Cannot fetch a row using a bookmark from OLE DB provider (Msg 7333, Level 16, State 2)

Linked Server Msg 7333

If you are getting this error and are trying to resolve it. Let me give you a little background information. A bookmark in this error is a left over from SQL Server 2005 SP1 and earlier. Bookmarks after 2005 SP1 is really a referred to as a key lookup, not to be confused with primary/foreign keys. From now on I am going to refer to bookmarks as key lookups so we are uniform and using the modern terminology.

Key lookups are objects within a query that refer to a data page (where the data is actually stored in the database) from a non-clustered index.

When SQL Server is executing DML (Insert, Update, Delete) against a linked server successfully for days, months, or even years and then fails with this error:

 Cannot fetch a row using a bookmark from OLE DB provider (Msg 7333, Level 16, State 2)

Something in the query execution plan is wrong. So far I have only run into this problem using linked servers and it probably has to do with resource management within SQL Server. Perhaps someone could elaborate on this that has more experience than I. But within the query’s execution plan, SQL Server decides to switch to a different index to run the query. The query then fails because it is using an index that is lacking information to complete the request.

Resolution

Ultimately, you have to do the following:

  1. Look at your where criteria in your query that is dependent on the linked server.
  2. Review indices on the table, that the where criteria is referring to.
  3. Is your one of your criteria covered by more than one index?

If you answered yes to number 3, there is a good chance that the behavior I am describing is actually causing your problem.

To diagnose this, disable one of the indexes  that is in question.

NOTE: Never disable the clustered index, you will prevent data access to the table.

Run the query again, and it should work, if not rebuild the disabled index to bring it back online and then disable the other index. If the query works you are in business. Talk with your database administrator, developer, or vendor of the application to come up with a good solution. But this will temporarily get your query to work.

SQL Server CTE ~ Hooray for Common Table Expressions

What is a SQL Server CTE?

CTE stands for Common Table Expression.

A CTE is basically a subquery. So why is it special? This is why they are good:

  • Separate a subquery from the body of the main query.
  • Cleaner code – CTEs enable you to separate your logically congruent code segments.
  • Can be used for improving the execution plan of your query.
  • Can be used for recursive, hierarchical queries.

When aren’t they good? Almost never. Basically, if you don’t need a subquery don’t use a CTE. Also don’t bother using a CTE when you are performing a query with resources outside of SQL Server, there are negative performance issues in SQL Server 2005, but is fixed in SQL Server 2008. Here is everything you will need to know about CTEs.

CTE Basics

Books Online

Recursive CTE

OracleConnection ~ Connection String Nightmare

I spent quite a bit of time today debugging an oracleconnection object. It would fail everytime the connection string property was set. This is what I would recieve in the error log:

Faulting module name: KERNELBASE.dll, version: 6.1.7601.17651, time stamp: 0x4e2111c0

Exception code: 0xe0434f4d

Fault offset: 0x0000d36f

I didn’t get to far Duck Duck Go’ing (or Googling) on this so I threw the assignment into an try catch block. I got an error  “Connection string is not well formed”. Connectionstrings.com had a great section on Oracle so I copied/pasted it into my code changing the pertinent data. Same issue.

After trying several things I discovered that the connection strings for Oracle cannot have a trailing semi-colon.

Data Source=IDWorks;User Id=IDWorks;Password=idw;

Data Source=IDWorks;User Id=IDWorks;Password=idw

Hope this saved you some time!

SQL Server Distributed Queries!!!

Distributed queries in SQL Server are a blessing and a curse. What can I say, they are fascinating. For those of you who are not familiar with distributed queries, here is the crash course.

What is a distributed query?

So here is the deal, a distributed query is query that relies on data that lives on another data source. The classic example of distributed queries is the Linked Servers in SQL Server. You can also use the OpenQuery and OpenRowset stored procedures to get to the remote data. The difference between OpenQuery and OpenRowset is that the OpenQuery uses linked servers (linked servers store the connection string data); whereas OpenRowset requires the entire connection string to be passed.

For sanity purposes I like the OpenQuery it is cleaner and is more abstract.

Example:

What not to do!

SELECT RTRIM(name), id FROM OPENQUERY(linkedserver, ‘SELECT name,id FROM foo’) WHERE id = 24234

–Much better!

SELECT name, id FROM OPENQUERY(linkedserver, ‘SELECT RTRIM(name),id FROM foo WHERE id = 24234’)

In the above example something might be popping in your head, and you’re right, it is a nested query! The theory that I use often is this: All you are doing is creating a temporary table from one server to use on another. All of the processing inside the OpenQuery occurs on the linked server first and then fills in a temporary table on the SQL Server from whence you are executing the distributed query. SQL Server then takes that temporary table (results of the OpenQuery) and runs the remainder of the query.

Here is the order of execution, prepare to have your mind blown:

  1. OPENQUERY
  2. FROM
  3. WHERE
  4. SELECT

Here is the mind blowing part, this also relates to performance. Let me explain. The more criteria (restricting the number of rows and fields) you impose in step 1, makes all the remaining steps faster. Likewise any data manipulation you do in step 1 will speed up the entire query. If you have more than one OPENQUERY linking to eachother then join them if possible (they have to be from the same data source), see the example below. Try to get to one OPENQUERY. Remember what I said from earlier, it downloads all that data and puts it into temporary tables. You could take 5 Mb of data transferred from a couple OPENQUERY’s and combine them to be under 1 Mb, this has a huge difference in network traffic and computing time on the SQL Server.

–This is really bad.

SELECT * FROM OPENQUERY(linkedserver1,’SELECT name, id from foo’) AS foo INNER JOIN OPENQUERY(linkedserver1,’SELECT name2, id from goo’) AS goo

ON foo.id = goo.id

–This is much better.

SELECT * FROM OPENQUERY(linkedserver1,’SELECT name, name2 FROM foo INNER JOIN goo ON foo.id = goo.id’)

Gotcha’s with Eloquence and the SQL/R ODBC

Only inner joins are accepted in the very old style: FROM foo, goo WHERE foo.id = goo.id

You can only link 5 tables per OPENQUERY

If you have any questions, just post a comment.