I know it is a week late. Let’s just agree that I have been busy with other things. For those that are unaware, Meme Monday is where I assign a writing assignment for those that need to be given a topic to write about in order to help you blog more often. With Meme Monday posts you don’t need to tag others, you don’t need to link back to this post, and you don’t have to wait to be tagged. Just start writing!

Here is your assignment for next week:

SQLCLR: Good? Or Evil?

Give an example of how you have used CLR, or seen CLR used well, or not so well. Tell us why you may have decided to not use CLR at all.

The issue I had with CLR is that when a developer would build a CLR object (say, a stored procedure) and then complain about lousy performance on the database side, we couldn’t see the code that was in the assembly. And we didn’t have access to the source code (because we weren’t developers), so we would have to walk over to the developers cube to look at the .NET code in order to try to decipher what was being done. This made performance tuning and troubleshooting rather difficult, if not impossible.

I still remember being at the MTC in Waltham and the Microsoft folks many years ago telling me about how wonderful CLR was going to be and my asking the simple question “yeah, so how do I help tune code and queries I can’t see?” The look on their faces was priceless. “You could get a refactoring tool…I guess….we don’t know…why would you need to see it? It will execute just fine.”

Yeah. Just fine. Until it doesn’t run just fine. And then you need to fix it like a mechanic performing a tune-up in the dark with both hands behind their back.

Despite all of that I find CLR to be “mostly good”. I try to tell developers to use CLR any time they would be thinking of creating a .DLL for something. If whatever you are doing doesn’t need a .DLL, or an extended stored procedure, then you don’t need to be thinking about creating a CLR.

Last week I posted a puzzle of sorts. I showed you a handful of graphs from PAL and asked if anyone could determine what had happened. Out of all the comments that were left only a handful of people seemed to be on the right track, or showed a methodology in their thinking that should have gotten them to the same conclusion.

The answer to the scenario last week is this: Someone ran a SELECT statement, and that statement needed to read pages from disk into memory, which flushed all the existing pages out.

There was no issue with hardware, there was no magic knob to turn to make performance better, or to have prevented any end user from running such a statement. The answer wasn’t complex, either. Just one user, running a valid query that needed to pull in a lot of data.

Sometimes, as a DBA, all you can do is educate. You are not always able to press a button to make problems go away completely.

Here is another one for you today. Can anyone explain what is happening here, and what actions a DBA would take when presented with these details…or if any action should be taken?

Checkpoint pages/sec

Lazy Writes/sec

Page Writes/sec

Leave your thoughts and answers in the comments, I would love to hear what you believe is happening to my server.