Deadlock when querying INFORMATION_SCHEMA - sql-server-2005

Question

I have a process which dynamically alters my SQL2K5 table structure according to changes in a published meta-data layer.
For example, if a new column needs to be added and the table has NO dependancies - the steps would be:
1. Create scripts using T-SQL for any indexes & primary keys that already exist on the table [these scripts are included below]
2. Drop the table
3. Re-create the table from the meta-layer that has the new column
4. Execute the scripts created in step#1
5. Populate the table using BulkCopy
The above is initiated via a .NET assembly and runs in 3 concurrent streams on a daily basis.
I am receiving a deadlock error in step #1 - when I access the INFORMATION_SCHEMA tables to script out the indexes/keys. I have used the hint WITH(NOLOCK) in these scripts thinking this should prevent any locking when 3 streams of these actions are running concurrently. A table can only be processed (the create or scripting) in 1 stream.
Is there something more I need to do???
Any comments greatly appreciated.
[Scripts]
ALTER Procedure [dbo].[s$spScriptPrimaryKeyForTable]
#Tablename varchar(100)
AS
-- Get all existing primary keys
DECLARE cPK CURSOR FOR
SELECT TABLE_NAME, CONSTRAINT_NAME
FROM INFORMATION_SCHEMA.TABLE_CONSTRAINTS WITH(NOLOCK)
WHERE upper(TABLE_NAME)=upper(#Tablename)
ORDER BY TABLE_NAME
DECLARE #PkTable SYSNAME
DECLARE #PkName SYSNAME
-- Loop through all the primary keys
OPEN cPK
FETCH NEXT FROM cPK INTO #PkTable, #PkName
WHILE (##FETCH_STATUS = 0)
BEGIN
DECLARE #PKSQL NVARCHAR(4000) SET #PKSQL = ''
SET #PKSQL = 'ALTER TABLE ' + #PkTable + ' ADD CONSTRAINT ' + #PkName + ' PRIMARY KEY CLUSTERED ('
-- Get all columns for the current primary key
DECLARE cPKColumn CURSOR FOR
SELECT COLUMN_NAME
FROM INFORMATION_SCHEMA.KEY_COLUMN_USAGE WITH(NOLOCK)
WHERE TABLE_NAME = #PkTable AND CONSTRAINT_NAME = #PkName
ORDER BY ORDINAL_POSITION
OPEN cPKColumn
DECLARE #PkColumn SYSNAME
DECLARE #PkFirstColumn BIT SET #PkFirstColumn = 1
-- Loop through all columns and append the sql statement
FETCH NEXT FROM cPKColumn INTO #PkColumn
WHILE (##FETCH_STATUS = 0)
BEGIN
IF (#PkFirstColumn = 1)
SET #PkFirstColumn = 0
ELSE
SET #PKSQL = #PKSQL + ', '
SET #PKSQL = #PKSQL + #PkColumn
FETCH NEXT FROM cPKColumn INTO #PkColumn
END
CLOSE cPKColumn
DEALLOCATE cPKColumn
SET #PKSQL = #PKSQL + ')'
-- Print the primary key statement
-- PRINT #PKSQL
FETCH NEXT FROM cPK INTO #PkTable, #PkName
END
CLOSE cPK
DEALLOCATE cPK
SELECT ISNULL(#PKSQL,' ')
================
ALTER Procedure [dbo].[s$spScriptIndexesForTable]
#Tablename varchar(100)
AS
DECLARE #RetVal varchar(4000)
SET #RetVal = ''
-- Get all existing indexes, but NOT the primary keys
DECLARE cIX CURSOR FOR
SELECT OBJECT_NAME(SI.Object_ID), SI.Object_ID, SI.Name, SI.Index_ID
FROM Sys.Indexes SI WITH(NOLOCK)
LEFT JOIN INFORMATION_SCHEMA.TABLE_CONSTRAINTS TC WITH(NOLOCK) ON SI.Name = TC.CONSTRAINT_NAME AND OBJECT_NAME(SI.Object_ID) = TC.TABLE_NAME
WHERE TC.CONSTRAINT_NAME IS NULL
AND OBJECTPROPERTY(SI.Object_ID, 'IsUserTable') = 1
AND upper(OBJECT_NAME(SI.Object_ID))=upper(#Tablename)
ORDER BY OBJECT_NAME(SI.Object_ID), SI.Index_ID
DECLARE #IxTable SYSNAME
DECLARE #IxTableID INT
DECLARE #IxName SYSNAME
DECLARE #IxID INT
-- Loop through all indexes
OPEN cIX
FETCH NEXT FROM cIX INTO #IxTable, #IxTableID, #IxName, #IxID
WHILE (##FETCH_STATUS = 0)
BEGIN
DECLARE #IXSQL NVARCHAR(4000)
--SET #PKSQL = ''
SET #IXSQL = 'CREATE '
-- Check if the index is unique
IF (INDEXPROPERTY(#IxTableID, #IxName, 'IsUnique') = 1)
SET #IXSQL = #IXSQL + 'UNIQUE '
-- Check if the index is clustered
IF (INDEXPROPERTY(#IxTableID, #IxName, 'IsClustered') = 1)
SET #IXSQL = #IXSQL + 'CLUSTERED '
SET #IXSQL = #IXSQL + 'INDEX ' + #IxName + ' ON [' + #IxTable + '] ('
-- Get all columns of the index
DECLARE cIxColumn CURSOR FOR
SELECT SC.Name,IC.[is_included_column],IC.is_descending_key
FROM Sys.Index_Columns IC WITH(NOLOCK)
JOIN Sys.Columns SC WITH(NOLOCK) ON IC.Object_ID = SC.Object_ID AND IC.Column_ID = SC.Column_ID
WHERE IC.Object_ID = #IxTableID AND Index_ID = #IxID
ORDER BY IC.Index_Column_ID,IC.is_included_column
DECLARE #IxColumn SYSNAME
DECLARE #IxIncl bit
DECLARE #Desc bit
DECLARE #IxIsIncl bit set #IxIsIncl = 0
DECLARE #IxFirstColumn BIT SET #IxFirstColumn = 1
-- Loop throug all columns of the index and append them to the CREATE statement
OPEN cIxColumn
FETCH NEXT FROM cIxColumn INTO #IxColumn, #IxIncl, #Desc
WHILE (##FETCH_STATUS = 0)
BEGIN
IF (#IxFirstColumn = 1)
BEGIN
SET #IxFirstColumn = 0
END
ELSE
BEGIN
--check to see if it's an included column
IF ((#IxIsIncl = 0) AND (#IxIncl = 1))
BEGIN
SET #IxIsIncl = 1
SET #IXSQL = #IXSQL + ') INCLUDE ('
END
ELSE
BEGIN
SET #IXSQL = #IXSQL + ', '
END
END
SET #IXSQL = #IXSQL + '[' + #IxColumn + ']'
--check to see if it's DESC
IF #Desc = 1
SET #IXSQL = #IXSQL + ' DESC'
FETCH NEXT FROM cIxColumn INTO #IxColumn, #IxIncl, #Desc
END
CLOSE cIxColumn
DEALLOCATE cIxColumn
SET #IXSQL = #IXSQL + ')'
-- Print out the CREATE statement for the index
--SELECT 'IXSQL: ' + #IXSQL
IF #RetVal IS NULL
SET #RetVal = ''
--SELECT 'Retval: ' + #RetVal
SET #RetVal = #RetVal + #IXSQL + ' '
FETCH NEXT FROM cIX INTO #IxTable, #IxTableID, #IxName, #IxID
END
CLOSE cIX
DEALLOCATE cIX
SELECT ISNULL(#RetVal,' ')

Answer 1

INFORMATION_SCHEMA views are just that - views. You can't update them so they are unlikely to cause any deadlocks. If you want to determine the real source (which I assume has something to do with your alters, or other code within the cursor that you didn't show, or other code you're calling in combination with calling these procedures - since selects against views and then selecting variables can't be the cause), I suggest reading Gail Shaw's blog post on interpreting deadlocks.
In spite of (1) I still suggest using more modern catalog views than INFORMATION_SCHEMA. The same information can be derived from, for example, sys.key_constraints.
You're using the default cursor options; and you're nesting cursors. If you end up still using cursors, you should get in the habit of using a less resource intensive cursor (e.g. LOCAL STATIC FORWARD_ONLY READ_ONLY).
You don't actually need a cursor to do this. Here is how I would re-write the PK table script:
CREATE PROCEDURE dbo.ScriptPKForTable
#TableName SYSNAME
AS
BEGIN
SET NOCOUNT ON;
DECLARE
#pkName SYSNAME,
#clustered BIT,
#object_id INT,
#sql NVARCHAR(MAX);
SELECT
#object_id = OBJECT_ID(UPPER(#TableName));
SELECT
#pkName = kc.name,
#clustered = CASE i.[type]
WHEN 1 THEN 1 ELSE 0 END
FROM
sys.key_constraints AS kc
INNER JOIN
sys.indexes AS i
ON kc.parent_object_id = i.[object_id]
AND kc.unique_index_id = i.index_id
WHERE
kc.parent_object_id = #object_id
AND kc.[type] = 'pk';
SET #sql = N'ALTER TABLE ' + QUOTENAME(#TableName)
+ ' ADD CONSTRAINT ' + #pkName
+ ' PRIMARY KEY ' + CASE #clustered
WHEN 1 THEN 'CLUSTERED' ELSE '' END + ' (';
SELECT
#sql = #sql + c.name + ','
FROM
sys.index_columns AS ic
INNER JOIN
sys.indexes AS i
ON ic.index_id = i.index_id
AND ic.[object_id] = i.[object_id]
INNER JOIN
sys.key_constraints AS kc
ON i.[object_id] = kc.[parent_object_id]
AND kc.unique_index_id = i.index_id
INNER JOIN
sys.columns AS c
ON i.[object_id] = c.[object_id]
AND ic.column_id = c.column_id
WHERE
kc.[type] = 'PK'
AND kc.parent_object_id = #object_id
ORDER BY key_ordinal;
SET #sql = LEFT(#sql, LEN(#sql) - 1) + ');';
SELECT COALESCE(#sql, ' ');
END
GO
As for the index creation script, I think there is a better way to do this (again without explicit cursors, not that avoiding the cursor is the goal, but the code is going to be a LOT cleaner). First you need a function to build either key or include columns from the index:
CREATE FUNCTION dbo.BuildIndexColumns
(
#object_id INT,
#index_id INT,
#included_columns BIT
)
RETURNS NVARCHAR(MAX)
AS
BEGIN
DECLARE #s NVARCHAR(MAX);
SELECT #s = N'';
SELECT #s = #s + c.name + CASE ic.is_descending_key
WHEN 1 THEN ' DESC' ELSE '' END + ','
FROM sys.index_columns AS ic
INNER JOIN sys.columns AS c
ON ic.[object_id] = c.[object_id]
AND ic.column_id = c.column_id
WHERE c.[object_id] = #object_id
AND ic.[object_id] = #object_id
AND ic.index_id = #index_id
AND ic.is_included_column = #included_columns
ORDER BY ic.key_ordinal;
IF #s > N''
SET #s = LEFT(#s, LEN(#s)-1);
RETURN (NULLIF(#s, N''));
END
GO
With that function in place, a ScriptIndexes procedure is pretty easy:
CREATE PROCEDURE dbo.ScriptIndexesForTable
#TableName SYSNAME
AS
BEGIN
SET NOCOUNT ON;
DECLARE
#sql NVARCHAR(MAX),
#object_id INT;
SELECT #sql = N'', #object_id = OBJECT_ID(UPPER(#TableName));
SELECT #sql = #sql + 'CREATE '
+ CASE i.is_unique WHEN 1 THEN 'UNIQUE ' ELSE '' END
+ CASE i.[type] WHEN 1 THEN 'CLUSTERED ' ELSE '' END
+ ' INDEX ' + i.name + ' ON ' + QUOTENAME(#TableName) + ' ('
+ dbo.BuildIndexColumns(#object_id, i.index_id, 0)
+ ')' + COALESCE(' INCLUDE('
+ dbo.BuildIndexColumns(#object_id, i.index_id, 1)
+ ')', '') + ';' + CHAR(13) + CHAR(10)
FROM
sys.indexes AS i
WHERE
i.[object_id] = #object_id
-- since this will be covered by ScriptPKForTable:
AND i.is_primary_key = 0
ORDER BY i.index_id;
SELECT COALESCE(#sql, ' ');
END
GO
Note that my solution does not assume the PK is clustered (your PK script hard-codes CLUSTERED but then your index script assumes that any of the indexes could be clustered). I also ignore additional properties such as filegroup, partitioning, or filtered indexes (not supported in 2005 anyway).