Transaction Isolation Level

#11

"dhek" <dhek (AT) dhek (DOT) dk> wrote

Quote:

ok, let me try to be more clear.

The thing is I have a database containing various types of information -
for instance orders.
Instead of letting each record in an order contain a timestamp indicating
the datatime the record was replicated to another system, I have a table
containing replication times for various types of objects.

Why not add such a column with a timestamp then?

Quote:

This means that while I initiate a transaction selecting all new orders it
is possible for others to add other new orders to the database which I
will never be aware of.
- in my transaction after having gotten new orders I set a timestamp
indicating the datetime of my last replication. If someone has added new
records meanwhile I executed my select statement and performed the update
statement, then these new records will never be replicated because their
timestamp will be less than than the one I set in my update-statement when
finished.

If you're simply looking to do replication, use SQL Server's built in
replication functionality.

It may save you a lot of effort.

Quote:

Hope this clearfies my issue.

"dhek" <dhek (AT) dhek (DOT) dk> wrote in message
news:472c5c63$0$90274$14726298 (AT) news (DOT) sunsite.dk...
Well, my issue is that I'm reading data from tables A, B, C and D and
updates table E.

What I need to prevent is 2 things:

1) New rows must not be added to either table A and B while my
transaction is executing
2) Existing rows must not be modified

My query is executed on a SQL server 2005. I do not have the option to
change the configuration of it.

Am I supposed to use table locks in my query

SELECT A.c1, B.c1, C.c1, D.d1
FROM ...
WHERE ...
WITH (TABLOCK, UPDLOCK, HOLDLOCK)

UPDATE E.1
SET ...
WHERE ...

"Erland Sommarskog" <esquel (AT) sommarskog (DOT) se> wrote in message
news:Xns99DD791FBE224Yazorman (AT) 127 (DOT) 0.0.1...
dhek (dhek (AT) dhek (DOT) dk) writes:
I have 1 SQL statement selecting data from various tables and updating
other tables.

The question then is how do I prevent other applications from modifying
the tables that I'm working on (that is while my transaction is being
executed)?

I know that the isolation level should be either REPEATABLE READ or
SERIALIZABLE. But I need confirmation on if one of these actually solve
my
issue - prevents other applications/threads from modifying/inserting
data
into the same tables that I'm working on.

It's difficult to give a single answer, since I don't know your exact
requirements, so I have to answer in genric terms.

If you want a consistent snapshot of how the data looks in this precise
moment, the isolation level you should use is snapshot isolation.
Snapshot
isolation is available only in SQL 2005 and later. Furthermore the
database
must be configured to permit snapshot isolation. When you have snapshot
is
created when the transaction starts, or at latest when you start to read
data. If data is updated while your query runs, you will not see these
updates. This gives you a consistent view - but it may also give you
outdated data, depending on how you look at it.

On SQL 2000, snapshot isolation is not available, and the only foolproof
way to get consistent data, is to set the database in single-user mode.

In the default isolation level, READ COMMITTED, if you read the same
row twice, you may get different results in different accesses. For
instance, if you run:

SELECT O.OrderID, E.EmployeeID, E.LastName
FROM Orders O
JOIN Employees E ON O.EmployeeID = E.EmployeeID

You may see different last names for the same employee ID if the query
plan uses a loop join, and the last name is updated while the query is
running.

In the next level, REPEATABLE READ locks are held, and you are
guaranteed
that reading the same row twice will yield the same result. However, if
the last name of employee 8 was Grønkjær when the query started, and
updated to Gravesen before you have read any orders with employee 8,
you would see Gravesen in the result set.

SERIALIZABLE adds protection against "phantom insert", so if you read
the
same range twice, you will get the same result. That is, if you run
SELECT MAX(OrderID) FROM Orders twice in the same serializable
transaction,
you will get the same result. But if a order is added after you started
the
transaction, but before your query runs, the order will show up.

--
Erland Sommarskog, SQL Server MVP, esquel (AT) sommarskog (DOT) se

Books Online for SQL Server 2005 at
http://www.microsoft.com/technet/pro...ads/books.mspx
Books Online for SQL Server 2000 at
http://www.microsoft.com/sql/prodinf...ons/books.mspx

--
Greg Moore
SQL Server DBA Consulting Remote and Onsite available!
Email: sql (at) greenms.com http://www.greenms.com/sqlserver.html

#12

Quote:

The thing is I have a database containing various types of information -
for instance orders.
Instead of letting each record in an order contain a timestamp indicating
the datatime the record was replicated to another system, I have a table
containing replication times for various types of objects.

Why not add such a column with a timestamp then?

Yes, I'd love to - but its not just up to me, since I don't have ownership
of the database.

Quote:

This means that while I initiate a transaction selecting all new orders
it is possible for others to add other new orders to the database which I
will never be aware of.
- in my transaction after having gotten new orders I set a timestamp
indicating the datetime of my last replication. If someone has added new
records meanwhile I executed my select statement and performed the update
statement, then these new records will never be replicated because their
timestamp will be less than than the one I set in my update-statement
when finished.

If you're simply looking to do replication, use SQL Server's built in
replication functionality.

It may save you a lot of effort.

Well, I'm not sure what the possibilities are really with regards to
replication from SQL server. I doubt that it is usefull in my case, since I
pull data from an SAP system.

#13

"dhek" <dhek (AT) dhek (DOT) dk> wrote

Quote:

The thing is I have a database containing various types of information -
for instance orders.
Instead of letting each record in an order contain a timestamp
indicating the datatime the record was replicated to another system, I
have a table containing replication times for various types of objects.

Why not add such a column with a timestamp then?

Yes, I'd love to - but its not just up to me, since I don't have ownership
of the database.

I'm confused then. If you don't have ownership of the DB, how can you
expect to set transaction levels and locking?

Quote:

Well, I'm not sure what the possibilities are really with regards to
replication from SQL server. I doubt that it is usefull in my case, since
I pull data from an SAP system.

Then I'm doubly confused.

Quote:

#14

dhek (dhek (AT) dhek (DOT) dk) writes:

Quote:

The thing is I have a database containing various types of information -
for instance orders.
Instead of letting each record in an order contain a timestamp indicating
the datatime the record was replicated to another system, I have a table
containing replication times for various types of objects.

This means that while I initiate a transaction selecting all new orders
it is possible for others to add other new orders to the database which
I will never be aware of. - in my transaction after having gotten new
orders I set a timestamp indicating the datetime of my last replication.

Just to make things clear: is this the timestamp data type, or a datetime
value? (It sounds like the latter, but I want to be sure.)

From this description, it sounds that SERIALIZABLE would do the job, but
I think there is a better solution. Do this:

SELECT @mynewtimestamp = getdate()

SELECT ... FROM tbl
WHERE regdate BETWEEN @myoldtimestamp AND @mytimestaamp

By first determining the upper limit of the span you will read,
there is no problem with rows that are added later, because they will
be outside your range.

I would strongly recommend against using serializable, because if
conflicting updates really is a seroius possibility, using serializable
is likely to wreak havoc with the concurrency of the system. Your
replication may work, but if people can place orders while your
replication is running, that can be a serious problem.

An interesting observation is that as I understand you scenario is
that snapshot isolation would *not* work, despite what I said before.
With snapshot you would miss rows that had been inserted when you started
reading, but which had not been committed.

This shows just how important it is to give a completely picture of the
problem you are trying to solve.

--
Erland Sommarskog, SQL Server MVP, esquel (AT) sommarskog (DOT) se

Books Online for SQL Server 2005 at
http://www.microsoft.com/technet/pro...ads/books.mspx
Books Online for SQL Server 2000 at
http://www.microsoft.com/sql/prodinf...ons/books.mspx

#15

Quote:

Just to make things clear: is this the timestamp data type, or a datetime
value? (It sounds like the latter, but I want to be sure.)

this is a datetime value obtained by the function GETUTCTIME()

Quote:

From this description, it sounds that SERIALIZABLE would do the job, but
I think there is a better solution. Do this:

SELECT @mynewtimestamp = getdate()

SELECT ... FROM tbl
WHERE regdate BETWEEN @myoldtimestamp AND @mytimestaamp

By first determining the upper limit of the span you will read,
there is no problem with rows that are added later, because they will
be outside your range.

I most definitely agree with u - its so simple and so right. I does however
still leave a small time window for the problem to still exist, but I guess
we're now dealing with a more theoretical issue than one that is likely to
actually occur.
In any case - this model should be used:-)

Quote:

I would strongly recommend against using serializable, because if
conflicting updates really is a seroius possibility, using serializable
is likely to wreak havoc with the concurrency of the system. Your
replication may work, but if people can place orders while your
replication is running, that can be a serious problem.

I agree - this is also my fear and is exactly why I didn't really felt too
happy about this solution. But I do believe it would also solve my issue,
since it would disallow for new data to be added that conflict with the
dataset at hand (my where-clause in the SELECT-statement part).

Quote:

An interesting observation is that as I understand you scenario is
that snapshot isolation would *not* work, despite what I said before.
With snapshot you would miss rows that had been inserted when you started
reading, but which had not been committed.

This shows just how important it is to give a completely picture of the
problem you are trying to solve.

hehe, point taken:-)

#16

Quote:

Why not add such a column with a timestamp then?

Yes, I'd love to - but its not just up to me, since I don't have
ownership of the database.

I'm confused then. If you don't have ownership of the DB, how can you
expect to set transaction levels and locking?

I cannot modify the configuration of the database - this also includes
tables and their design. I can however change my statements all I want and
for instance do

SET TRANSACTION ISOLATION LEVEL SERIALIZABLE
GO

BEGIN TRANSACTION trans1
SELECT ...
FROM ...
WITH (TABLOCK, HOLDLOCK, UPDLOCK)

UPDATE ...
SET ...
WHERE ...
COMMIT TRANSACTION trans1

which affects transactions and locking in the database.

Quote:

Well, I'm not sure what the possibilities are really with regards to
replication from SQL server. I doubt that it is usefull in my case, since
I pull data from an SAP system.

Then I'm doubly confused.

Now me too:-)

Thread Tools
Show Printable Version Email this Page
Display Modes
Linear Mode Switch to Hybrid Mode Switch to Threaded Mode

Transaction Isolation Level

comp.databases.ms-sqlserver comp.databases.ms-sqlserver

Re: Transaction Isolation Level - 11-04-2007 , 07:11 AM

Re: Transaction Isolation Level - 11-04-2007 , 08:11 AM

Re: Transaction Isolation Level - 11-04-2007 , 09:50 AM

Re: Transaction Isolation Level - 11-04-2007 , 11:38 AM

Re: Transaction Isolation Level - 11-04-2007 , 01:55 PM

Re: Transaction Isolation Level - 11-04-2007 , 02:02 PM