InnoDBMemcachedPlugin源码实现调研

背景 MySQL 5.6版本，新增了一个NoSQL的接口，通过将memcached嵌入到MySQL系统之中，用户可以直接使用memcached接口直接操作MySQL中的InnoDB表，绕过MySQL Server层面的SQL解析，优化，甚至绕过InnoDB Handler层，直接操作InnoDB内部的方法，从而达到更优的

背景

MySQL 5.6版本，新增了一个NoSQL的接口，通过将memcached嵌入到MySQL系统之中，用户可以直接使用memcached接口直接操作MySQL中的InnoDB表，绕过MySQL Server层面的SQL解析，优化，甚至绕过InnoDB Handler层，直接操作InnoDB内部的方法，从而达到更优的响应时间与效率。关于此功能的官方介绍，请见：InnoDB Integration with memcached 。嵌入memcached之后，整个MySQL的架构如下图所示：

本文接下来的部分，将从源码的角度，详细分析InnoDB Integration with memcached的实现细节问题。

导读

(一) InnoDB Handler层面改动

InnoDB为了支持Memcached接口，在其Handler层面提供了新的Callback方法，这些方法的定义如下：

????????????/** Set up InnoDB API callback function array */

????????????ib_cb_t innodb_api_cb[] = {

????????????????(ib_cb_t) ib_cursor_open_table,

????????????????(ib_cb_t) ib_cursor_read_row,

????????????????(ib_cb_t) ib_cursor_insert_row,

????????????????(ib_cb_t) ib_cursor_delete_row,

????????????????(ib_cb_t) ib_cursor_update_row,

????????????????(ib_cb_t) ib_cursor_next,

????????????????(ib_cb_t) ib_cursor_last,

????????????????(ib_cb_t) ib_tuple_get_n_cols,

????????????????(ib_cb_t) ib_col_set_value,

????????????????(ib_cb_t) ib_col_get_value,

????????????????(ib_cb_t) ib_col_get_meta,

????????????????(ib_cb_t) ib_trx_begin,

????????????????(ib_cb_t) ib_trx_commit,

????????????????(ib_cb_t) ib_trx_rollback,

????????????????(ib_cb_t) ib_trx_start,

????????????????…

????????????};

同时，innodb_api_cb[]数组，被存储在InnoDB Handlerton的data字段中，可以向上传递给InnoDB Memcached Engine。InnoDB Memcached Engine可以调用这些Callback方法，达到直接操作InnoDB Engine的目的。

????????ha_innodb.cc::innobase_init();

????????????…

????????????innobase_hton->data = &innodb_api_cb;

(二) Memcached Plugin的初始化流程

Memcached Plugin，同样注册为MySQL的一个插件(memcached_mysql.cc)，此插件的定义如下：

????????mysql_declare_plugin(daemon_memcached)

????????{

????????????MYSQL_DAEMON_PLUGIN,

????????????&daemon_memcached_plugin,

????????????”daemon_memcached”,

????????????”Oracle Corporation”,

????????????”Memcached Daemon”,

????????????PLUGIN_LICENSE_GPL,

????????????daemon_memcached_plugin_init,????????/* Plugin Init */

????????????daemon_memcached_plugin_deinit,????/* Plugin Deinit */

????????????0×0100 ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?/* 1.0 */,

????????????NULL, ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? /* status variables */

????????????daemon_memcached_sys_var, ? ? ? ? ? ? ?/* system variables */

????????????NULL ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?/* config options */

????????}

????????mysql_declare_plugin_end;

Memcached Plugin插件，初始化为daemon_memcached_plugin_init函数，该函数同样需要一个Handlerton输入，初始化Memcached Plugin插件，尤其是初始化InnoDB Memcached Engine。详细的初始化流程如下所示:

????struct mysql_memcached_context * con;

????// plugin为InnoDB的Handlerton，plugin->data指向InnoDB Handlerton中的data，

????// 在InnoDB的init函数中被初始化为innodb_api_cb数组。innodb_api_cb数组中，

????// 注册了各种InnoDB提供的方法。

????…

????con->memcached_conf.m_innodb_api_cb = plugin->data;

????…

????// 创建Memcached Plugin后台线程，线程的主函数为daemon_memcached_main()，

????// 线程传入参数，即为从InnoDB Handlerton处获取的InnoDB提供的Callback方法

????pthread_create(&con->memcached_thread, &attr,

????????????????daemon_memcached_main, (void *)&con->memcached_conf);

????????…

????????// 创建、加载InnoDB Memcached Engine

????????Engine_loader.c::load_engine();

????????????// Create InnoDB Memcached Engine

????????????// 注册InnoDB Memcached Engine的各种方法，例如：

????????????// 初始化方法：innodb_eng->engine.initialize = innodb_initialize;

????????????innodb_engine.c::create_instance();

????????????????…

????????// 初始化InnoDB Memcached Engine

????????Engine_loader.c::init_engine(engine_handle,engine_config…);

????????????Innodb_engine.c::innodb_initialize(engine, config_str);

????????????????// 注册InnoDB Engine提供的内部方法至InnoDB Memcached Engine，

????????????????// 包括：读取、插入、删除、更新、事务创建/提交/回滚等操作。

????????????????// 至此，后续的InnoDB Memcached Engine的所有操作，均可被映射

????????????????// 为InnoDB Storage Engine提供的内部方法进行实际操作。

????????????????Innodb_api.c::register_innodb_cb();

????????????????// Fetch InnoDB specific settings

????????????????innodb_eng->meta_info = innodb_config(NULL, 0, &innodb_eng->meta_hash);

????????????????????// Find the table and column info that used for memcached data

????????????????????// Cantainers表：InnoDB内部存储元数据的系统表，

????????????????????// MCI_CFG_DB_NAME；MCI_CFG_CONTAINER_TABLE；…

????????????????????innodb_config.c::innodb_config_container();

????????????????…

????????????????// 开启InnoDB Memcached Engine的后台自动提交线程

????????????????innodb_engine.c::innodb_bk_thread();

????????????????????…

????????…

????// 在InnoDB Handlerton中存储更多Memcached Plugin所需要的信息

????plugin->data = (void *)con;

(三) Memcached Plugin如何获取使用InnoDB Handlerton中的innodb_api_cb？

// 在此函数中完成innodb_api_cb的传递

sql_plugin.cc::plugin_initialize();

????// 此函数指针，直接调用到Plugin注册的init方法

????(*plugin_type_initialize[plugin->plugin->type])(plugin);

????…

????// 判断出当前是InnoDB引擎完成了初始化，则将InnoDB Handlerton中

? ? // 的data拷贝到临时变量中

????if (strcmp(plugin->name.str, “InnoDB”) == 0)

????????innodb_callback_data = ((Handlerton *)plugin->data)->data;

????// 判断出当前Plugin为Memcached Plugin，则将临时变量赋值到memcached plugin

? ? // 的data中，完成InnoDB Handlerton提供的Callback方法的传递

????else if (plugin->plugin->init)

????????if (strcmp(plugin->name.str, “daemon_memcached”) == 0)

????????????plugin->data = (void *)innodb_callback_data;

????????// 初始化Memcached Plugin

????????plugin->plugin->init();

(四) InnoDB Memcached Plugin提供的方法集合

????????innodb_engine.c::create_instance();

????????????innodb_eng->engine.interface.interface = 1;

????????????innodb_eng->engine.get_info = innodb_get_info;

????????????innodb_eng->engine.initialize = innodb_initialize;

????????????innodb_eng->engine.destroy = innodb_destroy;

????????????innodb_eng->engine.allocate = innodb_allocate;

????????????innodb_eng->engine.remove = innodb_remove;

????????????innodb_eng->engine.release = innodb_release;

????????????innodb_eng->engine.clean_engine= innodb_clean_engine;

????????????innodb_eng->engine.get = innodb_get;

????????????innodb_eng->engine.get_stats = innodb_get_stats;

????????????innodb_eng->engine.reset_stats = innodb_reset_stats;

????????????innodb_eng->engine.store = innodb_store;

????????????innodb_eng->engine.arithmetic = innodb_arithmetic;

????????????innodb_eng->engine.flush = innodb_flush;

????????????innodb_eng->engine.unknown_command = innodb_unknown_command;

????????????innodb_eng->engine.item_set_cas = item_set_cas;

????????????innodb_eng->engine.get_item_info = innodb_get_item_info;

????????????innodb_eng->engine.get_stats_struct = NULL;

????????????innodb_eng->engine.errinfo = NULL;

????????????innodb_eng->engine.bind = innodb_bind;

所有InnoDB Memcached Engine提供的方法，都是类Memcached方法。例如：get/remove/store方法等。

(五) 一个InnoDB Memcached Plugin的操作的流程

通过Memcached Plugin进行一个删除操作的函数处理流程，如下：

注1：删除操作，对应的InnoDB Handlerton提供的方法为：????remove()

注2：删除操作，对应的InnoDB Memcached Engine的方法为：????ib_cursor_delete_row()

// 通过Memcached接口删除操作的处理流程

memcached.c::process_delete_command();

????settings.engine.v1->remove(settings.engine.v0, c, key, nkey, 0, 0);

????// InnoDB Memcached Engine层面提供的删除方法

????innodb_engine.c::innodb_remove(ENGINE_HANDLE*, cookie, key, nkey, …);

????????…

????????// 初始化InnoDB的连接，传入参数分析：

????????// CONN_MODE_WRITE：当前为写操作

????????// IB_LOCK_X：????????当前操作需要对记录行加X锁

????????conn_data = innodb_conn_init(innodb_eng, cookie, CONN_MODE_WRITE, …);

????????????if (conn_option == CONN_MODE_WRITE)

????????????????// 开始一个事务

????????????????innodb_api.c::innodb_cb_trx_begin();

????????????????????ib_cb_trx_begin() -> api0api.cc::ib_trx_begin();

????????????????// 打开当前表，并设置游标

????????????????innodb_api.c::innodb_api_begin();

????????????????????…

????????// 进行真正的删除操作

????????innodb_api.c::innodb_api_delete(innodb_eng, conn_data, key, nkey);

????????????// 根据传入的key，将游标定位到正确的位置

????????????innodb_api.c::innodb_api_search();

????????????????…

????????????// 如果开启了binlog，则需要将定位到的记录拷贝出来

????????????if (engine->enable_binlog)

????????????????…

????????????// 删除记录

????????????????ib_cb_delete_row() -> api0api.cc::ib_cursor_delete_row();

????????????// 记录binlog

????????????????handler_api.cc::handler_binlog_row();

????????????// 删除构造出来的Tuple对象

????????????ib_cb_tuple_delete() -> api0api.cc::ib_tuple_delete();

(六) InnoDB Engine vs Default Engine

在InnoDB Engine结构的内部(innodb_engine.h::struct innodb_engine)，有两个实例化的Engine Handle，分别为：

????????ENGINE_HANDLE_V1????engine;

????????ENGINE_HANDLE*????????default_engine;

两个engine handle有何区别：

????/** Tells if we will used Memcached default engine or InnoDB Memcached engine to handle the request */

????typedef enum meta_cache_opt {

????????META_CACHE_OPT_INNODB = 1,????????/*!< Use InnoDB Memcached Engine only */

????????META_CACHE_OPT_DEFAULT, ? ? ? ? ? ? /*!< Use Default Memcached Engine only */

????????META_CACHE_OPT_MIX, ? ? ? ? ? ? ? ? ? ? ? ?/*!< Use both, first use default memcached engine */

????????META_CACHE_OPT_DISABLE, ? ? ? ? ? ? ? /*!< This operation is disabled */

????????META_CACHE_NUM_OPT ? ? ? ? ? ? ? ? ? ? ? ?/*!< Number of options */

????} meta_cache_opt_t;

默认的参数值为META_CACHE_OPT_INNODB，仅仅启用InnoDB Engine，数据通过InnoDB引擎提供的Callback方法操作；若同时启用了Memcached Default Engine(META_CACHE_OPT_MIX)，那么数据读取时，首先从Default Engine中读取；记录删除时，也需要先删除Default Engine中的记录；

(七) Container表详解

InnoDB Memcached Engine，包含一个Container Table，用于存储Memcached与InnoDB Table之间的映射关系。

Container Table，必须包含9列，分别是：

????/** Columns in the “containers” system table, this maps the Memcached

????operation to a consistent InnoDB table */

????typedef enum container {

????????CONTAINER_NAME, ? ? ? ? /*!< name for this mapping */

????????CONTAINER_DB, ? ? ? ? ? ? ? ?/*!< database name */

????????CONTAINER_TABLE,????????/*!< table name */

????????CONTAINER_KEY, ? ? ? ? ? ? /*!< column name for column maps to????memcached “key” */

????????CONTAINER_VALUE, ? ? ? /*!< column name for column maps to????memcached “value” */

????????CONTAINER_FLAG, ? ? ? ? ?/*!< column name for column maps to????memcached “flag” value */

????????CONTAINER_CAS, ? ? ? ? ? ? /*!< column name for column maps to????memcached “cas” value */

????????CONTAINER_EXP, ? ? ? ? ? ? /*!< column name for column maps to “expiration” value */

????????CONTAINER_NUM_COLS????/*!< number of columns */

????} container_t;

其中：

????innodb_config.c::innodb_config(NULL, 0, &innodb_eng->meta_hash);

????????innodb_config_meta_hash_init();

????????????…

????????????// 打开CONTAINER Table

????????????innodb_api.c::innodb_api_begin(…, MCI_CFG_CONTAINER_TABLE, …);

????????????innodb_api.c::innodb_cb_read_row();

????????????innodb_config.c::innodb_config_add_item();

????????????????…

????????????????// 针对Value列，需要解析此列对应于InnoDB Table的哪些列

????????????????// InnoDB中的不同列，以” ;,|\n”作为分隔符

????????????????if (i == CONTAINER_VALUE)

????????????????????innodb_config_parse_value_col();

????????????????…

在innodb_engine.cc::innodb_initialize函数，完成Container Table的读取以及Meta Hash的填充之后，后续的Memcached方法，才可以根据规则操作，完成记录在Memcached与InnoDB引擎间的传递。