mysql global status 状态变量详解一，innodb%writ相关

Innodb_buffer_pool_write_requests：写入buffer pool的请求数。理论上这个数量也等于每一个提交的事务修改的buffer pool中页*次。同一个页在一个事务中只累加一次，多个事务修改同一个page累加多次。原因是每一个事务在commit时都会将自身修改的page添加到flush列表，每次添加一个page，都会累加Innodb_buffer_pool_write_requests。具体的实现是，每一个mtr有一个自身的memory stack来存储修改的page，这个结构是这样的：
m_list :{mtr_buf_t::block_t,mtr_buf_t::block_t,mtr_buf_t::block_t}
mtr_buf_t::block_t:{mtr_memo_slot_t,mtr_memo_slot_t,mtr_memo_slot_t}
mtr_memo_slot_t:<object,type>
其中，m_list是一个集合，
这个操作是在事务提交时，mtr_buf_t::block_t是mtr_buf_t::block_t类型的元素，mtr_memo_slot_t是一个struct结构体，其中有两个元素，一个是指向page的指针，一个是page的类型。描述一下，就是一个mtr事务，有一个m_list名字的集合类型成员；这个集合成员包含了多个mtr_buf_t::block_t类型的元素；mtr_buf_t实际是dyn_buf_t类型，一个mtr_buf_t::block_t类型的元素可以包含的数据默认512字节，最大64k；一个mtr_memo_slot_t元素就是一个对buffer中page的包装，包括一个void指针，一个无符号long，总共16字节。
Innodb_buffer_pool_write_requests在代码中以srv_stats.buf_pool_write_requests变量指代。在下面的buf_flush_note_modification函数中累加。将mtr修改的page塞到m_list中发生在void mtr_t::memo_push(void *object, mtr_memo_type_t type)函数中。

buf0flu.ic
/** This function should be called at a mini-transaction commit, if a page was
 modified in it. Puts the block to the list of modified blocks, if it is not
 already in it. */
UNIV_INLINE
void buf_flush_note_modification(
    buf_block_t *block,      /*!< in: block which is modified */
    lsn_t start_lsn,         /*!< in: start lsn of the mtr that
                             modified this block */
    lsn_t end_lsn,           /*!< in: end lsn of the mtr that
                             modified this block */
    FlushObserver *observer) /*!< in: flush observer */
{
#ifdef UNIV_DEBUG
  {
    /* Allow write to proceed to shared temporary tablespace
    in read-only mode. */
    ut_ad(!srv_read_only_mode ||
          fsp_is_system_temporary(block->page.id.space()));
    ut_ad(buf_block_get_state(block) == BUF_BLOCK_FILE_PAGE);
    ut_ad(block->page.buf_fix_count > 0);

    buf_pool_t *buf_pool = buf_pool_from_block(block);

    ut_ad(!buf_flush_list_mutex_own(buf_pool));
  }
#endif /* UNIV_DEBUG */

  mutex_enter(&block->mutex);

  if (end_lsn != 0) {
    ut_ad(block->page.newest_modification <= end_lsn);
    block->page.newest_modification = end_lsn;
  } else {
    /* Do nothing. This is the case with no-redo mtr and it
    possibly could re-modify some earlier dirtied page, in
    which case we don't want to change oldest/newest lsns.
    If that's not the case, we will set newest_modification
    within buf_flush_insert_into_flush_list call. That's
    because we can't read the value now, because we don't
    hold the flush list mutex yet. */
  }

  /* Don't allow to set flush observer from non-null to null,
  or from one observer to another. */
  ut_ad(block->page.flush_observer == nullptr ||
        block->page.flush_observer == observer);
  block->page.flush_observer = observer;

  if (block->page.oldest_modification == 0) {
    buf_pool_t *buf_pool = buf_pool_from_block(block);

    buf_flush_insert_into_flush_list(buf_pool, block, start_lsn);
  } else if (start_lsn != 0) {
    ut_ad(block->page.oldest_modification <= start_lsn);
  }

  buf_page_mutex_exit(block);

  srv_stats.buf_pool_write_requests.inc();
}

//mtr0mtr.h
struct mtr_memo_slot_t {
  /** pointer to the object */
  void *object;

  /** type of the stored object (MTR_MEMO_S_LOCK, ...) */
  ulint type;
};


/**
Pushes an object to an mtr memo stack. */
void mtr_t::memo_push(void *object, mtr_memo_type_t type) {
  ut_ad(is_active());
  ut_ad(object != nullptr);
  ut_ad(type >= MTR_MEMO_PAGE_S_FIX);
  ut_ad(type <= MTR_MEMO_SX_LOCK);
  ut_ad(m_impl.m_magic_n == MTR_MAGIC_N);
  ut_ad(ut_is_2pow(type));

  /* If this mtr has x-fixed a clean page then we set
  the made_dirty flag. This tells us if we need
  insert the dirtied page to the flush list in mtr_co