sql-server sql-server-2000 full-text-search full-text-indexing

Tell me SQL Server Full-Text searcher is crazy, not me

i have some customers with a particular address that the user is searching for:

123 generic way

There are 5 rows in the database that match:

ResidentialAddress1
=============================
123 GENERIC WAY
123 GENERIC WAY
123 GENERIC WAY
123 GENERIC WAY
123 GENERIC WAY

i run a FT query to look for these rows. i'll show you each step as i add more criteria to the search:

SELECT ResidentialAddress1 FROM Patrons
WHERE CONTAINS(Patrons.ResidentialAddress1, '"123*"')

ResidentialAddress1
=========================
123 MAPLE STREET
12345 TEST
123 MINE STREET
123 GENERIC WAY
123 FAKE STREET
...

(30 row(s) affected)

Okay, so far so good, now adding the word "generic":

SELECT ResidentialAddress1 FROM Patrons
WHERE  CONTAINS(Patrons.ResidentialAddress1, '"123*"')
AND CONTAINS(Patrons.ResidentialAddress1, '"generic*"')

ResidentialAddress1
=============================
123 GENERIC WAY
123 GENERIC WAY
123 GENERIC WAY
123 GENERIC WAY
123 GENERIC WAY

(5 row(s) affected)

Excellent. And now i'l add the final keyword that the user wants to make sure exists:

SELECT ResidentialAddress1 FROM Patrons
WHERE  CONTAINS(Patrons.ResidentialAddress1, '"123*"')
AND CONTAINS(Patrons.ResidentialAddress1, '"generic*"')
AND CONTAINS(Patrons.ResidentialAddress1, '"way*"')


ResidentialAddress1            
------------------------------ 

(0 row(s) affected)

Huh? No rows? What if i query for just "way*":

SELECT ResidentialAddress1 FROM Patrons
WHERE CONTAINS(Patrons.ResidentialAddress1, '"way*"')

ResidentialAddress1            
------------------------------ 

(0 row(s) affected)

At first i thought that perhaps it's because of the *, and it's requiring that the root way have more characters after it. But that's not true:

Searching for "123*" matches "123"
Searching for "generic*" matches "generic"
Books online says, The asterisk matches zero, one, or more characters

What if i remove the * just for s&g:

SELECT ResidentialAddress1 FROM Patrons
WHERE CONTAINS(Patrons.ResidentialAddress1, '"way"')

Server: Msg 7619, Level 16, State 1, Line 1
A clause of the query contained only ignored words.

So one might think that you are just not allowed to even search for way, either alone, or as a root. But this isn't true either:

SELECT * FROM Patrons
WHERE CONTAINS(Patrons.*, '"way*"')

AccountNumber FirstName Lastname
------------- --------- --------
33589         JOHN      WAYNE

So sum up, the user is searching for rows that contain all the words:

123 generic way

Which i, correctly, translate into the WHERE clauses:

SELECT * FROM Patrons
WHERE CONTAINS(Patrons.*, '"123*"')
AND CONTAINS(Patrons.*, '"generic*"')
AND CONTAINS(Patrons.*, '"way*"')

which returns no rows. Tell me this just isn't going to work, that it's not my fault, and SQL Server is crazy.

Note: i've emptied the FT index and rebuilt it.

Update One

SELECT Lastname, ResidentialAddress1 FROM Patrons
WHERE CONTAINS(Patrons.*, '"gen*"')

Lastname                  ResidentialAddress1            
------------------------- ------------------------------ 
SAVE                      123 GENERIC WAY
Genders                   
SAVE                      123 GENERIC WAY
Patron                    123 GENERIC WAY
SAVE                      123 GENERIC WAY
SAVE                      234 GENERIC WAY
SAVE                      123 GENERIC WAY

(7 row(s) affected)

Update Two

Pretending the user typed in:

123 generic wa

SELECT ResidentialAddress1 FROM Patrons
WHERE  CONTAINS(Patrons.ResidentialAddress1, '"123*"')
AND CONTAINS(Patrons.ResidentialAddress1, '"generic*"')
AND CONTAINS(Patrons.ResidentialAddress1, '"wa*"')

ResidentialAddress1            
------------------------------ 

(0 row(s) affected)

The real problem is that the user is typing in something perfectly valid, and they would expect to see what any human being would expect to see.

Update Three

Someone asked for all this, it's not my fault!:

CREATE TABLE [dbo].[Patrons] (
    [PatronGUID]  uniqueidentifier ROWGUIDCOL  NOT NULL ,
    [AccountNumber] [bigint] NULL ,
    [FirstName] [varchar] (25) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [MiddleInitial] [varchar] (1) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [Lastname] [varchar] (25) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [EyeColor] [varchar] (30) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [HairColor] [varchar] (30) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [Gender] [varchar] (1) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [Birthday] [datetime] NULL ,
    [Height] [int] NULL ,
    [Weight] [int] NULL ,
    [FacialHair] [tinyint] NULL ,
    [Nationality] [varchar] (50) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [IdentifyingMarks] [varchar] (30) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [DriversLicenseNumber] [varchar] (25) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [DriversLicenseRegion] [varchar] (20) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [DriversLicenseCountry] [varchar] (2) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [DriversLicenseExpires] [datetime] NULL ,
    [DriversLicenseDateVerified] [datetime] NULL ,
    [PassportNumber] [varchar] (25) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [PassportRegion] [varchar] (20) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [PassportCountry] [varchar] (2) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [PassportExpires] [datetime] NULL ,
    [PassportDateVerified] [datetime] NULL ,
    [OtherIdentificationNumber] [varchar] (25) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [OtherIdentificationRegion] [varchar] (20) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [OtherIdentificationCountry] [varchar] (2) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [OtherIdentificationType] [varchar] (50) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [OtherIdentificationExpires] [datetime] NULL ,
    [OtherIdentificationDateVerified] [datetime] NULL ,
    [ResidentialAddress1] [varchar] (30) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [ResidentialAddress2] [varchar] (30) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [ResidentialAddress3] [varchar] (30) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [ResidentialCity] [varchar] (25) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [ResidentialZipCode] [varchar] (15) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [ResidentialRegion] [varchar] (20) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [ResidentialCountry] [varchar] (50) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [ResidentialPhoneNumber] [varchar] (20) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [CountryOfResidence] [varchar] (50) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [BusinessAddress1] [varchar] (30) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [BusinessAddress2] [varchar] (30) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [BusinessAddress3] [varchar] (30) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [BusinessCity] [varchar] (25) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [BusinessRegion] [varchar] (20) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [BusinessZipCode] [varchar] (15) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [BusinessCountry] [varchar] (50) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [BusinessName] [varchar] (25) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [BusinessPhone] [varchar] (20) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [PositionWithFirm] [varchar] (30) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [EmployerTelephone] [varchar] (20) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [MemberCardType] [varchar] (1) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [PlayerStatusCode] [varchar] (50) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [AccountType] [varchar] (1) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [AccountStatus1] [varchar] (1) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [AccountStatus2] [varchar] (50) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [IsVIPExchangeRate] [tinyint] NULL ,
    [ChangedUserGUID_Depricated] [uniqueidentifier] NULL ,
    [ChangedUser] [varchar] (50) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [ChangedDate] [datetime] NULL ,
    [ChangedWorkstation] [varchar] (50) COLLATE SQL_Latin1_General_CP1_CI_AS NULL ,
    [PendingUpdates_Depricated] [varchar] (255) COLLATE SQL_Latin1_General_CP1_CI_AS NULL 
) ON [PRIMARY]
GO

ALTER TABLE [dbo].[Patrons] ADD 
    CONSTRAINT [DF_Patrons_PatronGUID] DEFAULT (newid()) FOR [PatronGUID],
    CONSTRAINT [PK_Patrons] PRIMARY KEY  NONCLUSTERED 
    (
        [PatronGUID]
    ) WITH  FILLFACTOR = 90  ON [PRIMARY] 
GO

if (select DATABASEPROPERTY(DB_NAME(), N'IsFullTextEnabled')) <> 1 
exec sp_fulltext_database N'enable' 

GO

if not exists (select * from dbo.sysfulltextcatalogs where name = N'TheFullTextCatalog')
exec sp_fulltext_catalog N'TheFullTextCatalog', N'create' 

GO

exec sp_fulltext_table N'[dbo].[Patrons]', N'create', N'TheFullTextCatalog', N'PK_Patrons'
GO

exec sp_fulltext_column N'[dbo].[Patrons]', N'FirstName', N'add', 1033  
GO

exec sp_fulltext_column N'[dbo].[Patrons]', N'MiddleInitial', N'add', 1033  
GO

exec sp_fulltext_column N'[dbo].[Patrons]', N'Lastname', N'add', 1033  
GO

exec sp_fulltext_column N'[dbo].[Patrons]', N'EyeColor', N'add', 1033  
GO

exec sp_fulltext_column N'[dbo].[Patrons]', N'IdentifyingMarks', N'add', 1033  
GO

exec sp_fulltext_column N'[dbo].[Patrons]', N'ResidentialAddress1', N'add', 1033  
GO

exec sp_fulltext_column N'[dbo].[Patrons]', N'ResidentialAddress2', N'add', 1033  
GO

exec sp_fulltext_column N'[dbo].[Patrons]', N'ResidentialAddress3', N'add', 1033  
GO

exec sp_fulltext_column N'[dbo].[Patrons]', N'ResidentialCity', N'add', 1033  
GO

exec sp_fulltext_column N'[dbo].[Patrons]', N'ResidentialZipCode', N'add', 1033  
GO

exec sp_fulltext_column N'[dbo].[Patrons]', N'ResidentialRegion', N'add', 1033  
GO

exec sp_fulltext_column N'[dbo].[Patrons]', N'ResidentialCountry', N'add', 1033  
GO

exec sp_fulltext_column N'[dbo].[Patrons]', N'ResidentialPhoneNumber', N'add', 1033  
GO

exec sp_fulltext_column N'[dbo].[Patrons]', N'CountryOfResidence', N'add', 1033  
GO

exec sp_fulltext_table N'[dbo].[Patrons]', N'activate'  
GO

Here's the screenshots for the guy who didn't believe me:

The query that should work but doesn't:

The query that works, but isn't useful:

The query that works, but isn't useful, with the proof content:

Update Four

The query cannot be written as

CONTAINS(Patrons.*, 'words...')

Since there are items not logically or physically covered by the FT index. e.g. the user queries for:

6/4/2010 ian boyd 619

Presents four keywords:

6/4/2010
ian
boyd
619

This means they want all the conditions to hold true, with pseudo-code being:

WHERE 6/4/2010 is in the row
AND ian is in the row
AND boyd is in the row
AND 619 is in the row

Which is translated into a partial query of:

WHERE --Keyword 1: 6/4/2010
(
   ((ChangedDate >= '20100604') AND (ChangedDate < '20100605'))
   OR 
   ((LastTransactionDate >= '20100604') AND (LastTransactionDate < '20100605'))
   OR 
   (CONTAINS(Patrons.*, '"6/4/2010*"')
)
AND --Keyword 2: ian
(
    CONTAINS(Patrons.*, '"ian*"')
)
AND --Keyword 3: boyd
(
    CONTAINS(Patrons.*, '"boyd*"')
)
AND --Keyword 4: 619
(
    (AccountNumber IN (SELECT CAST(619 AS bigint)))
    OR
    (CONTAINS(Patrons.*, '"619*"'))
)

One of the answerers was looking at the simplified example presented in the original question; not the real world. To say that it is incorrect to have multiple AND clauses is nieve.

Solution

The message is telling you that "way" is a stopword, which means it's ignored and not indexed. That's why you can find "wayne" but not "way".

So, no, it's not crazy and neither are you. There's just a simple misunderstanding.