Question

I have a table defined as follows:

CREATE TABLE [dbo].[T](
[FileID] [bigint] NOT NULL,
[Line] [nvarchar](250) NULL
) ON [PRIMARY]

FileID is a foreign key, and the table has > 7,000,000 records

Doing a select count(1) on this table causes a memory spike of 1.5GB+. Any workaround for this?

Answer 1

If you don’t need to block all your users while you’re getting a count, and can accept minor differences due to transactions currently occurring, then this is much better:

SELECT SUM([rows]) FROM sys.partitions
  WHERE [object_id] = OBJECT_ID('dbo.T')
  AND index_id IN (0,1);

Otherwise Marc is right – put an index on the FileID column and SQL Server will choose to scan that index instead of scanning the entire table.

Is there really no index on this table? There are a few use cases where no clustered index makes sense, but I don’t suspect this is one of them. While creating a foreign key can help the optimizer, it does not create an index (though that is a common misconception).