Question

当我在SQL Server Profiler中测量Writes和RowCounts时，我发现一个语句（删除）报告Writes = 26035，但RowCounts为0.这怎么可能？显然，该语句是在写行时删除行，那么为什么这些行都不计入rowcount列？

Answer 1

~~因为rowcount指向返回给客户端的行数。除非使用OUTPUT子句，否则删除不会返回任何行。~~

更新：==＆gt;我站得更正测试显示当delete没有返回行时也会设置rowcount。

证明：

SELECT * 
  FROM Accounts 
 WHERE Category= 'COA'  
   AND Code between 1500 and 2000 
-- 18 Reads, 0 Writes, 51 RowCount

DELETE FROM Accounts 
 WHERE Category = 'COA' 
   AND Code between 1500 and 2000
-- 22 Reads, 1 Writes, 51 RowCount

DELETE FROM Accounts
OUTPUT DELETED.* 
WHERE Category = 'COA' 
  AND Code between 2000 and 4000
-- 24 Reads, 3 Writes, 103 RowCount

在包含500万条记录的2个表上运行删除语句，这些记录不会删除任何记录，而是提供以下查询计划::

delete clust from clust, heap where clust.Key= heap.Key
-- 19854 Reads, 0 Writes, 0 RowCount

 |--Clustered Index Delete(OBJECT:([dbo].[clust].[idx_clust]), OBJECT:([dbo].[clust].[idx2_clust]))
      |--Top(ROWCOUNT est 0)
           |--Parallelism(Gather Streams)
                |--Hash Match(Right Semi Join, HASH:([dbo].[heap].[Key])=([dbo].[clust].[Key]))
                     |--Bitmap(HASH:([dbo].[heap].[Key]), DEFINE:([Bitmap1012]))
                     |    |--Parallelism(Repartition Streams, Hash Partitioning, PARTITION COLUMNS:([dbo].[heap].[Key]))
                     |         |--Stream Aggregate(GROUP BY:([dbo].[heap].[Key]))
                     |              |--Index Scan(OBJECT:([dbo].[heap].[idx_heap]), ORDERED FORWARD)
                     |--Parallelism(Repartition Streams, Hash Partitioning, PARTITION COLUMNS:([dbo].[clust].[Key]))
                         |--Index Scan(OBJECT:([dbo].[clust].[idx2_clust]),  WHERE:(PROBE([Bitmap1012],[dbo].[clust].[Key],N'[IN ROW]')) ORDERED FORWARD)

在2个小表上运行相同的查询，每个表有10行，结果如下：

delete smallclust from smallclust, smallheap where smallclust.srp_key = smallheap.common_key
-- 45 Reads, 0 Writes, 0 RowCount

 |--Table Delete(OBJECT:([dbo].[smallclust]))
      |--Top(ROWCOUNT est 0)
           |--Nested Loops(Left Semi Join, WHERE:([dbo].[smallheap].[Key]=[dbo].[smallclust].[Key]))
                |--Table Scan(OBJECT:([dbo].[smallclust]))
                |--Table Scan(OBJECT:([dbo].[smallheap]))

因此，导致写入的哈希表的假设仍然不确定。

SQL Server Profiler写入与RowCounts

1 个答案: