 192.168.1.2 9998
Trying 192.168.1.2...
Connected to xxx.dena.jp (192.168.1.2).Escape character is &#39;^]&#39;.
P       0       test    user    PRIMARY user_name,user_email,created
0       1
0       =       1       101
0       3       Yukari Takeba   yukari.takeba@dena.jp   2010-02-03 11:22:33

Copy after login

Benchmarking

测试环境：

CPU: Nehalem 8 cores, E5540 @ 2.53GHz
RAM: 32GB (all data fit in the buffer pool)
MySQL Version: 5.1.50 with InnoDB Plugin
memcached/libmemcached version: 1.4.5(memcached), 0.44(libmemcached)
Network: Broadcom NetXtreme II BCM5709 1000Base-T x 3

Copy after login

仍然使用上面的user表，数据大概100w行。
SQL语句为：

SELECT user_name, user_email, created FROM user WHERE  userid=?

Copy after login

memcached和HandlerSocket的客户端代码都用C/C++编写。
所有的客户端程序都位于另外一台机器上。通过TCP/IP与MYSQL/memcached服务器连接。

测试结果如下：

approx qps server CPU util
MySQL via SQL 105,000 %us 60% %sy 28%
memcached 420,000 %us 8% %sy 88%
HandlerSocket 750,000 %us 45% %sy 53%

Copy after login

HandlerScket比传统的mysql快了7.5倍,而且%us的cpu使用率为mysql的3/4。
说明SQL层还是相当耗时的。避开SQL层之后，性能得到了明显的提升。

HandlerSocket比memcached快了78%。而且%sy占用的cpu比memcached要少得多。
虽然memcached是一个优秀的产品，但是还有很大的提升空间。

花荣注：比如七夜就改写了memcached的部分代码。

再来看一下HandlerSocket测试的时候oprofile的输出：

samples  %        app name                 symbol name
984785    5.9118  bnx2                     /bnx2
847486    5.0876  ha_innodb_plugin.so.0.0.0 ut_delay
545303    3.2735  ha_innodb_plugin.so.0.0.0 btr_search_guess_on_hash
317570    1.9064  ha_innodb_plugin.so.0.0.0 row_search_for_mysql
298271    1.7906  vmlinux                  tcp_ack
291739    1.7513  libc-2.5.so              vfprintf
264704    1.5891  vmlinux                  .text.super_90_sync
248546    1.4921  vmlinux                  blk_recount_segments
244474    1.4676  libc-2.5.so              _int_malloc
226738    1.3611  ha_innodb_plugin.so.0.0.0 _ZL14build_template
P19row_prebuilt_structP3THDP8st_tablej
206057    1.2370  HandlerSocket.so         dena::hstcpsvr_worker::run_one_ep()
183330    1.1006  ha_innodb_plugin.so.0.0.0 mutex_spin_wait
175738    1.0550  HandlerSocket.so         dena::dbcontext::cmd_find_internal(dena::dbcallback_i&, dena::prep_stmt const&, ha_rkey_function, dena::cmd_exec_args const&)
169967    1.0203  ha_innodb_plugin.so.0.0.0 buf_page_get_known_nowait
165337    0.9925  libc-2.5.so              memcpy
149611    0.8981  ha_innodb_plugin.so.0.0.0 row_sel_store_mysql_rec
148967    0.8943  vmlinux                  generic_make_request

Copy after login

大部分的CPU用在了网络数据包的处理，取出数据，等。
排在首位的bnx2是网卡驱动程序。

由于HandlerSocket只是一个插件，最终还会调用innodb引擎的函数去取数据，
所以我们仍然可以使用mysql命令来获取到统计数据。

$ mysqladmin extended-status -uroot -i 1 -r -p| grep “InnoDB_rows_read”
…| Innodb_rows_read | 750192 |
| Innodb_rows_read | 751510 |
| Innodb_rows_read | 757558 |
| Innodb_rows_read | 747060 |
| Innodb_rows_read | 748474 |
| Innodb_rows_read | 759344 |
| Innodb_rows_read | 753081 |
| Innodb_rows_read | 754375 |

Copy after login

作者注：memcached和HandlerSocket的性能都会受到网络IO的限制。
上面的测试是在3块千M网卡的环境下进行的。如果换成单网卡，HandlerSocket大概每秒处理260,000次查询，memcached每秒处理220,000次查询。

HandlerSocket的优势和特性

支持很多类型的查询

PK肯定没问题
unique key也可以。
普通key也可以。
limit 语句也可以。
IN 也没问题。
INSERT/UPDATE/DELETE也没问题。
不过，一定要用到索引才行。
不使用索引的操作不被支持。 Operations that do not use any index are not supported。
详细说明看这里：//m.sbmmt.com/

高并发

HandlerSocket employs epoll() and worker-thread/thread-pooling architecture,the number of MySQL internal threads is limited 。
所以放心地使用persistent connection吧。不会导致mysql connection太高的。

高性能

Not only HandlerSocket eliminates SQL related function calls, but also it optimizes around network/concurrency issues
HandlerSocket不仅避开了SQL层，而且优化了网络层，并解决了并发的一些问题。

更小的网络数据包

与mysql协议相比，HandlerSocket协议的数据包更简单更小。
从总体上来看，网络传输的数据量会减少。

Running limited number of MySQL internal threads

不会导致mysql connection太高。

Grouping client requests

当HandlerSocket接受到请求的时候，每一个处理线程都会收集尽可能多的请求，
然后一次把这些请求执行完毕，并返回数据。
看起来没什么神奇的，但是却可以大大地提升性能。
代价是每个请求的响应时间略微变长。
*** Can reduce the number of fsync() calls
*** Can reduce replication delay

至于详细情况，作者会再写其它文章。很期待。

No duplicate cache

我们使用memcached的时候，数据会同时缓存到innodb的buffer pool与memcached中。
算是重复的cache吧。
HandlerSocket自身没有缓存，它完全听从InnoDB storage engine。

No data inconsistency

缓存的一大问题就是数据何时过期。
HandlerSocket没有缓存，也就不存在这种问题。

Crash-safe

InnoDB还是相当可靠的。
innodb-flush-log-at-trx-commit=1 加上这句就更保险了。
最多会丢失死机前1秒内的数据。

SQL can be used from mysql clients

我们仍然可以使用SQL语句进行复杂的查询。
确实方便很多啊。

All operational benefits from MySQL

HandlerSocket作为插件运行于mysql内部，所以mysql的操作，比如SQL, 热备份，主从，Nagios监视等等，都是支持的。
通过 show global status, show engine innodb status , show processlist 这些都可以看到HandlerSocket的状态。

No need to modify/rebuild MySQL

无需改动mysql的源代码或者重新编译mysql。
线上的环境可以直接使用HandlerSocket。

Independent from storage engines

理论上讲，HandlerSocket支持任何的存储引擎，比如myisam,memory?
不过目前只在mysql5.1 5.5 InnoDB的环境下进行过测试和应用。

缺点与注意事项

Need to learn HandlerSocket APIs

目前只有C++和PERL库可用。以后肯定会有php扩展出现。

No security

就像其它的NoSQL数据一样，HandlerSocket没有任何安全方面的功能。比如权限控制之类的。
HandlerSocket的工作线程使用system权限，所以你的应用可以存取所有的表。
当然可以使用防火墙来限制访问。

No benefit for HDD bound workloads

如果你的内存不够大，内存要与硬盘交换数据，那么HandlerSocket的优势就体现不出来了。
We use HandlerSocket on servers that almost all data fit in memory.

DeNA is using HandlerSocket in production

DeNA公司已经在生产环境上使用HandlerSocket了。
The results are great!
有了HandlerSocket可以省掉很大一批memcached服务器与mysql从服务器。
而且带宽占用也得到了缓解。

We’ve been very satisfied with the results。
Since HandlerSocket plugin is Open Source, feel free to try. We’d be appreciated if you give us any feedback.

Copy after login

以上就是[转]Mysql的HandlerSocket插件的内容，更多相关内容请关注PHP中文网（m.sbmmt.com）！