Microsoft KB Archive/153096

= FIX: SQL Server 6.5 Service Pack 1 Fixlist =

Article ID: 153096

Article Last Modified on 8/9/2006

-

APPLIES TO


 * Microsoft SQL Server 6.5 Service Pack 1

-



This article was previously published under Q153096



The following is a list of fixes and other various improvements that have been made in the Microsoft SQL Server version 6.5 Service Pack 1. Service Pack 1 is now available from your primary support provider. For more information, contact your primary support provider.

Please note that workarounds described have been provided for your information only. It is not necessary to implement these workarounds if you have the updated software.

DB-LIBRARY COMPONENTS
Q151301: FIX: dbcursorfetchex Can Cause Blocking in DB-Library

ODBC COMPONENTS
Q153694: FIX: SQLNumResultCols Fails with Incorrect Syntax Near 'SET' Q153908: FIX: 16-bit Driver Times Out on Async Calls to SQLTables

SERVER COMPONENTS
Q149938: FIX: AV on INSERT of UNIONed SELECTs to Table With DEFAULTs Q149939: FIX: ROLLBACK of TRUNCATE TABLE May Cause Error 3301, AV Q150775: FIX: CREATE TABLE in Stored Procedure May Fail With Error 1750 Q150894: FIX: LOAD TRAN May Fail With Handled AV Q150896: FIX: SH_PAGE Locks May Be Held on Inner Tables of JOINs Q150897: FIX: LOAD TRAN May Fail With Error 1511 Q150900: FIX: UPDATE May Fail With Error 107 Q151111: FIX: DECLARE CURSOR on Temp Table with FOR UPDATE Causes AV Q151590: FIX: Error 4409 Generated When Using Multiple Database Views Q151591: FIX: DBCC Checkident Always Reports the Seed Value as Current Q151693: FIX: Err Msg "Unknown Token Received from SQL Server" Q151765: FIX: SELECT INTO Inside a Temp Procedure Causes Client to Hang Q151985: FIX: SELECTs from SYSPROCESSES Result in Access Violation Q151988: FIX: 1108 Errors with Heavy Tempdb Activity Q152353: FIX: Select with CASE Statement Inside View Can Cause Client AV Q152416: FIX: sp_processmail Will Only Process One Query per Execution Q152615: FIX: Syntax Error in SELECT List May Cause Handled AV Q152690: FIX: Count(*) May Not Return Result Set Q152709: FIX: Update of BLOB Data w/SQL Server 2.65.0201 Driver Stops Q153006: FIX: Dump or Load w/Tape Devices Results in Errors, AVs Q153079: FIX: SQL Server Stops w/ Temporary Procedure in a Transaction Q153186: FIX: Filtered Stored Procedures Cannot Reference Multiple Tables Q153455: FIX: Multiple "LEFT OUTER JOIN" in FROM Clause May Cause Error Q153780: FIX: sp_droppublisher Does Not Clear 'pub' Server Option Q153802: FIX: Deadlock During Cursor Update Causes CPU Spin and Spinloop Q153803: FIX: GRANT ALL Fails to Give Permissions on Stored Procedures Q153836: FIX: SQLGetData Fails on Multiple Active Statement Handles Q153851: FIX: AV if 4 or More Correlated Subqueries with Self-Join Q153855: FIX: IRL can cause 3307, OS error 6, unhandled server level AV Q153954: FIX: Checkpoint Process Deadlock Results in Errors 603, 3314 Q153961: FIX: Full Memory and Cursors Cause AVs, 707, 706, and Spins Q153987: FIX: AV Error Using Temp Table and Cursor in Stored Procedure Q153992: FIX: Cannot Rename a Column with a Quoted Identifier Q154018: FIX: AV in Update Mode When Script Is Run Twice Q154047: FIX: SQL Terminates on Delete Table with 15 Self-References

FIX: dbcursorfetchex Can Cause Blocking in DB-Library
151301 BUG #: NT 15039 (6.50)



SYMPTOMS
If a DB-library application calls dbcursorfetch or dbcursorfetchex successfully in one thread, subsequent calls to dbcursoropen or dbclose from other threads using the same DBPROCESS would be blocked. Sp_who and sp_lock show no blockage on the server side, and the DB-library application would appear to hang.

Calling SQLNumResultCols after SQLPrepare of a SELECT statement that contains a subquery in the select list will fail with:

szSqlState = "37000", pfNativeError = 156

szErrorMsg="[Microsoft][ODBC SQL Server Driver][SQL Server]

Incorrect syntax near the keyword 'SET'."

When a 16-bit ODBC application calls SQLTables using the Microsoft SQL Server ODBC Driver version 2.65.0201 set in asynchronous mode, a "Timeout expired" error message appears immediately after the second async retry.

This problem occurs with any network library configuration and Windows NT local pipes. The Timeout error message also appears regardless of other login or query timeouts you configure on the client computer.

INSERTing to a table from a UNION of two or more SELECT statements can result in a handled Access Violation.

A TRUNCATE TABLE command that is aborted or rolled back may get a 3301 error "Invalid log record found in Syslogs (logop 42)" and an access violation.

If the truncated table has an IDENTITY column and the TRUNCATE aborts, or is inside a user transaction which later does a ROLLBACK, this problem can occur.

EXECuting a stored procedure that creates a temp or permanent table with defaults may fail with the error:

[INTERNAL ERROR] unable to locate original param in voidptr list.

Msg 1750, Level 16, State 0

Unable to create constraint. See previous errors.

LOAD TRANSACTION may fail with the errors:

udwritem: Operating system error 6(The handle is invalid.) on device ....

ex_testhandle: stack overflow, top=0x17a7f10, end=0x17a7f10

EXCEPTION_ACCESS_VIOLATION

The actual stack of the AV in this case can be many different places. The SQL Server may now become unresponsive and have to be restarted. The database being loaded will have to be recreated.

A join of tables may hold the shared page locks on the inner tables of the join(s) for the duration of the SELECT.

If the transaction log being LOADed contains the log records for a CREATE CLUSTERED INDEX on a large table (> 150 MB), then the transaction may fail to load with the 1511 error.

Error : 1511, Severity: 20, State: 8

Sort cannot be reconciled with transaction log

If using the 204 trace flag for backwards compatibility, UPDATEs using JOINs may fail with an error 107.

Msg 107, Level 15, State 1

The column prefix 'so' does not match with a table name or alias name used in the query

A DECLARE CURSOR statement that has a FOR UPDATE clause with multiple columns in the column list can generate a handled access violation if the table in the SELECT clause is a temporary table.

Under certain conditions, you may receive 4409 errors. The problem arises when the first view in the chain of views can be executed but subsequent views are unavailable for use. For example:

If you have viewA and viewC in the master database and viewB in the pubs database, where viewA selects * from viewB and viewB selects * from viewC and viewC selects * from sysdatabases. And, SQL Server has been stopped while a client continues to try to execute a select * from viewA. When SQL Server is restarted, the master database is always recovered first and then the subsequent databases. As soon as master is recovered, the client attempts to execute the select and receives the 4409 error because pubs has not yet been recovered.

The same behavior can occur when you take a database on and offline, or you try to drop and create viewC in the above scenario while someone is trying to access it.

When running dbcc checkident, the current identity value will always be reported as the original seed value.

Checking identity information: current identity value '1', maximum column value '14'. DBCC execution completed. If DBCC printed error messages, see your System Administrator.

The rest of the dbcc completes successfully and if need be the next identity value is corrected. However, the message will continue to report the seed value as the current identity value.

FIX: Err Msg "Unknown Token Received from SQL Server"
151693 BUG #: NT: 15056 (6.50)

When you set the statement options to use a server-side cursor and prepare a select statement on a SQL Server system table, the first execution of the select creates the cursor successfully. After you close this cursor, if you execute the prepared statement again, the following error message appears:

unknown token received from SQL Server.

In SQL Server version 6.5, executing a temporary stored procedure containing a SELECT INTO statement causes the client to stop responding. The server never finishes the execution and the control is not returned to the client. CPU utilization on the server computer goes above 95 percent and persists, with the result that the server slows down drastically. This does not prevent other clients from connecting to SQL Server, but the queries executed from these clients will be slow.

On busy symmetric multiprocessing (SMP) computers, queries against the virtual table sysprocesses may infrequently get a handled access violation, causing the client connection to be terminated. Other clients are unaffected. The call stack of the access violation will look similar to the following:

findwaitfor+0x2f ins_sysproc+0x7fc make_fake+0xe8 s_setuptables+0x2bc s_execute+0x7d8 sequencer+0x23f execproc+0xfb7 s_execute+0xb8c sequencer+0x23f language_exec+0x65e

FIX: 1108 Errors with Heavy Tempdb Activity
151988 BUG #: NT: 15209 (6.50)

SQL Servers with lots of tempdb activity, sorts, and deadlocks can get 1108 errors in the error log and at the client. The following error message appears:

Error : 1108, Severity: 21, State: 1

Cannot deallocate extent 944, database 2. Object id 0, index id 0,

status 0 in extent does not match object id -49604, index id 0, status 0

in object being deallocated. Run DBCC CHECKALLOC

You can get 1108 errors in tempdb when there are sorts of small work tables and some other event occurs such as a deadlock or cancel, causing a backout. Normally, the client is already backing out for some other reason so they don't even notice the error, but it does show up in the errorlog.

FIX: Select with CASE Statement Inside View Can Cause Client AV
152353 BUG #: WINDOWS: 15383 (6.00 and 6.50)

A View that has a Select statement with a CASE construct that does a sub- select with an IN clause may cause the client to access violate (AV) during a Select from the View.

The following sample is the problem View definition:

create view MyView as  Select   Column1, Column2= Case When SomeColumn In           (Select  SomeOtherColumn              From   SomeOtherTable) Then SomeValue Else Null End From MyTable

A Select from MyView may cause the client to access violate.

The stored procedure sp_processmail will only process one query sent via e- mail each time it is run if SQLMail is running with Microsoft Exchange Client software.

The absence of a comma between columns in the SELECT list may cause a handled AV under some narrow circumstances.

The parser incorrectly perceives that an alias is intended. A reference to the object later in the SELECT query, usually in a GROUP BY or ORDER BY clause triggers the parser confusion.

Count(*) may not return a result set. This problem occurs on tables with 34 or more columns. When a query is executed, it may return the following message:

This command did not return data, and it did not return any rows

This behavior has also been observed for SELECT-INTO from that table to another table.

When you select from a view which has a definition like select * from another view you may not return results. The base view definition would be on a table with 34 columns or more.

Update of a BLOB data causes SQL Server 2.65.0201 driver to stop with a syntax error. When an update statement is sent to the driver, it generates an invalid statement that results in the syntax error. This happens during an update of a large BLOB data (larger than 64K).

The syntax error generated is

Incorrect syntax near the keyword 'UPDATE'.

State:37000, Native:156, Origin:[Microsoft][ODBC SQL Server

Driver][SQL Server]

Line 1"Incorrect syntax near '='

State:37000, Native:170, Origin:[Microsoft][ODBC SQL Server

Driver][SQL Server]

If a Transact-SQL query contains a GROUP BY clause and an index is not defined on the column involved in GROUP BY, the query requires many more i/o[ASCII 146]s than expected, esulting in slower performance. This problem occurs only only on the Alpha platform.

Attempts to dump or load with tape devices may generate errors 3201, tbswritecheck, or access violations (AVs). Msg 3201 and the tbswritecheck errors are seen when attempting to dump; the AV may occur during the load.

When a temporary stored procedure is created and executed within a user- defined transaction and the client (DBLIB or ODBC) disconnects without either committing or rolling back the transaction, SQL Server stops running. The SQL Service manager shows a red light.

An attempt to kill the process that initiated the transaction also causes SQL Server to stop running.

The Filtered Stored procedures used to facilitate horizontal partitioning in SQL Server 6.50 replication may cause the Logreader task to keep retrying with the following message:

ConnectionTransact (GetOverLappedResult).Possible network error:

Write to SQL Server. Failed. Connection broken.

This error prevents logreader from processing transactions, effectively stopping replication. The problem occurs when rows that do not match the restriction clause are inserted. This behavior is only exhibited in SQL Server version 6.50.

Multiple LEFT OUTER JOIN tables in the FROM clause of a SELECT query may cause SQL Server error 803. For example, the following query:

select authors.au_id from (((( authors left outer join titleauthor on authors.au_id=titleauthor.au_id)  left outer join titles on titleauthor.title_id=titles.title_id) left outer join authors a1 on a1.au_id=titleauthor.au_id)  left outer join roysched on titles.title_id=roysched.title_id) where (roysched.royalty = 20)

will cause the following SQL Server error:

Msg 803, Level 20, State 2

Unable to place buffer 0x0 holding logical page 424 in sdes for object

'titleauthor' - either there is no room in sdes or buffer already in

requested slot.

The SQL Server is terminating this process.



WORKAROUND
Use a separate DBPROCESS for each thread.

FIX: SQLNumResultCols Fails with Incorrect Syntax Near 'SET'
153694 BUG #: NT: 15314 (6.50)

Where possible convert the offending query to a SQL Server VIEW.

FIX: 16-bit Driver Times Out on Async Calls to SQLTables
153908 BUG #: NT: 15563 (2.65.0201)

To work around this problem, do not call the ODBC driver in asynchronous mode. This can be done with different applications and they apply as follows:


 * For ODBC programs written directly to the ODBC API, call SQLSetStmtOption with SQL_ASYNC_ENABLE (set to SQL_ASYNC_ENABLE_OFF).
 * For Visual Basic 3.0 and Access 2.0 applications, set DisableAsync to 1 in the [ODBC] section of in the Vb.ini, App.ini or Msacc20.ini respectively.
 * For Visual Basic 4.0, set the INIPATH property of DBENGINE to point to Vb.ini or App.ini that has DisableAsync set to 1.

For MSACC20.INI entry settings applicable to Access 95, please see the following article in the Microsoft Knowledge Base:

139044 : INF: How to Add Former MSACC20.INI ODBC Section to Registry

Jet 3.x (a base component of Access 95 and Visual Basic 4.0) registry information is on Appendix C of the Microsoft Jet Database Engine Programmer[ASCII 146]s Guide. You can obtain Jet 2.x database engine information from the Microsoft Developer Network Library Level 1 by querying the following:

jet database engine connectivity neil black

You can also reference the Technical Backgrounder called "Jet Database Engine ODBC Connectivity."

Implement the INSERT as distinct INSERT SELECTs for each table in the UNION. If there is a UNIQUE INDEX on the target table, use the IGNORE_DUP_KEY option on that INDEX to filter out duplicate rows that would previously have been filtered out by the UNION. In many cases, this method can be much faster anyway because it can eliminate one or more intermedate work tables.

FIX: ROLLBACK of TRUNCATE TABLE May Cause Error 3301, AV
149939 BUG #: NT: 14849 (6.50)

If the purpose of the TRUNCATE TABLE is just to remove all the rows, use DELETE with no WHERE clause. If the purpose is also to reset the IDENTITY value, use DELETE with no WHERE clause then TRUNCATE TABLE, and make sure there is no user defined transaction at the time of the TRUNCATE TABLE via logic such as

delete t1 while @@trancount > 0 begin commit tran end truncate table t1

FIX: CREATE TABLE in Stored Procedure May Fail With Error 1750
150775 BUG #: NT: 14884 (6.50)

Use some other method than a CONSTRAINT, such as ISNULL, to get the default values you want into the table.

FIX: LOAD TRAN May Fail With Handled AV
150894 BUG #: 14952 (6.50)

DUMP the database and re-sync the transaction logs.

FIX: SH_PAGE Locks May Be Held on Inner Tables of JOINs
150896 BUG #: NT: 15329 (6.50)

Use non-unique indexes instead of unique.

FIX: LOAD TRAN May Fail With Error 1511
150897 BUG #: NT: 15114 (6.50)

DUMP the whole DATABASE after a CREATE CLUSTERED INDEX on a large table.

FIX: UPDATE May Fail With Error 107
150900 BUG #: NT: 14984 (6.50)

Turn off the 204 trace flag or rewrite the JOIN as a subquery.

FIX:DECLARE CURSOR on Temp Table with FOR UPDATE Causes AV
151111 BUG #: NT: 15086 (6.00)

Do not use the FOR UPDATE clause for more than one column in case of a temporary table. Use a permanent table if more than one column is absolutely necessary in the FOR UPDATE clause.

FIX: Error 4409 Generated When Using Multiple Database Views
151590 BUG #: NT: 14645 (6.00)

Drop and add the views when you are sure no one is accessing them.

FIX: DBCC Checkident Always Reports the Seed Value as Current
151591 BUG #: 14759 (6.50)

Use a forward-only cursor instead of a static, keyset, or dynamic cursor on the system tables. Note that the problem does not occur if a server-side cursor is created on a user-defined table or view.

If a forward-only cursor is unacceptable, prepare the select statement again on the system table and execute it to create a server-side cursor. Note that the first execution works fine and applications typically do not need to create a cursor on a system table repeatedly.

FIX: SELECT INTO Inside a Temp Proc Causes Client to Hang
151765 BUG #: NT: 15113 (6.50)

Use permanent stored procedure in place of temporary stored procedure. For ODBC clients, clear the Generate Stored Procedures for Prepared Statements option check box in the ODBC SQL Server Driver Setup dialog box, or set the SQL_USE_PROCEDURE_FOR_PREPARE option in the SQLSetConnectOption function to SQL_UP_OFF.

Try to avoid the CASE construct inside a View Definition or perform some pre-processing for the CASE logic before the View Definition by using the IF-ELSE construct and Temporary Tables, and then define the View on the Temporary Table.

FIX: sp_processmail Will Only Process One Query per Execution
152416 BUG#: 15475

If you are scheduling sp_processmail as a task, you have the following options:


 * 1) Change the task to run every minute. Note that this will work as long as there is not more than one query per minute sent to SQL Server via e-mail.
 * 2) Make multiple tasks to run sp_processmail that run every minute. This provides a workaround for option 1 above.
 * 3) Change the sp_processmail stored procedure not to delete the mail it responds to by commenting the xp_deletemail call.
 * 4) Change the sp_processmail stored procedure to not run in a loop but to instead do several iterations.

FIX: Syntax Error in SELECT List May Cause Handled AV
152615 BUG#: 15027 (6.0)

Add the comma to the SELECT clause.

FIX: Count(*) May Not Return Result Set
152690 BUG#: 15418 (6.5)

You can use the following query instead of the Count(*) on the table

select rows from sysindexes where name like 'My_Table_name'

To perform SELECT-INTO from the source table to the destination table, create the destination table and perform the following query:

insert Destination_Table select * from Source_Table

FIX: Update of BLOB Data w/SQL Server 2.65.0201 Driver Stops
152709

Sort the columns in the GROUP BY clause. Alternately, define an index encompassing the columns involved in the GROUP BY clause and make sure the index is used when grouping the columns in the query.

FIX: Dump or Load w/Tape Devices Results in Errors, AVs
153006 BUG #: NT: 15499 (6.50)

Do not dump to a tape device; dump and load using a different type of dump device, such as disk. If dumping to disk, use another backup package, such as Windows NT Backup, to archive the dump file to tape.

Use a permanent stored procedure instead of a temporary stored procedure within a transaction.

You can also commit or roll back the transaction before the client disconnects.

To avoid encountering the bug you can:


 * Publish the entire table instead of using horizontal partitioning.

-OR-
 * Create a permanent table with the data from the original tables selectively siphoned out (using triggers with the partition condition) and then publish the new table.

FIX: Multiple "LEFT OUTER JOIN" in FROM Clause May Cause Error
153455 BUG #: NT: 15565 (6.50)

In the master database of any server participating in replication, please replace the sp_droppublisher procedure by running the following script:

use master go if exists (select * from sysobjects where sysstat & 0xf = 4 and name = 'sp_droppublisher') drop procedure sp_droppublisher go create procedure sp_droppublisher (  @publisher varchar (30),        /* publisher server name */    @type varchar (5) = null     /* null or 'dist' */         ) as    declare @distaccount varchar(127) declare @proc varchar (255) declare @retcode int declare @privilege varchar (30)

/*   ** parameter check:  @publisher. ** check to make sure that the publisher exists, that the name isn't   ** null, and that the name conforms to the rules for identifiers. */

if @publisher is null begin raiserror (14043, 16, -1, 'the publisher') return (1) end

execute @retcode = sp_validname @publisher

if @retcode <> 0 return (1)

/*   ** perform special logic if dropping a publisher for a distribution ** server. */    if lower(@type) = 'dist' begin if not exists (select *     from master..sysservers                where srvname = @publisher                and srvstatus & 16 <> 0) begin raiserror (14080, 11, -1) return (1) end

execute @retcode = sp_serveroption @publisher, 'dpub', false if @@error <> 0 or @retcode <> 0 return (1)

if exists (select * from master..sysremotelogins         where remoteserverid = (select srvid from master..sysservers where srvname = @publisher)         and remoteusername = 'sa'          and suid = 1)    /* 'sa' */ begin execute @retcode = sp_dropremotelogin @publisher, sa, sa         if @@error <> 0 or @retcode <> 0 return (1) end if exists (select * from master..sysremotelogins         where remoteserverid = (select srvid from master..sysservers where srvname = @publisher)         and remoteusername = 'probe'          and suid = 10)   /* 'probe' */ begin execute @retcode = sp_dropremotelogin @publisher, probe, probe if @@error <> 0 or @retcode <> 0 return (1) end

return (0) end

/*   ** make sure the server is defined as a 'publisher'. */    if not exists (select *                     from master..sysservers                    where srvname = @publisher                      and srvstatus & 2 <> 0) begin raiserror (14080, 11, -1) return (1) end

/*   ** turn off the server option to indicate that this is a publisher. */    execute @retcode = sp_serveroption @publisher, 'pub', false if @@error <> 0 or @retcode <> 0 return (1)

/*   ** fetch the publisher's distributor account. */

select @proc = rtrim(@publisher) + '.master..sp_helpdistributor ' exec @retcode = @proc @account = @distaccount output if @@error <> 0 or @retcode <> 0 begin raiserror (14071, 16, -1) return (1) end

/*   ** if @distaccount = 'localsystem' assume 'admin' privilege */    if @distaccount = 'localsystem' return (0)

/*   ** if @distaccount has 'admin' privilege, do not revoke */    execute @retcode = master.dbo.xp_logininfo @distaccount, 'all', @privilege = @privilege output if @@error <> 0 or @retcode <> 0 return (1)

if @privilege = 'admin' return (0) /*   ** revoke replication privilege to the distributor nt account. */    exec @retcode = master.dbo.xp_revokelogin @distaccount if @@error <> 0 or @retcode <> 0 return (1) go

To clear the problem before applying the procedure, you can manually disable the 'pub' server option by executing the following statement on the subscription server:

use master go sp_serveroption, 'pub', false go

Example:

use master go sp_serveroption AIKMAN, 'pub', false go



MORE INFORMATION
When the SQL Server ODBC driver constructs the T-SQL statement that will be sent to the server to resolve the number of columns that will be returned in the resultset, it incorrectly parses the initial query and excludes the final table name. For example:

SQLPrepare(hstmt, "SELECT t1.c1, (SELECT t2.c1 FROM t2) FROM t1", 44) SQLNumResultCols(hstmt, pcol1)

This results in the following statement being sent to the server on the call to SQLNumResultCols:

SET FMTONLY ON SELECT t1.c1, (SELECT t2.c1 FROM t2) FROM SET FMTONLY OFF

Hence, causing SQL Server to report "Incorrect syntax near the Keyword 'SET'."

Microsoft client/server database applications, such as Access 2.0, Visual Basic 3.0 and Visual Basic 4.0 (16-bit), are designed on top of the Jet Database engine. Therefore, they attempt to make calls to SQLTables during several ODBC operations, which include attaching to a SQL Server 6.5 table. These applications utilize ODBC in asynchronous mode, by default. Unless the default asynchronous behavior is turned off, the "Timeout Expired" error message will occur with the driver version 2.65.0201.

NOTE: Microsoft Excel and Microsoft Query are not designed on top of the Jet Database engine, thus the problem does not occur.

For more information about the Jet Database Engine, please refer to the Microsoft Jet Database Engine Programmer's Guide published by Microsoft Press. The ISBN number is 1-55615-877-7. You can order guide by calling (800) MSP-RESS.

FIX: AV on INSERT of UNIONed SELECTs to Table With DEFAULTs
149938 BUG #: NT: 14833 (6.00 and 6.50)

At a certain threshold of data, a work table needs to be created to resolve the UNION correctly. If the target table has DEFAULTs, these are not applied to intermediate work tables but other attributes of the target table, such as NOT NULL, are applied, and this would cause the query to fail with a different error if there was not the AV.

If there is some activity in the stored procedure before the CREATE TABLE, this error can occur. In the simplest case, executing the following stored procedure will fail:

create proc sp_test1 as begin IF ( @@error <> 0 ) BEGIN GOTO cleanup END CREATE TABLE  #t1 (c1         int NOT NULL DEFAULT -1) cleanup: end go declare @rc int EXEC @rc = sp_test1 SELECT @rc go

The page locks on the inner table(s) of joins where a unique index matches the join on the inner table are not released until the end of the SELECT. On earlier versions, SQL Server would obtain and release the locks one at a time as the page chain is traversed. The outer table's page locks are not held. If the tables are large, this can be a significant concurrency impact. This did not happen in prior versions of SQL Server.

This behavior is also seen in Microsoft SQL Server 6.0.

With ODBC clients the same behavior is seen if:


 * 1) Generate Stored Procedures for Prepared Statements option checkbox in the ODBC SQL Server Driver Setup dialog box is checked.
 * 2) SQL_USE_PROCEDURE_FOR_PREPARE option in the SQLSetConnectOption function is set to SQL_UP_ON.

and a SELECT INTO statement is prepared and executed.

FIX: SELECTs from SYSPROCESSES Result in Access Violation
ARTICLE-ID: 151985 BUG #: NT: 15280 (6.00 and 6.50)

If the SQL Server 4030 trace is turned on, the 2.65.0201 SQL server driver generates the following syntax when an update statement of BLOB data is issued:

SELECT TEXTPTR(imagecol) FROM imagetest where intcol=update imagetest set imagecol=0x00

This is an invalid syntax.

This does not happen in 2.50.0121 (6.0 drivers).They generate:

SELECT TEXTPTR(imagecol) FROM imagetest where imagecol LIKE 0x000000

WRITETEXT BULK imagetest.imagecol 0xa

FIX: SQL Server Stops w/ Temporary Procedure in a Transaction
153079 BUG# 15570 (6.50)

The SQL Server errorlog (under SQL..\LOG directory on the server) would look like:

Error : 631, Severity: 21, State: 1

spid12 The length of 116 passed to delete row routine for the row at

offset 32 is incorrect on the following page: Page pointer = 0xa8b000,

pageno = 104, status = 0x101, objectid = 6, indexid = 0

kernel WARNING: Process being freed while holding Dataserver semaphore

Error : 631, Severity: 21, State: 1

The length of 116 passed to delete row routine for the row at offset 32 is

incorrect on the following page: Page pointer = 0xa8b000, pageno = 104,

status = 0x101, objectid = 6, indexid = 0

kernel udread: Operating system error 6(The handle is invalid.) on device

'C:\SQL60\DATA\MASTER.DAT' (virtpage 0x000002fb).

Buffer 8d8700 from database 'master' has page number 0 in the page header

and page number 759 in the buffer header

Recursive error 822 in ex_print

kernel mirrorproc: i/o error on primary device 'C:\SQL60\DATA\MASTER.DAT'

FIX: Filtered Stored Procedures Cannot Reference Multiple Tables
153186 BUG #: NT: 15451 (6.50)

FIX: Deadlock During Cursor Update Causes CPU Spin and Spinloop
153802 BUG #: 15422 (6.50)

SYMPTOMS
A group of updates through cursors that result in the cursor being a deadlock victim can cause the following error to appear in the error log:

closetable:table already closed for sdes %d

After this error occurs, the process can become unkillable, and its status in sysprocesses is marked as "spinloop." At this point SQL Server becomes very unresponsive and will often stop responding entirely, allowing no one to log on or shut it down. Processes that were accessing the tables become blocked.

WORKAROUND
When using cursors for updates make sure they will not get into a deadlock situation.

FIX: GRANT ALL Fails to Give Permissions on Stored Procedures
153803 BUG #: 15102 (6.50)

SYMPTOMS
When a user attempts to GRANT ALL to a stored procedure, the permissions can fail to be applied. No errors are reported; SQL Server simply does not give permission to the stored procedure. Once this happens to a stored procedure, it will always happen, and GRANT ALL will never work for that stored procedure.

WORKAROUND
To grant permissions to a stored procedure, use GRANT EXEC instead of GRANT ALL.

FIX: SQLGetData Fails on Multiple Active Statement Handles
153836 BUG #: 15053 (6.50)

SYMPTOMS
When there are two active statement handles, SQLGetData fails to fetch results from the two statement handles simultaneously. The following error is generated:

szErrorMsg="[Microsoft][ODBC SQL Server Driver]Connection is busy with results for another hstmt"

WORKAROUND
Bind the result set columns using SQLBindCol before a fetch. This allows multiple active statement handles.

MORE INFORMATION
When you are using server-based cursors, the connection between the client and server does not remain busy between operations. This allows you to have multiple cursors statements active at the same time. However, with the SQL Server ODBC driver 2.65.0121v, an attempt to fetch data using SQLGetData between multiple statement handles fails. If SQLFetch is being done on a statement handle hstmt1 and if, before a result of SQL_NO_DATA_FOUND is returned, another statement hstmt2 is allocated and another fetch operation is done, and then a simultaneous SQLGetData is done on hstmt1, the result will be a "Connection is busy with results" error.

FIX: AV if 4 or More Correlated Subqueries with Self-Join
153851 BUG #: 14802 (6.50)

SYMPTOMS
If a complex query does a self-join and has four or more subqueries, all referencing the same table, it may get a thread-level access violation (AV) if the table has only a non-clustered index.

WORKAROUND
Convert the non-clustered index to a clustered index, or drop it.

FIX: IRL can cause 3307, OS error 6, unhandled server level AV
153855 BUG #: NT: 15569 (6.50)

SYMPTOMS
If a table is set to enable Insert - Row Locking (IRL) with "sp_tableoption 'table_name', 'Insert row lock', true" within a transaction and the user fails to end the transaction with 'commit tran' or 'rollback tran' before exiting, it could cause 3307 "Process %ld was expected to hold logical lock on page %ld.", OS error 6, 602 "Could not find row in Sysindexes for dbid '%d', object '%Id',index '%d'." and an unhandled server level access violation (AV).

MORE INFORMATION
On a single processor computer, when the user exits without 'commit tran', it can cause the following error in the errorlog:

Error : 3307, Severity: 21, State: 1

Process 10 was expected to hold logical lock on page 336.

WARNING: Process being freed while holding Dataserver semaphore

udread: Operating system error 6(The handle is invalid.) on device

'C:\MSSQL\DATA\MASTER.DAT' (virtpage 0x000009a4).

udread: Operating system error 6(The handle is invalid.) on device

'C:\MSSQL\DATA\MASTER.DAT' (virtpage 0x00000394).

udread: Operating system error 6(The handle is invalid.) on device

'C:\MSSQL\DATA\MASTER.DAT' (virtpage 0x00000383).

Buffer 8d3b20 from database 'master' has page number 0 in the page header

and page number 895 in the buffer header

Recursive error 822 in ex_print

mirrorproc: i/o error on primary device 'C:\MSSQL\DATA\MASTER.DAT'

SQL Server will terminate itself afterwards. When SQL Server is restarted and DBCC CHECKDB("pubs") is executed, the results sometimes report error 602: Could not find row in Sysindexes for dbid '%d', object '%Id',index '%d'.

After a user exits on a multi-processor computer, sp_who shows the spid is still valid and the status shows 'spinloop', which does not allow you to kill the spid. DBCC CHECKDB or other queries will either hang, or become extremely slow. There will also be many bufwait errors in the errorlog. Stopping SQL Server would either cause an AV or cause SQl Server to stop responding.

FIX: Checkpoint Process Deadlock Results in Errors 603, 3314
153954 BUG #: NT: 15307 (6.50)

SYMPTOMS
A stored procedure that performs the following actions:


 * 1) creates a temporary table
 * 2) inserts data into the table
 * 3) creates a cursor for read-only access that is then used to update the contents of the temporary table
 * 4) drops the cursor
 * 5) drops the temporary table

can, if executed by multiple users, result in the Checkpoint Process being chosen as a deadlock victim (spid 3). In this case, the following message will come up three times:

Your server command (process id 3) was deadlocked with another process and has been chosen as a deadlock victim. Re-run your command.

The message will be followed by the following errors:

603: There are not enough system session descriptors available to run this query. The maximum number available to a process is %d. Split query and rerun.

3314: Error while undoing log row in database '%.*s'. %S_RID.

A thread-level Access Violation will also be displayed.

If you run sp_who prior to receiving 1105 errors in tempdb, the Checkpoint Process is not present, and if the client processes continue to run, the result is that tempdb eventually fills because the log is not being truncated.

FIX: Full Memory and Cursors Cause AVs, 707, 706, and Spins
153961 BUG #: NT: 14828 (6.50)

SYMPTOMS
A user application that uses engine side cursors in a stored procedure can cause a myriad of problems when available SQL Server system memory becomes full. Usually it starts with error 707:

System error detected during attempt to free memory at address 0x%1x. Please consult the SQL Server error log for more details.

Or Error 706:

Process %d tried to remove PROC_HDR 0x%lx that it does not hold in Pss.

These are then followed by handled access violations. The access violations can also occur without the 707 or 706 errors. After the access violations, SQL Server will often lock up and become unusable as it goes into a 100 percent CPU spin.

MORE INFORMATION
When the SQL Server procedure cache needs to swap out a stored procedure it can incorrectly deallocate the procedure twice. This can cause a memory leak as well as the access violations. This will only occur if the stored procedure is using engine side cursors.

WORKAROUND
Increasing the amount of procedure cache available to SQL Server will reduce the chance of the problem occurring. This can be accomplished by giving SQL Server more memory to use, or by increasing the procedure cache using sp_configure.

FIX: AV Error Using Temp Table and Cursor in Stored Procedure
153987 BUG #: NT: 15510 (6.50)

SYMPTOMS
A stored procedure that performs selects against a cursor with a temporary table in the cursor will fail with a handled access violation or with the following error message:

Msg 208 "Invalidobject name #temptable"

The problem will only occur after SQL Server has been shut down and restarted after the initial creation of the stored procedure. If the stored procedure is then dropped and re-created, it will work again until SQL Server is shut down and restarted.

WORKAROUND
The problem only occurs if both the temporary table and the cursor are both created inside a stored procedure. Creating the temporary table outside of the stored procedure will allow the stored procedure to function correctly. An alternative workaround would be to convert the temporary table to a permanent table.

FIX: Cannot Rename a Column With a Quoted Identifier
153992 BUG #: NT: 15100 (6.50)

SYMPTOMS
If you try to rename a table's column to use a quoted identifier column name, sp_rename gives an error indicating that the column name is invalid. The errors are as folows:

Msg 15006, Level 16, State 1

' ' is not a valid name since it contains invalid characters

Msg 15224, Level 11, State 1

Error, the value for parameter NewName contains invalid characters or

violates a basic restriction

WORKAROUND
Export the data. Drop and re-create the table with quoted identifiers on necessary columns. Import the data back in.

FIX: AV in Update Mode When Script Is Run Twice
154018 BUG#: 14827 (6.50)

SYMPTOMS
This problem arises with simple scripts that use trace flag 323 to determine if UPDATE IN PLACE is occurring. When you run such a script twice, a handled access violation (AV) occurs.

WORKAROUND
To avoid this problem, do not use trace flag 323. Without the trace flag, the script will run successfully.

FIX: SQL Terminates on Delete Table with 15 Self-References
154047 BUG #: NT: 15629 (6.50)

SYMPTOMS
Trying to delete a table that contains 15 self-references can cause the system to return the following errors and warnings:

Error : 1202, Severity: 20, State: 2

Table_lock was called with illegal mode 0.

WARNING: Process being freed while holding Dataserver semaphore

udwritem: Operating system error 6(The handle is invalid.) on device

'C:\MSSQL\DATA\MASTER.DAT' (virtpage 0x00002739).

udread: Operating system error 6(The handle is invalid.) on device

'C:\MSSQL\DATA\MASTER.DAT' (virtpage 0x00000394).

Error: 5157, Severity: 20, State: 0

Recursive error 1202 in ex_print

This action also results in an unhandled exception, essentially terminating the server.

WORKAROUND
The workaround for this problem is to avoid having more than 14 self- references on a table. Establish another table that can contain the additional required references.

MORE INFORMATION
This problem is specific to 15 self-references and hence is extremely uncommon.

NOTE: If a default debugger is set up, it will generate a debug log (like Dr. Watson log).

FIX: SELECT INTO Locking Behavior
153441 BUG #: 14818 (DCR, 6.50)

SUMMARY
In SQL Server 6.5, SELECT INTO wraps within a transaction. Tables created by using SELECT INTO hold to the ACID (atomicity, consistency, isolation, durability) transaction properties. This also means that system resources, such as pages, extents, and locks, are held for the duration of the SELECT INTO statement. With larger system objects, this leads to the condition where many internal tasks can be blocked by other users performing SELECT INTO statements. For example, on high-activity servers, many users running the SQL Enterprise Manager tool to monitor system processes can block on each other, which leads to a condition where the SEM application appears to hang.

MORE INFORMATION
When you upgrade to SQL Server 6.5 Service Pack 1, SELECT INTO locking characteristics is a system settable feature. Wrapping the SELECT INTO with a transaction remains the default behavior. For users wishing not to hold system catalog locks on the activity, a trace flag has been added to allow for such operations. To apply the trace flag, start the server with the -T5302 command line parameter, or from within a query window, use the following commands:

dbcc traceon (3604) go dbcc traceon (5302) go

When the 5302 trace flag is applied and a SELECT INTO fails, the table is still created. Note that the locking behavior you select is applied for all databases within the server.

Additional query words: sp1 database patch service pack sqlfaqtop

Keywords: kbfix KB153096

-

[mailto:TECHNET@MICROSOFT.COM Send feedback to Microsoft]

© Microsoft Corporation. All rights reserved.