Full Text Searching Eventually Utilizes 100% CPU

For one day, this SPROC executes very quickly to return results on a Full Text catalog.

ALTER Procedure dbo.sp_RSSHarvestedHeadlines_FullTextSearch
(
@ORKeywords varchar(4000) = 'xxxx',
@ANDKeywords varchar(4000) = 'xxxx',
@NOTKeywords varchar(4000) = 'xxxx',
@SourceID int = -1,
@IsHidden bit = null
)
As
set nocount on

SELECT HHL.HeadlineID,
HHL.Title,
HHL.Link,
HHL.[Description],
HHL.PubDate,
HHL.GMTDateAdded,
RSSSources.SourceTitle,
RSSSources.SourceLink
FROM RSSHarvestedHeadlines HHL
INNER JOIN RSSSources ON HHL.SourceID = RSSSources.SourceID
WHERE HHL.PublishedFlag = 0
AND (@IsHidden is null OR HHL.HideFlag = @IsHidden)
AND (@SourceID = -1 OR HHL.SourceID = @SourceID)
AND (@ORKeywords = 'xxxx' OR (CONTAINS(HHL.Title, @ORKeywords) OR CONTAINS(HHL.Description, @ORKeywords)))
AND (@ANDKeywords = 'xxxx' OR (CONTAINS(HHL.Title, @ANDKeywords) OR CONTAINS(HHL.Description, @ANDKeywords)))
AND (
@NOTKeywords = 'xxxx'
OR (
(NOT CONTAINS(HHL.Title, @NOTKeywords)
AND
NOT CONTAINS(HHL.Description, @NOTKeywords))
)
)

ORDER BY HHL.GMTDateAdded DESC, HHL.PubDate DESC



But somethign happens overnight and in the morning the sproc times out. While running (even from a new query window on the SQL 2005 server itself) it utilizes 100% CPU until it times out.


When I pass default parameters to the sproc (not using any part of the query that uses Full Text) the sproc returns every record in the database very quickly. No hang ups.

But the moment I add any text in say the @ORKeywords parameter, for example, the sproc utilizes 100% CPU for maybe 15 seconds and then times out.

By accident I discovered that I can fix this temporarily by copying the database. I don't use the new copy or anything. Just the act of copying the database fixes it. The sproc then executes normally, and quickly. But the next morning it's back to slow again.

Note, over night I am adding about 1000 records to the table.

Would automatic updates to the FT Catalog choke on 1000 records?

Also note that one of the fields being cataloged is a "Text" field (blob). Would that cause this?

Would what text is being added to the table matter? What if an invalid character was added (like some european character or a control character)? Would FT indexing hang up on that?

I am at a loss.


The following code will recreate the table:

CREATE TABLE [dbo].[RSSHarvestedHeadlines](
[HeadlineID] [int] IDENTITY(1,1) NOT NULL,
[SourceID] [int] NOT NULL,
[Title] [varchar](500) COLLATE SQL_Latin1_General_CP1_CI_AS NOT NULL,
[Link] [varchar](1000) COLLATE SQL_Latin1_General_CP1_CI_AS NOT NULL,
[Description] [ntext] COLLATE SQL_Latin1_General_CP1_CI_AS NULL,
[PubDate] [datetime] NULL,
[GMTDateAdded] [datetime] NOT NULL CONSTRAINT [DF_RSSHarvestedHeadlines_GMTDateAdded] DEFAULT (getutcdate()),
[GMTLastHarvested] [datetime] NOT NULL CONSTRAINT [DF_RSSHarvestedHeadlines_GMTLastHarvested] DEFAULT (getutcdate()),
[HideFlag] [bit] NOT NULL CONSTRAINT [DF_RSSHarvestedHeadlines_HideFlag] DEFAULT ((0)),
[PublishedFlag] [bit] NOT NULL CONSTRAINT [DF_RSSHarvestedHeadlines_PublishedFlag] DEFAULT ((0)),
[EditStamp] [timestamp] NOT NULL,
CONSTRAINT [PK_RSSHarvestedHeadlines] PRIMARY KEY CLUSTERED
(
[HeadlineID] ASC
)WITH (IGNORE_DUP_KEY = OFF) ON [PRIMARY]
) ON [PRIMARY] TEXTIMAGE_ON [PRIMARY]

GO