TokuDB存储引擎

TokuDB是Tokutek公司开发的基于ft-index（Fractal Tree Index）键值对的存储引擎。

它使用索引加快查询速度，具有高扩展性，并支持hot scheme modification，具有以下特点：

1. 插入性能快20~80倍；

2. 压缩数据减少存储空间；

3. 数据量可以扩展到几个TB；

4. 不会产生索引碎片；

5. 支持hot column addition，hot indexing，mvcc

适用场景：

1. 如果你要存储blob，不要使用TokuDB，因为它限制记录不能太大；

2. 如果你的记录数量过亿，使用TokuDB；

3. 如果你注重update的性能，不要使用TokuDB，它没有Innodb快；

4. 如果你要存储旧的记录，使用TokuDB；

5. 如果你想要缩小数据占用的存储空间，使用TokuDB；

下面看看TokuDB与InnoDB的对比情况，

注：上述总结和性能数据来源于淘宝梁智超的《RethinkDB-&-TokuDB调研测试报告.pptx》

关于TokuDB和InnoDB性能压测报告，可参考：

https://www.percona.com/blog/2016/02/01/innodb-and-tokudb-on-aws/

以下安装基于Percona 5.6.31-77.0

安装libjemalloc库

该库在EPEL中有提供

# wget http://dl.fedoraproject.org/pub/epel/epel-release-latest-6.noarch.rpm

# rpm -ivh epel-release-latest-6.noarch.rpm

# yum install jemalloc -y

通过yum安装，生成的库文件为/usr/lib64/libjemalloc.so.1

[root@localhost ~]# rpm -qa |grep jemalloc
jemalloc-3.6.-.el6.x86_64
[root@localhost ~]# rpm -ql jemalloc-3.6.-.el6.x86_64
/usr/bin/jemalloc.sh
/usr/lib64/libjemalloc.so.1
/usr/share/doc/jemalloc-3.6.
/usr/share/doc/jemalloc-3.6./COPYING
/usr/share/doc/jemalloc-3.6./README
/usr/share/doc/jemalloc-3.6./VERSION
/usr/share/doc/jemalloc-3.6./jemalloc.html

修改配置文件my.cnf

在[mysqld_safe]下设置malloc-lib变量

malloc-lib=/usr/lib64/libjemalloc.so.

重启mysqld_safe进程

启动过程中，会输出以下信息：

 :: mysqld_safe Adding '/usr/lib64/libjemalloc.so.1' to LD_PRELOAD for mysqld

如果不安装该库的话，则在加载TokuDB插件时会报如下错误：

mysql> INSTALL PLUGIN tokudb SONAME 'ha_tokudb.so';
2016-08-10 16:25:02 1861 [ERROR] TokuDB is not initialized because jemalloc is not loaded
2016-08-10 16:25:02 1861 [ERROR] Plugin 'TokuDB' init function returned error.
2016-08-10 16:25:02 1861 [ERROR] Plugin 'TokuDB' registration as a STORAGE ENGINE failed.
2016-08-10 16:25:02 1861 [Note] Shutting down plugin 'TokuDB'
ERROR 1123 (HY000): Can't initialize function 'tokudb'; Plugin initialization function failed.

禁用Transparent huge pages

查看当前内核是否已启用Transparent huge pages

# cat /sys/kernel/mm/transparent_hugepage/enabled
[always] madvise never

always代表已启用

如何禁用呢？

echo never > /sys/kernel/mm/transparent_hugepage/enabled
echo never > /sys/kernel/mm/transparent_hugepage/defrag

如果没有禁用的话，在后续安装插件的时候会报错：

mysql> INSTALL PLUGIN tokudb SONAME 'ha_tokudb.so';
ERROR 1123 (HY000): Can't initialize function 'tokudb'; Plugin initialization function failed.

并且，错误日志中打印如下信息：

-- ::  [ERROR] TokuDB: Huge pages are enabled, disable them before continuing

-- ::  [ERROR] ************************************************************
-- ::  [ERROR]
-- ::  [ERROR]                         @@@@@@@@@@@
-- ::  [ERROR]                       @@'         '@@
-- ::  [ERROR]                      @@    _     _  @@
-- ::  [ERROR]                      |    (.)   (.)  |
-- ::  [ERROR]                      |    (.)   (.)  |
-- ::  [ERROR]                      |             ` |
-- ::  [ERROR]                      |        >    ' |
-- ::  [ERROR]                      |     .----.    |
-- ::  [ERROR]                      ..   |.----.|  ..
-- ::  [ERROR]                       ..  '      ' ..
-- ::  [ERROR]                         .._______,.
-- ::  [ERROR]
-- ::  [ERROR] TokuDB will not run with transparent huge pages enabled.
-- ::  [ERROR] Please disable them to continue.
-- ::  [ERROR] (echo never > /sys/kernel/mm/transparent_hugepage/enabled)
-- ::  [ERROR]
-- ::  [ERROR] ************************************************************
-- ::  [ERROR] Plugin 'TokuDB' init function returned error.
-- ::  [ERROR] Plugin 'TokuDB' registration as a STORAGE ENGINE failed.
-- ::  [Note] Shutting down plugin 'TokuDB'

启用TokuDB

Percona是5.6.17-66.0将TokuDB插件引入到自己的二进制版本中，5.6.19-67.0起才能作为GA版本使用。5.6.22-72.0起开发了一个脚本ps_tokudb_admin来简化TokuDB的安装工作。

首先，来看看TokuDB如何手动安装

主要是加载插件，TokuDB插件名为ha_tokudb.so，默认放到plugin_dir下。

登录mysql客户端，执行如下命令

INSTALL PLUGIN tokudb SONAME 'ha_tokudb.so';
INSTALL PLUGIN tokudb_file_map SONAME 'ha_tokudb.so';
INSTALL PLUGIN tokudb_fractal_tree_info SONAME 'ha_tokudb.so';
INSTALL PLUGIN tokudb_fractal_tree_block_map SONAME 'ha_tokudb.so';
INSTALL PLUGIN tokudb_trx SONAME 'ha_tokudb.so';
INSTALL PLUGIN tokudb_locks SONAME 'ha_tokudb.so';
INSTALL PLUGIN tokudb_lock_waits SONAME 'ha_tokudb.so';
INSTALL PLUGIN tokudb_background_job_status SONAME 'ha_tokudb.so';

查看插件是否加载成功

mysql> show engines;

| TokuDB             | YES     | Percona TokuDB Storage Engine with Fractal Tree(tm) Technology             | YES          | YES  | YES        |

mysql> show plugins;

| TokuDB                        | ACTIVE   | STORAGE ENGINE     | ha_tokudb.so | GPL     |
| TokuDB_file_map               | ACTIVE   | INFORMATION SCHEMA | ha_tokudb.so | GPL     |
| TokuDB_fractal_tree_info      | ACTIVE   | INFORMATION SCHEMA | ha_tokudb.so | GPL     |
| TokuDB_fractal_tree_block_map | ACTIVE   | INFORMATION SCHEMA | ha_tokudb.so | GPL     |
| TokuDB_trx                    | ACTIVE   | INFORMATION SCHEMA | ha_tokudb.so | GPL     |
| TokuDB_locks                  | ACTIVE   | INFORMATION SCHEMA | ha_tokudb.so | GPL     |
| TokuDB_lock_waits             | ACTIVE   | INFORMATION SCHEMA | ha_tokudb.so | GPL     |
| TokuDB_background_job_status  | ACTIVE   | INFORMATION SCHEMA | ha_tokudb.so | GPL     |

创建一张表，测试一下

mysql> create table test.test(id int) engine=tokudb;
Query OK, 0 rows affected (0.23 sec)

至此，TokuDB手动安装完毕~

脚本安装

其实，Percona官方在5.6.22-72.0版本中提供了一个脚本，可用于自动安装TokuDB插件，涉及的操作包括禁用透明大页，加载插件。

该脚本在二进制包的bin目录下。

[root@localhost bin]# ./ps_tokudb_admin --help
This script is used for installing and uninstalling TokuDB plugin for Percona Server 5.6.
It can also be used to install or uninstall the Percona TokuBackup plugin (requires mysql server restart).
If transparent huge pages are enabled on the system it adds thp-setting=never option to my.cnf
to disable it on runtime.

Valid options are:
  --user=user_name, -u user_name     mysql admin username
  --password[=password], -p[password]     mysql admin password (on empty will prompt to enter)
  --socket=path, -S path         the socket file to use for connection
  --host=host_name, -h host_name     connect to given host
  --port=port_num, -P port_num         port number to use for connection
  --defaults-file=file          specify defaults file instead of guessing
  --enable, -e                 enable TokuDB plugin and disable transparent huge pages in my.cnf
  --enable-backup, -b             enable Percona TokuBackup and add preload-hotbackup option to my.cnf
                     (this option includes --enable option)
  --disable, d                 disable TokuDB plugin and remove thp-setting=never option in my.cnf
                     (this option includes --disable-backup option)
  --disable-backup, r             disable Percona TokuBackup and remove preload-hotbackup option in my.cnf
  --help                 show this help

For TokuDB requirements and manual steps for installation please visit this webpage:
http://www.percona.com/doc/percona-server/5.6/tokudb/tokudb_installation.html

该脚本是用shell写的，参数也比较简单，其中，--enable-backup还包括安装TokuDB在线热备插件。

脚本安装过程如下

 ./ps_tokudb_admin --enable --host=127.0.0.1 --port= --user=root -p
Enter password:
Continuing without password...

Checking SELinux status...
INFO: SELinux is disabled.

Checking if Percona Server is running with jemalloc enabled...
INFO: Percona Server is running with jemalloc enabled.

Checking transparent huge pages status on the system...
INFO: Transparent huge pages are currently disabled on the system.

Checking if thp-setting=never option is already set in config file...
INFO: Option thp-setting=never is set in the config file.

Checking TokuDB engine plugin status...
INFO: TokuDB engine plugin is installed.

总结

TokuDB最新版本是7.5.7，实际上现在是以插件的形式包含在Percona二进制发行版中。但该插件并不能直接在MySQL社区版中使用。

TokuDB基于MySQL 5.5.30源码，重新打包了个二进制版本，mysql-5.5.30-tokudb-7.1.0-linux-x86_64.tar.gz，基于TokuDB 7.1.0。

但也仅限如此，MySQL 5.6和5.7的社区版中并不能直接使用TokuDB插件。

以后若要使用较新版本的MySQL和TokuDB，只能使用Percona版本的。

下面是官档中的说明

Please note that Tokutek made some changes to the MySQL source that are required to either fix bugs or aid in performance, so if you are compiling from source you must use the Tokutek version of MySQL that is based on the MySQL 5.5. source.

参考

1. https://www.percona.com/blog/2013/07/02/tokumx-fractal-treer-indexes-what-are-they/

2. http://dbaplus.cn/news-21-418-1.html

3. https://www.izhangheng.com/highly-scalable-storage-engine-tokudb