Exclusives Locks With Transaction Isolation Level Readuncommitted
Mar 12, 2008
Hello,
I have some locks issues on production database (win 2k3 SP1, sql server 2k5).
In fact, I have an asynchronous process that makes SELECT TOP 1 in a table and UPDATE the selected row. The transaction isolation level for doing this action is READUNCOMMITTED.
The isolation level readuncommitted is ignored for the update if I'm not wrong.
On the other hand, I have some transactional activities with the isolation level read uncommitted too.
But when I control the database activity, I find very often locks between the asynchronous part and the transactional part. This is the transactional activity that is locking the asynchronous activity.
The transactional activity is a simple SELECT and this type of query, in spite of the isolation level readuncommitted, makes exclusives locks when the asynchronous makes LCK_M_U.
I tried to modify the strored procedure for the SELECT/UPDATE of the asynchronous process with a "UPDATE my_table ... FROM my_table" query in order to reduce the transaction time. But the problem is always present.
Can someone help me to understand how a select query with the isolation level readuncommtted can make exclusives locks ?
Have the need for going to a table to get an identity value. This is for updating an existing database, blah blah blah. Here is the schema of the table we are using:CREATE TABLE [TableIdentityValue] ( [TableName] [varchar] (50) , [NextNegativeIdentity] [int] NOT NULL , [NextPositiveIdentity] [int] NOT NULL , CONSTRAINT [PK_TableIdentityValue] PRIMARY KEY CLUSTERED ( [TableName] ) ON [PRIMARY] ) ON [PRIMARY]GO Now, depending on the type of data we are inserting into a table, we need to get either a negative or positive number for the PK. There are two sprocs that control the obtaining of those values:CREATE PROCEDURE GetNegativeIdentity @tableName varchar(50)AS DECLARE @nextNegativeIdentityValue int BEGIN TRANSACTION SET TRANSACTION ISOLATION LEVEL SERIALIZABLE SET @nextNegativeIdentityValue = ( SELECT NextNegativeIdentity FROM TableIdentityValue WITH (ROWLOCK) WHERE TableName = @tableName ) UPDATE TableIdentityValue SET NextNegativeIdentity = @nextNegativeIdentityValue - 1 WHERE TableName = @tableName COMMIT TRANSACTION RETURN @nextNegativeIdentityValueGOCREATE PROCEDURE GetPositiveIdentity @tableName varchar(50)AS DECLARE @nextPositiveIdentityValue int BEGIN TRANSACTION SET TRANSACTION ISOLATION LEVEL SERIALIZABLE SET @nextPositiveIdentityValue = ( SELECT NextPositiveIdentity FROM TableIdentityValue WHERE TableName = @tableName ) UPDATE TableIdentityValue SET NextPositiveIdentity = @nextPositiveIdentityValue + 1 WHERE TableName = @tableName COMMIT TRANSACTION RETURN @nextPositiveIdentityValueGOSo, the thing is, we need the read and update of the value from the specific TableIdentityValue row to be atomic - we don't want anyone else reading or modifying that data. The problem is knowing which level of isolation to use and/or locking, and how to implement that. I have tried a few different things that seemed to make sense, like placing a ROWLOCK on the SELECT statement, but is that lock going to hold for the entire length of the transaction? Also, I read that using some of the lock hints can be accomplished in the sense that some isolation levels are the same as some lock hints (e.g. setting isolation level to SERIALIZABLE "has the same effect as setting HOLDLOCK on all tables in all SELECT statements in a transaction" according to SQL Books Online.Any help is appreciated!
What is the TRANSACTION ISOLATION LEVEL settings for MSSQL like the default setting in Oracle. In Oracle the default setting allows one session to read consistent data without waiting for the other sessions to commit/rollback the data.
For eg: In Mssql, if I update table A in the first session, and in another session (second session) if I select from table A, the second session waits till the first session completes the updates and commit or rollbacks.
But in Oracle , if I update table A in the first session, and in another session (second session) if I select from table A, the second session will perform a read from the ROLLBACK SEGS and give a read consistent data without waiting for the first session to commit or rollback the transaction.
Is this type of behaviour is possible is MSSQL. And If YES how can I do it?
Not sure if this is more a .Net question or SQL Server, but I think it belongs here.
I have a small .Net app that reads records from a bunch of files from disk and inserts them into a database table. There could be several hundred files resulting in 100,000 records or more each time its run. Since it's a large table there are of course a few indexes on it so the insert takes a while. For larger sessions it could run as long as an hour. I need it to run in a transaction so that if anything happens while it's running the records from that run were committed on an all or nothing basis. However, I don't want to lock the table at all while the insert is happening. These aren't transaction records or anything like that, and the batches are separated by client so there will be no conflicts (no need to lock the table).
Unfortunately, no matter what I use for the isolation level of the transaction the table always ends up locked for reads. Data from previous runs is live at this point and we can't allow that. I have the choice of the following isolation levels when I create the transaction, but none seems to work: Chaos ReadCommitted ReadUncommitted RepeatableRead Serializable Snapshot Unspecified
I would expect Chaos, ReadUncommitted, or Snapshot be okay here, but I can't seem to get it working. Any thoughts?
Hi,I have 1 SQL statement selecting data from various tables and updating othertables.The question then is how do I prevent other applications from modifying thetables that I'm working on (that is while my transaction is being executed)?I know that the isolation level should be either REPEATABLE READ orSERIALIZABLE. But I need confirmation on if one of these actually solve myissue - prevents other applications/threads from modifying/inserting datainto the same tables that I'm working on.Thanks in advance,Daniel
I need to set the Isolation Level (in ADO) for the Non-transaction queries to SNAPSHOT.
Both the ADO.Connection.IsolationLevel Property and the SQL Server SET TRANSACTION ISOLATION LEVEL command set the Isolation Level for the Transaction queries but no for the non-transaction queries.
I cannot use the READ_COMMITTED_SNAPSHOT database option, becaus when I am in a transaction I need the READ COMMITTED Isolation Level not the SNAPSHOT Isolation Level.
I don't want to rewrite the entire code of my existing application to add (NOLOCK).
I have a question about the "readCommitted" transaction isolation level.I have a client that is updating a record on a table.I suspend the execution after the UPDATE but before the commit statement.Than another client is trying to read the same record.As transaction isolation is set to "readCommited" I expected that the secondclient will read the old version of the record (before the update).Instead, the second client hangs and wait until the first client do thecommit.I expect this behavior if transaction isolation is set to "serializable"Is this behavior correct?Thanks,D.
Is there a way to read data from a linked server,within a transaction, without using DTC?The data on the linked server is static, thereforethere is no need for two-phase commit. There isno need for locking data on the linked server, becauseit is not being updated (either from the remote server,or from the local server).I don't want to run DTC because:1.) there have been security-related flaws with DTC inthe past2.) the application doesn't do distributed updates, andbecause the data on on the remote server is static,there is really no data integrity exposure withoutDTC.I cannot specify "WITH (NOLOCK)" on the select fromthe linked server:Server: Msg 7377, Level 16, State 1, Line 6Cannot specify an index or locking hint for a remote data source.I tried setting the isolation level:SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTEDbut that seems to have no effect on the requirement to use DTC.I still get the message:MSDTC on server 'LOCALSERVER' is unavailable.Is there some other way around this? Is it possible to provide someconnection string parameter, in the linked server setup, that wouldspecify the "READ UNCOMMITTED" isolation level for the linked server,so that DTC wouldn't be necessary.(In other words, can I tell SQL Server, "trust me, this won't hurt"?)Environment: SQL Server 2000 sp4The SQL does something like:declare @x char(4), @k int, @rc1 int, @rc2 intset @k=123BEGIN TRANselect @x=xfrom remoteserver.remotedatabase.dbo.tablewhere k=@kupdate localdatabase.dbo.table1set x=@xwhere k=@kset @rc1=@@errorupdate localdatabase.dbo.table2set x=@xwhere k=@kset @rc2=@@errorif (@rc1 = 0 AND @rc2=0) COMMIT TRANelse ROLLBACK TRAN
Hi,we are executing the following query in a stored procedure using snapshot isolation level:DELETE FROM tBackgroundProcessProgressReportFROM tBackgroundProcessProgressReport LEFT OUTER JOIN tBackgroundProcess ON tBackgroundProcess.BackgroundProcessProgressReportID = tBackgroundProcessProgressReport.BackgroundProcessProgressReportID LEFT OUTER JOIN tBackgroundProcessProgressReportItem ON tBackgroundProcessProgressReport.BackgroundProcessProgressReportID = tBackgroundProcessProgressReportItem.BackgroundProcessProgressReportIDWHERE (tBackgroundProcess.BackgroundProcessID IS NULL) AND (tBackgroundProcessProgressReportItem.BackgroundProcessProgressReportItemID IS NULL)The query should delete records from tBackgroundProcessProgressReport which are not connected with the other two tables.However, for some reasone we get the following exception:System.Data.SqlClient.SqlException: Snapshot isolation transaction aborted due to update conflict. You cannot use snapshot isolation to access table 'dbo.tBackgroundProcess' directly or indirectly in database 'RHSS_PRD_PT_Engine' to update, delete, or insert the row that has been modified or deleted by another transaction. Retry the transaction or change the isolation level for the update/delete statement.The exception specifies that we are not allowed to update/delete/insert records in tBackgroundProcess, but the query indeed deletes records from tBackgroundProcessProgressReport, not from the table in the exception.Is the exception raised because of the join?Has someone encountered this issue before?Thanks,Yani
I'm investigating a poorly performing procedure that I have never seen before. The procedure sets the transaction isolation level, and I suspect it might be doing so incorrectly, but I can't be sure. I'm pasting a bastardized version of the proc below, with all the names changed and the SQL mucked up enough to get through the corporate web filters.
The transaction isolation level is set, but there is no explicit transaction. Am I right that there are two implicit transactions in this procedure and each of them uses snapshot isolation?
SET NOCOUNT ON; SET TRANSACTION ISOLATION LEVEL SNAPSHOT; DECLARE @l_some_type varchar(20), @some_type_code varchar(3), @error int, @error_msg varchar(50);
We have a service that inserts some rows into a parent table (P) and child table (C). This operation is atomic and performed within a transaction.
We also have a service that queries these tables such that rows are (should only be) returned from P where there are no children for that parent.
The SQL that performs this is simplified below:
SELECT P.SomeCol FROM P LEFT OUTER JOIN C ON P.PKofP_Value = C.PkofP_Value WHERE C.PkofPValue IS NULL AND P.SomeOtherCol=0
Our expectation is that the query service should only return rows from P where there are no rows in C.
However, this seems not to be the case, and occasionally we find that rows from P are returned where there are matching rows in C.
We are sure that the process that inserts rows into P and C does so within a single transaction.
We have traced this with SQLTrace and can see the txn stag and committing and all operations using the same transactionid within the transaction.
We are running the default isolation level committed.
In SQLTrace we can see the query process start, the inserter process start and complete and then the query process continue (after presumably being blocked).
So how can the query process "miss" the child rows and return the parent from the above query?
Is it possible that, in this isolation level, the inserter process can block the query process such that when the inserter process commits and when the query process continues it does not see the child rows inserted because they were inserted in the table/index "behind" where the query process has already read - some kind of phantom phenomenon?
I am trying to get my head around locking (row, table) and Isolation Levels. We have written a large .NET/SQL application and one day last week we had about two dozen people in our company do some semi "stress/load" testing of the app.
On quite a few occassions, a few of the users would receive the following error:
"Transaction (Process ID xx) was deadlocked on lock resources with another process and has been chosen as the deadlock victim. Rerun the transaction."
We are handling this on two fronts, the app and the database. The error handling in the app is being modified to capture this specific error and to retry the transaction.
However, from the database side, I am trying to find the most affective and efficient change to make regarding locking. I have been doing a lot of reading online and in BOL to get a better grasp of locking, but what I would really like is feedback from the community (forum) and get your thoughts on what changes I should make, if any, on the db side.
lets say user1 is reading row1, then user2 reads and updates row1, when user1 is about to update row1 i want him to be informed that his copy of row1 have been updated, so he has now the options to either get the new version of row1 or cancel his update process.
Hi, folks. Please guide. I have a VB application that is used for production and reporting. I 've been having alerts for deadlocks that popup after every 2 or 3 minutes. I am planning to seperate reporing server by using transactional replication from production server to the reporting server. However some reports update and insert data so i need reporting server to be enabled for DML.
Is there any option on the server-level where i can force each user to operate in READ-UNCOMMITTED mode instead of specifying WITH (NOLOCK) in the queries of my application. Dirty reads won't bother me in current situation, i guess the propotion of fast reads would be a better trade-off. New to SQL, Thanx for helping!!
Hi all, can anyone give me more information on set transaction isolation level serializable ?? I want to prove some lock to use on online insert and update. Thank you every much.
I have an issue in one of my stored procs. I set the Isolation level to read uncommitted at the beginning of the proc and then I try to reset this isolation level back to read committed. When reset the isolation level, I get and error. has anyone encountered this before?
lets say user1 is reading row1, then user2 reads and updates row1, when user1 is about to update row1 i want him to be informed that his copy of row1 have been updated, so he has now the options to either get the new version of row1 or cancel his update process or continue his update
by the way, im using typed dataset on my data access layer.
Is there a way to define Connection Manager with Read Uncommited isolation level? I do not want to specify (nolock) in all my commands and instead want to give a generic defenition at the Connection level.
I would like to be able to alter the default isolation level at connection time via the ADO connection string. Can this be done?
Why? I have various reporting applications (Crystal etc.) that queries against MS SQL server using ADO (SQLOLEDB). I would like to be able to alter the isolation level for these queries to readuncommitted. But many of the reporting applications does not have this option and they autogenerate the SQL making it impossible to use the use the WITH(table_hints) clause in the SELECT statement. So if I could set the isolation level in the connection string this could be a workaround.
To all SQL gurus:I have a Windows Service that uses a single SQL Server table toretrieve items of work. Each thread of the service checks this tablefor the earliest item of work that is not already in process, marksthat item as in process, then begins to work the item. My concern iswhether the threads will begin to step on each other's toes by pickingthe same item of work at the same time. To prevent this, I use thefollowing SQL table:[WorkItems]WorkItem varchar(512)DateSubmitted datetimeStatus intIn requesting the next work item, I use the following SQL syntax:DECLARE @workitem varchar(512)SET TRANSACTION ISOLATION LEVEL SERIALIZABLEBEGIN TRANSACTIONSELECT TOP 1 @workitem=WorkItemName FROM WorkItems WHERE Status=1ORDER BY DateSubmittedUPDATE WorkItems SET Status=2 WHERE WorkItemName=@workitemSELECT * FROM WorkItems WHERE WorkItemName=@workitemCOMMIT TRANSACTIONThe idea is that the Transaction Isolation Level, along with the threestatements in the transaction block, will only let one thread at atime request the next work item. The three statements in thetransaction block select the next work item, mark it as in process,then return the work item to the calling thread. In limited testing,all seems well. Before going into production, however, I would like tosee if anyone can confirm that my ideas will indeed prevent threadsfrom duplicating each other's work.Will the above SQL syntax allow me to run multiple threads all lookingto the same database table for work, but prevent them from selectingany of the same work at the same time? If you need more information,please ask.Reply to newsgroup, or directly at Join Bytes!.Matthew Roberts
Is there a way to alter the isolation level of a data source? I have queries with nolock hints that I would like to remove in favor of "set isolation level read uncommitted" but this is not allowed in the report builder. I cannot move to SPs as I'm working in a third-party apps database. I guess I'm wishing for the SSIS style of isolation setting where it's a member of the properties.
Hello all. I'm a litle confused about what's best to use, either isolation levels or locking per table. Cause there are some queries in the stored procedures where I don't need locking i.e. when I check the status of client, but other queries where I do need locking like when I check the existence of a product.
What's best to use, can I combine both? Could you explain it thecnically?
it it possible to to set for appropriate linked server fixed isolation level. Somewhere in linked server settings?
Say, if I want for every query from server SQLServer_A to linked server SQLServer_B to run with isolation level read uncommited.
It's clear, that I can state in very procedure, that uses SQLServer_B "set transaction isolation level read uncommitted". But it's not a way out, as I have thousand stored procedures writted a long time ago.
For the time being SQLServer_A is 2000 SP4 version and SQLServer_B (linked) 2005 SP2a version.
May be it is possible for both servers of 2005 version only?
Are there really any benefit on using Read Uncommitted Isolation Level or having a NOLOCK hints for retrieve queries when the default Isolation level just Read Committed (not using COM+). I'm confused why the Community Server uses this technique perhaps for perf issues but I couldn't see any reason why...
We are using sql 2008r2 standard edition.One of our Production database is using default isolation Readcommitted.The transactions also using read committed. But we want change isolation level to read comitted snapshot isolation and test it to avoid deadlocks.
Is it possible to set in the transaction level for some queries or do we need to change entire database isolation level by using alter database "ALTER DATABASE AdventureWorks2008R2 SET READ_COMMITTED_SNAPSHOT ON"
