innodb中master线程的调度的算法改进(mysql 5.6.26)

作者：周琳

QQ:715169549源码交流群：196380905

1.master线程的调度：/*********************************************************************//**

The master thread controlling the server.

os_thread_ret_t DECLARE_THREAD(srv_master_thread)

如下是核心代码：

ulint old_activity_count = srv_get_activity_count();

if (srv_force_recovery >= SRV_FORCE_NO_BACKGROUND) {

goto suspend_thread;

}

while (srv_shutdown_state == SRV_SHUTDOWN_NONE) {

/*在linux下就是1000000/1000000，在针对这个整除的值%1000000;

*这个就是os_thread_sleep( microseconds);的计算方式

srv_master_sleep(); //

MONITOR_INC(MONITOR_MASTER_THREAD_SLEEP);

/*很显然并没有按照1秒和10秒的算法来，而是通过当前服务器活动的数量来决定。

if (srv_check_activity(old_activity_count)) { //srv_check_activity(old_activity_count)返回：return(srv_sys->activity_count != old_activity_count);

old_activity_count = srv_get_activity_count(); //srv_get_activity_count返回:return(srv_sys->activity_count);

srv_master_do_active_tasks();

} else {

srv_master_do_idle_tasks();

}

}

while (srv_master_do_shutdown_tasks(&last_print_time)) {

/* Shouldn't loop here in case of very fast shutdown*/ut_ad(srv_fast_shutdown< 2);

}

}

srv_master_do_active_tasks做的事情：//#define PCT_IO(p) ((ulong) (srv_io_capacity * ((double) (p) / 100.0)))//# define SRV_MASTER_CHECKPOINT_INTERVAL (7)

1.log_free_check，检查是否有空闲的log buffer。2.ibuf_contract_in_background(0, FALSE),合并PCT_IO(5)个插入缓冲3.srv_sync_log_buffer_in_background,后台同步刷新log buffer日志到log file4.cur_time % SRV_MASTER_CHECKPOINT_INTERVAL == 0,就做log_checkpoint(TRUE, FALSE)，而log_checkpoint仅仅是检查bp中的lsn以及把lsn的值写入到redo file中。

真正做检查点刷新脏页到磁盘的是log_make_checkpoint_at函数。

srv_master_do_idle_tasks做的事情：1.log_free_check，检查是否有空闲的log buffer。2.ibuf_contract_in_background(0, TRUE),合并PCT_IO(100)个插入缓冲.3.srv_sync_log_buffer_in_background,刷新log buffer到磁盘。4.cur_time % SRV_MASTER_CHECKPOINT_INTERVAL == 0,就做log_checkpoint(TRUE, FALSE)，而log_checkpoint仅仅是检查bp中的lsn以及把lsn的值写入到redo file中。

真正做检查点刷新脏页到磁盘的是log_make_checkpoint_at函数。

如下是srv_sys_t结构：

元素包含一个线程表和一个task queue。/** The server system struct*/

structsrv_sys_t{

ib_mutex_t tasks_mutex;/*!< variable protecting the

tasks queue*/UT_LIST_BASE_NODE_T(que_thr_t)

tasks;/*!< task queue*/ib_mutex_t mutex;/*!< variable protecting the

fields below.*/ulint n_sys_threads;/*!< size of the sys_threads

array*/srv_slot_t* sys_threads; /*!< server thread table*/ulint n_threads_active[SRV_MASTER+ 1];/*!< number of threads active

in a thread class*/srv_stats_t::ulint_ctr_1_t

activity_count;/*!< For tracking server

activity*/};

接下来介绍下activity_count,这个函数是由srv_inc_activity_count(srv_sys->activity_count.inc();)进行自增的,这个函数会在哪些动作中调用呢？

我们看下调用该函数的函数流程：1.row_undo_step：在表中回滚一行的操作->srv_inc_activity_count()2.srv_active_wake_master_thread：如果需要主线程处理，切主线程处于挂起状态就唤醒主线程。->srv_inc_activity_count()3.srv_wake_master_thread:如果主线程挂起或者正在被挂起，唤醒主线程。->srv_inc_activity_count()

上述函数还被很多函数调用，activity_count可以理解为mysql运行中操作的计数器，每个操作都会累加。所以master线程依据时间间隔的

activity_count计数器来判断执行srv_master_do_active_tasks还是srv_master_do_idle_tasks。可以看出这个计数器如果在1秒以后这个值如果

不发生变化就切换到srv_master_do_idle_tasks。2.page cleaner线程：(buf0flu.c中)

核心代码：

ulint next_loop_time= ut_time_ms() + 1000;while (srv_shutdown_state == SRV_SHUTDOWN_NONE) {//服务器没有关闭的状态下

/*The page_cleaner skips sleep if the server is

idle and there are no pending IOs in the buffer pool

and there is work to do.*/

if (srv_check_activity(last_activity) //检查当前服务器的活动次数

||buf_get_n_pending_read_ios()|| n_flushed == 0) {

page_cleaner_sleep_if_needed(next_loop_time);//休眠min(1000000,(next_loop_time-当前时间)*1000))

}

next_loop_time= ut_time_ms() + 1000; //重新计算当前next_loop_time值

if (srv_check_activity(last_activity)) { //检查当前服务器活动数量

last_activity =srv_get_activity_count();/*Flush pages from end of LRU if required*/n_flushed=buf_flush_LRU_tail();/*Flush pages from flush_list if required*/n_flushed+=page_cleaner_flush_pages_if_needed();

}else{

n_flushed=page_cleaner_do_flush_batch(

PCT_IO(100), //刷新srv_io_capacity个脏页到磁盘

LSN_MAX);

}

}if (srv_fast_shutdown == 2) {/*In very fast shutdown we simulate a crash of

buffer pool. We are not required to do any flushing*/

gotothread_exit;

}do{

n_flushed= page_cleaner_do_flush_batch(PCT_IO(100), LSN_MAX); //刷新srv_io_capacity个脏页到磁盘

/*We sleep only if there are no pages to flush*/

if (n_flushed == 0) {

os_thread_sleep(100000); //休眠1秒

}

}while (srv_shutdown_state ==SRV_SHUTDOWN_CLEANUP);do{

success= buf_flush_list(PCT_IO(100), LSN_MAX, &n_flushed);

buf_flush_wait_batch_end(NULL, BUF_FLUSH_LIST);

}while (!success || n_flushed > 0);