I place a lot of emphasis on capturing actual execution plans because of the runtime metrics, but with Trace Flag 7412, we don't need the plan. This is great news, because capturing execution plans, even using extended events, is an expensive proposition. However, using either the query_thread_profile event, or, Trace Flag 7412, we can get the runtime metrics without the plan. Trace Flag 7412 Here's how it works.You can either be running the extended event, query_thread_profile (a debug event, but one documented and supported by Microsoft) or, enable the Trace Flag 7412. I like to use the extended event in a targeted fashion to easily see behaviors on a query without having to capture the plan. You can even capture the information and then combine it with an estimated plan…
One of the ways that you take more direct control over your SQL Server instances is through the use of trace flags. There are a number that people recommend you enable by default. Prior to Extended Events for example, I'd say you should turn on trace flag 1222 in order to capture deadlock information on your server (now I just recommend you use the system_health session). I absolutely think you should turn on trace flag 2371 to get better behavior out of your automated statistics updates. There are others that I'll leave to all the systems experts to advise you on. What about Azure SQL Database? I doubt you'll be shocked, but if I try this: DBCC TRACEON (2371,-1); I get the following error: Msg 2571, Level 14, State 3,…
One of my favorite little bits of information about Extended Events is the fact that everyone running a full instance of SQL Server has deadlock information available to them, even if they never enabled Trace Flag 1222 through the system_health session. That captures the xml_deadlock_report which has the full deadlock graph. However, what if you want to capture deadlock info, but, you're dealing the GDPR, and transmitting query values could be problematic? Enter xml_deadlock_report_filtered. xml_deadlock_report_filtered If you do a search for this event, you're not going to find much. Doesn't seem like anyone, including Microsoft, has bothered to document it. This is not going to be a comprehensive definition for all things xml_deadlock_report_filtered. However, I can show you why you might want to use it. This is a port of…
What's the story with AWS RDS and SQL Server deadlocks? I'm approaching AWS RDS like I was taking on a new role at a new organization. Do we have backups in place? Yes, great. Can I test them? Yes. Do they meet our RTO & RPO? Yes. Moving on. What have we got for monitoring? AWS RDS has a good percentage of the fundamentals. Now, it's laid out a little oddly. You have the stuff going through CloudWatch which is largely OS oriented. Then you have enhanced monitoring, which you have to turn on, which covers eight key metrics for SQL Server. Finally, you can enable Performance Insights which gives you metrics on query behaviors (and yeah, any or all of these may be the subject of upcoming blog posts).…
The general idea for this question came from dba.stackexchange.com: could we, and if we can, how, get row counts after execution. I was intrigued with the idea, so I ran some tests and did a little digging. I boiled it all down in the answer at the link, but I figured I could share a little here as well. Properly Retrieve Row Counts After Execution The right way to do this is obvious and simple. Before you need it, set up an Extended Events session. Done. The only question is what goes into the Session. First blush, sql_batch_completed and/or rpc_completed. Both will return a rows affected value. Although, interestingly, the row_count value is documented as rows returned. However, it's both. But, if you really want to get picky, batches and…
I've always felt responsible for making such a big deal about the differences between estimated and actual plans. I implied in the first edition of the execution plans book (get the new, vastly improved, 3rd edition in digital form for free here, or you can pay for the print version) that these things were so radically different that the estimated plan was useless. This is false. All plans are estimated plans. However, actual plans have some added runtime metrics. It's not that we're going to get a completely different execution plan when we look at an actual plan, it's just going to have those very valuable runtime metrics. The problem with getting those metrics is, you have to execute the query. However, this is no longer true in SQL Server…
I've posted a number of new videos to the Youtube channel that might be of interest if you're a data professional. First up, I've started migrating my Database Fundamentals posts over to Youtube as SQL Server Fundamentals. The first one is available: [embedyt] https://www.youtube.com/watch?v=NHkzj9ZRhbk[/embedyt] Speaking of DBAs, the job is not going away, but it is changing. Find out how to survive the change: [embedyt] https://www.youtube.com/watch?v=loUwmKQ_Eg4[/embedyt] Would you like to know how the Query Store works? Here's a getting started video: [embedyt] https://www.youtube.com/watch?v=XK8rBO9R43c[/embedyt] I also have a good video on how to combine capturing query metrics along with wait statistics using Extended Events: [embedyt]https://www.youtube.com/watch?v=nYyTSxry03A&t=97s[/embedyt] Wait until you see what Trace Flag 7412 can do for you. It's very cool. [embedyt]https://www.youtube.com/watch?v=ei2zJwZyRic&t=79s[/embedyt] Finally, if you are interested in attending one of my…
Query tuning ain't easy. Figuring out which index is getting used is one step, and generally simple, look at the execution plan to see which index is in use and whether it's being used in a SEEK or a SCAN. Done. However, when your index isn't being used, how do you tell how or why something else is being done? Well, that's largely down to row counts which brings us to statistics. Which Statistics are Used Years ago I was of the opinion that it wasn't really possible to see the statistics used in the generation of a query plan. If you read the comments here, I was corrected of that notion. However, I've never been a fan of using undocumented trace flags. Yeah, super heroes like Fabiano Amorim and…
Based on the number of times I see this question on forums, it must be occurring all the time. You have two different servers that, as far as you know, are identical in terms of their options and setup (although not necessarily in terms of power, think a test or pre-production system versus production). On these servers you have a database on each that, as far as you know, is the same as the other in terms of options, objects, maybe even data (although, this does mean that you have unmasked production information in your QA environment, which potentially means you're going to jail, might want to address this, especially now that I've told you about it, mens rea, you're welcome). On each database you run, as far as you know, the exact same query (whether…
How do you define an expert? My personal definition: An expert is the person that is a chapter ahead of you in the book. Why am I talking about this? Just that I’m feeling more stupid than usual lately. In the last two weeks I’ve had people bring up through various discussions, documents, what have you, four different SQL Server trace flags that will affect how SQL Server builds execution plans and I’ve never heard of any of them before. I’ve never, ever, thought of myself as an expert in execution plans, despite having written a book about them. I just thought I had a good grasp on how they worked and I was willing to share. I didn’t know everything and never pretended to, but I thought I knew…