MySQL中文件排序中单双路排序

zhbbupt 2013-03-21

在执行相关的SQL语句中,在mysql里进程容易出现一些file temp类似的字眼,该字眼告诉我们我们的宿主sql采用了文件排序等操作,这可能会极大的影响我们的需要获取结果集的效率问题。

mysql中有2中文件排序算法:单路排序和双路排序;他们的区别于MYSQL中的max_length_for_sort_date具有相关性;单路算法的排序可以为排序中的每一行创建固定的缓冲区;如果库中出现超大字符串的时候,比如;BLOB,TEXT等可以采用前缀排序的算法,但是其值不能超过参数规定的值大小,在生产环境下,建议采用更多的是单路的排序算法,他可以使磁盘的读写效率更高,sort_merge_passes的值会加大等!以下是相关innodb中给出的提示;

ut_ad(error == DB_SUCCESS);

 /* Commit the data dictionary transaction in order to release
 the table locks on the system tables.  This means that if
 MySQL crashes while creating a new primary key inside
 row_merge_build_indexes(), indexed_table will not be dropped
 by trx_rollback_active().  It will have to be recovered or
 dropped by the database administrator. */
 trx_commit_for_mysql(trx);

 row_mysql_unlock_data_dictionary(trx);
 dict_locked = FALSE;

 ut_a(trx->n_active_thrs == 0);
 ut_a(UT_LIST_GET_LEN(trx->signals) == 0);

 if (UNIV_UNLIKELY(new_primary)) {
  /* A primary key is to be built.  Acquire an exclusive
  table lock also on the table that is being created. */
  ut_ad(indexed_table != prebuilt->table);

  error = row_merge_lock_table(prebuilt->trx, indexed_table,
          LOCK_X);

  if (UNIV_UNLIKELY(error != DB_SUCCESS)) {

   goto error_handling;
  }
 }

 /* Read the clustered index of the table and build indexes
based on this information using temporary files and merge sort</span>. */
 error = row_merge_build_indexes(prebuilt->trx,
     prebuilt->table, indexed_table,
     index, num_of_idx, table); <span style="color: rgb(51, 102, 255);">----handler0alter.cc指定句柄操作过程中文件排序带来的merge的操作

  for (i = 0; i < n_index; i++) {
   row_merge_buf_t* buf = merge_buf[i];
   merge_file_t*  file = &files[i];
   const dict_index_t* index = buf->index;

   if (UNIV_LIKELY
      (row && row_merge_buf_add(buf, row, ext))) {
    file->n_rec++;
    continue;
   }

   /* The buffer must be sufficiently large
   to hold at least one record. */
   ut_ad(buf->n_tuples || !has_next);

   /* We have enough data tuples to form a block.
   Sort them and write to disk. */

   if (buf->n_tuples) {
    if (dict_index_is_unique(index)) {
     row_merge_dup_t dup;
     dup.index = buf->index;
     dup.table = table;
     dup.n_dup = 0;

     row_merge_buf_sort(buf, &dup);<span style="color: rgb(51, 102, 255);">---row0merge.c文件中的,merge算法中排序的指定

相关推荐