MySQL内核技术之“索引查询”

MySQL的索引查询内部使用的是quick range结构。先让我们看看调用流程：

join->optimize()-->
make_join_plan()-->
estimate_rowcount()-->
get_quick_record_count()-->
test_quick_select()-->
...
join_init_read_record()（这个函数设置要使用的读取操作为rr_quick()）-->
read_record（实际上是QUICK_RANGE_SELECT::get_next()）-->
ha_innobase::multi_range_read_next-->
DsMrr_impl::dsmrr_next-->
handler::multi_range_read_next-->
如果是第一次：mrr_funcs.next(mrr_iter, &mrr_cur_range)实际上是调用：quick_range_seq_next
如果不是：read_range_next

quick_range_seq_next最重要的作用是设置范围查询的上下边界：

uint quick_range_seq_next(range_seq_t rseq, KEY_MULTI_RANGE *range)
{
  QUICK_RANGE_SEQ_CTX *ctx= (QUICK_RANGE_SEQ_CTX*)rseq;

  if (ctx->cur == ctx->last)
    return 1; /* no more ranges */

  QUICK_RANGE *cur= *(ctx->cur);
  key_range *start_key= &range->start_key;
  key_range *end_key=   &range->end_key;

  start_key->key=    cur->min_key;
  start_key->length= cur->min_length;
  start_key->keypart_map= cur->min_keypart_map;
  start_key->flag=   ((cur->flag & NEAR_MIN) ? HA_READ_AFTER_KEY :
                      (cur->flag & EQ_RANGE) ?
                      HA_READ_KEY_EXACT : HA_READ_KEY_OR_NEXT);
  end_key->key=      cur->max_key;
  end_key->length=   cur->max_length;
  end_key->keypart_map= cur->max_keypart_map;
  /*
    We use HA_READ_AFTER_KEY here because if we are reading on a key
    prefix. We want to find all keys with this prefix.
  */
  end_key->flag=     (cur->flag & NEAR_MAX ? HA_READ_BEFORE_KEY :
                      HA_READ_AFTER_KEY);
  range->range_flag= cur->flag;
  ctx->cur++;
  return 0;
}

上面函数中，设置上下边界的的代码：

start_key->key=    cur->min_key; 为5
end_key->key=      cur->max_key; 为7
start_key->flag=HA_READ_KEY_OR_NEXT；
end_key->flag=HA_READ_BEFORE_KEY；

执行完quick_range_seq_next，开始真正执行读取操作：
read_range_first

/** @brief
  Read first row between two ranges.
  Store ranges for future calls to read_range_next.

  @param start_key        Start key. Is 0 if no min range
  @param end_key        End key.  Is 0 if no max range
  @param eq_range_arg            Set to 1 if start_key == end_key
  @param sorted        Set to 1 if result should be sorted per key

  @note
    Record is read into table->record[0]

  @retval
    0            Found row
  @retval
    HA_ERR_END_OF_FILE    No rows in range
  @retval
    \#            Error code
*/
int handler::read_range_first(const key_range *start_key,
                  const key_range *end_key,
                  bool eq_range_arg,
                              bool sorted /* ignored */)
{
  int result;
  DBUG_ENTER("handler::read_range_first");

  eq_range= eq_range_arg;
  set_end_range(end_key, RANGE_SCAN_ASC);

  range_key_part= table->key_info[active_index].key_part;

  if (!start_key)            // Read first record
    result= ha_index_first(table->record[0]);
  else
    result= ha_index_read_map(table->record[0],
                              start_key->key,
                              start_key->keypart_map,
                              start_key->flag);
  if (result)
    DBUG_RETURN((result == HA_ERR_KEY_NOT_FOUND) 
        ? HA_ERR_END_OF_FILE
        : result);

  if (compare_key(end_range) <= 0)
  {
    DBUG_RETURN(0);
  }
  else
  {
    /*
      The last read row does not fall in the range. So request
      storage engine to release row lock if possible.
    */
    unlock_row();
    DBUG_RETURN(HA_ERR_END_OF_FILE);
  }
}

这里会设置结束边界：set_end_range(end_key, RANGE_SCAN_ASC); end_key为7。接下来读取第一条记录：

  if (!start_key)            // Read first record
    result= ha_index_first(table->record[0]);
  else
    result= ha_index_read_map(table->record[0],
                              start_key->key,
                              start_key->keypart_map,
                              start_key->flag);

接下来sub_select会执行read_range_next来读取下一条数据。无论在read_range_next还是read_range_first中，最后面都有下面这个判断：

  if (compare_key(end_range) <= 0)
  {
    DBUG_RETURN(0);
  }
  else
  {
    /*
      The last read row does not fall in the range. So request
      storage engine to release row lock if possible.
    */
    unlock_row();
    DBUG_RETURN(HA_ERR_END_OF_FILE);
  }
}

这个就是用来判断读取的记录是不是到达了上界，如果是则返回HA_ERR_END_OF_FILE结束。