Tag Archives: Tuning

Practical SQL Server tuning written in a format that you dont have to be a tuning expert to understand.

Cleaning up the Buffer Pool to Increase PLE

by

Chances are you have extra information in the buffer pool for a bad query and it’s dragging down your PLE, causing SQL Server to run slower because it’s reading more from disk. Although this approach is taking a 180 from my post Fixing Page Life Expectancy it has the same effect, with that post focusing on fixing your worst queries and this one focused on fixing your most misused indexes.  One approach doesn’t replace the other, it’s more like burning the candle at both ends, except that you end up with a better running database instead of getting burnt out.

With this approach we start with what’s in the cache.  You’ll see the same types of issues in almost any database, and this just happens to be a production database I’m going through today.

ScreenHunter_01 2014-01-06 14.52.28

The results were found with this query:

SELECT count(1)/128 AS cached_MB 
    , name 
    , index_id 
FROM sys.dm_os_buffer_descriptors AS bd with (NOLOCK) 
    INNER JOIN 
    (
        SELECT name = OBJECT_SCHEMA_NAME(object_id) + '.' + object_name(object_id)
            --name = 'dbo.' + cast(object_id as varchar(100))
            , index_id 
            , allocation_unit_id
        FROM sys.allocation_units AS au with (NOLOCK)
            INNER JOIN sys.partitions AS p with (NOLOCK) 
                ON au.container_id = p.hobt_id 
                    AND (au.type = 1 OR au.type = 3)
        UNION ALL
        SELECT name = OBJECT_SCHEMA_NAME(object_id) + '.' + object_name(object_id) 
            --name = 'dbo.' + cast(object_id as varchar(100))   
            , index_id
            , allocation_unit_id
        FROM sys.allocation_units AS au with (NOLOCK)
            INNER JOIN sys.partitions AS p with (NOLOCK)
                ON au.container_id = p.partition_id 
                    AND au.type = 2
    ) AS obj 
        ON bd.allocation_unit_id = obj.allocation_unit_id
WHERE database_id = db_id()
GROUP BY name, index_id 
HAVING Count(*) > 128
ORDER BY 1 DESC;

Service Broker Errors Table

First, there is an errors table here with Service_Broker_Errors, and that should never be in the top 10. What are we doing wrong? The index ID of 1 tells me right away that this is a clustered index, the fact that the table is at least 1.5 GB tells me that we’re probably not purging old data, and 1.5 GB in memory on this table is hinting that we’re probably doing a clustered index scan.

I’ll start by looking at the proc cache to see what’s going on. There’s only one execution plan that used that index and it is, in fact, doing a clustered index scan as I expected.

ScreenHunter_01 2014-01-06 15.24.05

Predicates or Seek Predicates

In an execution plan you’ll have a seek predicate and just a plain old predicate. The seek predicate is what you were able to do taking advantage of the index being in order, and the predicate is what you had to scan for.

ScreenHunter_01 2014-01-06 15.26.02

This case is easy because we’re only searching by a single column, but others could have both a seek predicate and a predicate. For instance, if I had an index on my customers table with the key columns Active, First_Name, Last_Name then searched where Active = 1 and Last_Name = ‘Hood’ then it will show up as an index seek with a seek predicate of Active = 1 and a predicate of Last_Name = ‘Hood’. Anyways, lets get back on topic with the issue of my Service_Broker_Errors table…

Now this sounds like a reasonable query looking for errors. I’m looking for the errors that occurred in the last so many days. The CONVERT_IMPLICIT(datetime,[@1],0) shows me that someone typed this the lazy way of GetDate()-1, which isn’t as efficient as DateAdd(Day, -1, GetDate()), but you’re getting me off topic again.

Fixing a useless index

Looking at the indexes on this table I realize there is only one, and it has the single key column of ID. For uniqueness you can’t do much better than an ID column, but you have to ask yourself if you’ll ever use it to query by.  In this case the index has never had a seek against it, only scans.  Although there table is rarely queried with only 4 uses in the last 2 months (I limited my historical data for this query), it’s still pulling 1.5 GB into cache for every use.  After a couple seconds of shaking my head I start to create a change request to make add TimeStamp in as the first key column in the clustered index.

ScreenHunter_01 2014-01-06 15.35.27

However, I then remembered that this table is 1.5 GB. Is that right? It’s an error table, so if it’s really 1.5 GB then I should either be cleaning up old data or there are so many problems that there is no reason I should be spending my time tuning. Seeing that it has 0 updates in the last 2 months, I already know it’s old data.  To double-check on this I run a simple query, keeping in mind the ID is still the clustered index, to find the oldest record, and discover that we haven’t removed anything from this table in years.

SELECT timestamp
FROM Service_Broker_Errors
WHERE id = (SELECT Min(ID) FROM Service_Broker_Errors)

So I have to talk to people about how old an error can be before we say we just don’t care. It was determined that we probably don’t care about anything more than a month old. I’m paranoid, it comes with the job title, so I made it three months with my change being this:

DECLARE @ID Int

SELECT @ID = MAX(ID) FROM Service_Broker_Errors WHERE TimeStamp < GETDATE()-90 

WHILE @@ROWCOUNT > 0 BEGIN
    DELETE TOP (10000)
    FROM Service_Broker_Errors
    WHERE ID <= @ID 
END 

IF EXISTS (SELECT * FROM sys.indexes WHERE object_id = OBJECT_ID(N'[dbo].[Service_Broker_Errors]') AND name = N'PK_Service_Broker_Errors') 
    ALTER TABLE [dbo].[Service_Broker_Errors] 
    DROP CONSTRAINT [PK_Service_Broker_Errors] 
GO 

ALTER TABLE [dbo].[Service_Broker_Errors] 
ADD CONSTRAINT [PK_Service_Broker_Errors] PRIMARY KEY CLUSTERED 
( 
    [TimeStamp] ASC
    , [id] ASC 
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, IGNORE_DUP_KEY = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON) ON [PRIMARY] 
GO

The reason I did it like this is because I don’t want to fool around with the junk values as I drop the clustered index (rebuilding the table) then recreate the clustered index (rebuilding the table) with all of that information in there, then delete it all and wreak havoc on the new index. Sure, the fragmentation at the page level would be fine since I’m deleting in order, but that’s a lot of changes to the b-tree.

Finally, I add these lines to my nightly cleanup job to keep things trimmed up from now on:

DECLARE @BatchSize Int 

SELECT @BatchSize = 10000 

WHILE @@ROWCOUNT > 0 BEGIN
    DELETE TOP (@BatchSize)
    FROM Service_Broker_Errors
    WHERE TimeStamp < GETDATE()-90
END

I know, I just got done saying that GetDate()-90 was less efficient than DateAdd(Day, -90, GetDate()), but it’s an implicit conversion that’s done once per call. I admit this is a bad habit of mine, and any time there’s even a chance of this being called once per row I have to write it differently than I normally do. Nobody’s perfect, and the most important thing is that I’m getting rid of all those records using batches, right?

In Conclusion with Service Broker Errors

Now that all of this is done I took a 1.5 GB table with all of it in cache to a 20 MB table with only 1 MB in cache. Cleaning up the data had more of an impact that my index change, but it’s usually not that easy to clean stuff up. Even if I wasn’t able to clean up the data, the index change alone would have allowed for the cache usage to be decreased by at least 1.4 GB.

On to the next one

That’s not the only issue I’m seeing here, in fact I bet I could take half of these off the top 10 list. However, today I’m picking up low-hanging fruit and moving on to the next task on my list. In this case, External_Messages is the next one that doesn’t look right to me. Once again, it’s a clustered index (index_id = 1) of a table that isn’t part of the primary focus of the database, which is handling orders.

Starting with the same steps I look in the proc cache to see what put this here. Once again I see a single query in cache referencing the table, but this one is different. It’s the typical IF X = @X or @X IS NULL that we’ve all written, and it’s being used as the predicate.

ScreenHunter_01 2014-01-06 16.34.32

I check the indexes on the table and it has the ID as the only key column of the clustered index, so that’s not an issue. Why isn’t it able to use the clustered index with a seek predicate? After all, I’m rather certain that they almost always, if not always, call this with a parameter where it should be filtered down.

Lets look at the execution plan to make sure. The sniffed parameter value from the first time it was called, as found at the bottom of that statement in the XML version of the execution plan, is, in fact, a non-null value.

ScreenHunter_01 2014-01-06 16.39.02

However, SQL Server can’t guarantee that you’re going to pass it a non-null value, and it has to make an execution plan that can account for either possibility. I’ve seen this before, so I know the basic options for a single optional parameter (there are more options, with increasing complexity). I can either add OPTION (RECOMPILE) to the query or I can rewrite it to be two separate queries.

OPTION (RECOMPILE)

Here’s the option I didn’t choose. This will recompile the statement every run, which isn’t too much of a problem because it’s a quick compile on something that only runs a couple times a day. However, it will make it so this doesn’t show up in the proc cache anymore, which I take advantage of quite a bit (for instance, look at the blog post you’re currently reading). Also, it goes against my rule of using hints as a last resort.
That’s not saying you can’t do it this way, just that I didn’t. The reason it works is because SQL Server knows when it makes this execution plan that it will only be used once, so it doesn’t have to account for the possibility of it being a NULL value next run. In fact, it just makes the execution plan with your parameter as a constant value.

ScreenHunter_01 2014-01-06 16.51.09

Two queries

Here’s the method I decided to go with. Assuming third-normal form doesn’t apply to query text, you should be good to go here. If it’s simple enough then it will be very obvious to anyone editing this at a later point that they need to make any changes in two places.

CREATE PROC NotTheRealProcName
    @id INT = NULL
AS

IF @id IS NULL BEGIN 
    SELECT ID
        , Subject
        , Message
        , DateAdded 
    FROM ExternalMessage EM
END ELSE BEGIN
    SELECT ID
        , Subject
        , Message
        , DateAdded 
    FROM ExternalMessage EM
    WHERE ID = @ID
END

This solution isn’t too complex with just a single parameter as it only creates two queries, but if you get just three parameters and try to do this then you’re up to 8 queries. The OPTION (RECOMPILE) method starts to look pretty good as a quick and easy fix before too long. I would still prefer one of the long and drawn out methods, such as getting interim results to a temp table, making it look a little more complex, but getting it to run efficiently with a cached plan.

It worked as I expected, with no one actually getting all of the IDs. I left the code in place to make it possible to get all of them to avoid digging through all of the application code to find where it could return everything then make sure it gets changed there. The result I was looking for was cleaning up the cache, which I got when this dropped from 1.4 GB down to 25 MB in cache.

The End

Overall I was able to drop about 3 GB out of cache, making room for other data while raising the PLE, in two simple changes that were rather easy to track down just by saying “that doesn’t look like it belongs here”. Even the ones that do look like they belong there probably don’t need to have that large of a presence in the cache.

Although I didn’t actually measure the impact that it would have on the end users in these cases because I didn’t start with a performance complaint, I would imagine that SQL Server tends to run faster returning processing 1 row it found quickly instead of reading through a couple million. So, measured or not, I’ll count that as a success as well.

PS. Jonathan Kehayias helped me

My queries looking into the proc cache are slightly modified versions of Jonathan’s work on his blog. Honestly, it didn’t need tweaked or touched for any reason other than the fact that I learn by tinkering. I did help myself by adding in filtering by database because I have a couple servers that have multiple copies of the same database, but, again, it worked great before I touched it, which is something you should expect from Jonathan’s work.

Here’s my altered version:

SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTED;
DECLARE @IndexName SYSNAME = '[PK_ExternalMessage]'; 
DECLARE @DatabaseName SYSNAME;

SELECT @DatabaseName = '[' + DB_NAME() + ']';

WITH XMLNAMESPACES
   (DEFAULT 'http://schemas.microsoft.com/sqlserver/2004/07/showplan')
SELECT
    n.value('(@StatementText)[1]', 'VARCHAR(4000)') AS sql_text,
    n.query('.'),
    cp.plan_handle,
    i.value('(@PhysicalOp)[1]', 'VARCHAR(128)') AS PhysicalOp,
    i.value('(./IndexScan/@Lookup)[1]', 'VARCHAR(128)') AS IsLookup,
    i.value('(./IndexScan/Object/@Database)[1]', 'VARCHAR(128)') AS DatabaseName,
    i.value('(./IndexScan/Object/@Schema)[1]', 'VARCHAR(128)') AS SchemaName,
    i.value('(./IndexScan/Object/@Table)[1]', 'VARCHAR(128)') AS TableName,
    i.value('(./IndexScan/Object/@Index)[1]', 'VARCHAR(128)') as IndexName,
    i.query('.'),
    STUFF((SELECT DISTINCT ', ' + cg.value('(@Column)[1]', 'VARCHAR(128)')
       FROM i.nodes('./OutputList/ColumnReference') AS t(cg)
       FOR  XML PATH('')),1,2,'') AS output_columns,
    STUFF((SELECT DISTINCT ', ' + cg.value('(@Column)[1]', 'VARCHAR(128)')
       FROM i.nodes('./IndexScan/SeekPredicates/SeekPredicateNew//ColumnReference') AS t(cg)
       FOR  XML PATH('')),1,2,'') AS seek_columns,
    RIGHT(i.value('(./IndexScan/Predicate/ScalarOperator/@ScalarString)[1]', 'VARCHAR(4000)'), len(i.value('(./IndexScan/Predicate/ScalarOperator/@ScalarString)[1]', 'VARCHAR(4000)')) - charindex('.', i.value('(./IndexScan/Predicate/ScalarOperator/@ScalarString)[1]', 'VARCHAR(4000)'))) as Predicate,
    cp.usecounts,
    query_plan
FROM (  SELECT plan_handle, query_plan
        FROM (  SELECT DISTINCT plan_handle
                FROM sys.dm_exec_query_stats WITH(NOLOCK)) AS qs
        OUTER APPLY sys.dm_exec_query_plan(qs.plan_handle) tp
      ) as tab (plan_handle, query_plan)
INNER JOIN sys.dm_exec_cached_plans AS cp 
    ON tab.plan_handle = cp.plan_handle
CROSS APPLY query_plan.nodes('/ShowPlanXML/BatchSequence/Batch/Statements/*') AS q(n)
CROSS APPLY n.nodes('.//RelOp[IndexScan/Object[@Index=sql:variable("@IndexName") and @Database=sql:variable("@DatabaseName")]]' ) as s(i)
--WHERE i.value('(./IndexScan/@Lookup)[1]', 'VARCHAR(128)') = 1
OPTION(RECOMPILE, MAXDOP 1);
Related articles
Advertisement

Indexing Fundamentals

by

First, lets understand what the types of indexes are. If you take a reference book, you have the two types of indexes. The book itself is a clustered index, it contains all the data and has the key field of page number. The index in the back is your nonclustered index, it’s in order by the key field of subject, then has the implied second key field in there for page number, which nonclustered indexes in SQL Server have as well. On top of that, the table of contents at the beginning of the book is just like a b-tree built on every index that helps you get to the right page faster.

The key fields I mentioned are the fields everything is ordered by, just as you would expect a book to be ordered by page number. Nonclustered indexes also have implied key fields, which are all of the clustered index’s key fields that weren’t explicitly made key fields already, which is a great reason to keep your clustered indexes as narrow as possible in most cases, which means they have fewer columns.

If you have a phone book then you have a clustered index with the key fields last_name, first_name, address, phone_number. That means it’s ordered by each one, so any duplicate last names are still in order by first name, and so on. It’s wide for a clustered index, but doesn’t have any nonclustered indexes to worry about, so wide is fine here.

However, the important thing to notice is that if I asked you for people with the last name “Hood”, you’d find them rather quickly. If I asked you whose phone number is 412-555-6789, you’d cringe. The same thing happens in SQL Server where it will do the work you ask of it, but it won’t be cheap, quick and easy. This is why the internet has reverse phone number lookups where they have a nonclustered index on phone number which lets you find the name quickly by just knowing the number.

Indexes can also have included columns which are stored in the index, but no sorting is done on them. These fields don’t even exist in the b-tree for finding the information, only the page level where everything is stored. To simplify the process, this is where you put fields that you have in your SELECT clause but not your WHERE clause since don’t need it sorted to search quicker, you just need it to fulfill your query’s needs.

How are they used, and what do they cost?

A nonclustered index is used exactly the same way you read a reference book. You look up the key field in the back of the book and see what information is there. If you covered all the information that you needed just by looking at that index then you’re done. However, if you want to know more that wasn’t listed in the index then you take the key field of the clustered index (the page number) and do a key lookup by flipping to those pages and reading more about the subject at hand.

If you noticed, reading all that information in the back of the book seems pretty simple and efficient, especially when all of the information is there so you don’t have to touch the book itself. If you don’t have to touch the book then that’s called a covering index, which can be accomplished by adding the fields you want to retrieve to an index as either included or key columns. Depending on how many of these key lookups you have to do and how much extra information it has to get from the clustered index, it may make sense to make your index a covering index.

However, none of this is free. Every index you make has to be maintained. If you ignore filtered indexes, which I’ll get to soon, every time you insert or delete rows on a table then you have to write to each index. So if you have 6 nonclustered indexes, it doesn’t matter if you have a clustered index or a heap (merely a lack of a clustered index), you have to do one write to the table itself and one write to each nonclustered indexes, so 7 writes for that one row.

Add on top of that maintaining those indexes for fragmentation, extra storage needed on disks and backups, and more data that can possibly be held in the cache, and this stuff really starts to get expensive.

How many indexes should I have?

There’s no magic number of indexes you should have on any single table. Too few makes it hard to find the data, and too many makes a maintenance nightmare with data modifications taking longer. What indexes and how many will vary wildly, but there are methods to figure out what’s appropriate for each table. Chances are, many of your tables will benefit from both dropping indexes and creating them.

Consolidate what you have

First, look at the indexes themselves. If you have indexes that have the same first key field as the clustered index then you can consider removing those indexes with little to no detrimental effects. This is regardless as to if these indexes are being used or not.

Now, look at what indexes are never used. This can be done by looking at sys.dm_db_index_usage_stats to get you up to a month of data (assuming a monthly reboot for patching), or, my preferred method, by querying your tables you have set up that monitor this DMV. If an index is never being used it means it’s never helping you, but it is extra weight that’s adding up and slowing your database down. Just make sure you have enough data to say that the index really isn’t needed before relying on this alone.

Also, look for patterns in the indexes themselves by comparing the first couple key fields. If the couple key fields are the same, these are candidates for having the indexes combined into a single index. Depending on how quickly those key fields make the data semi-unique, you may be talking one to three key fields matching before I’m itching to reduce the number of indexes. You can do this by taking the columns that are in the second index that aren’t in the first and adding them to the first index as included columns, then drop the second index. Queries that used to hit either one of these indexes should run slower by an unnoticeable amount, yet you’re only storing data, doing updates, and maintaining a single index instead of two.

When you’re going through this process you just have to be careful that you understand the implementations of dropping a unique index. It is rarely a good idea to get rid of a unique index unless that uniqueness is enforced by other means as well.

Take this example.  I have a heap with a unique index on Col1 that gets scanned, but no seeks.  Also, there’s a NC index on Col2 that has a lot of seeks then is obviously doing a lot of lookups to get more information from the heap.  In this case I happen to know that the “unique” part of the index on Col1 is because that’s an identity column, so I’m not too concerned about maintaining the unique constraint here.  After seeing the usage stats over a month I’m able to restructure this table.  Both nonclustered indexes were dropped, a new unique clustered index was created on Col4 and Col1.  User Lookups were eliminated since it’s all just a clustered index now, we saved about 700 MB on disk, and the compressed equivalent on backups as well.

Heap_With_Unique_NC_Identity

Take another example where you can see that an index is duplicated.  No, none of these are 100% the same, but the second index is the unique (is_unique field), clustered (* for included columns, it’s just my way of doing it) has a single key column on Col1, which is the same as the first nonclustered index.  Seeing the usage like this, I’m going to drop the first index, saving 435 MB of space in my database.  This will have a negative impact in that it will move 626 index scans done throughout the month to an index that’s twice the size, and that’s a problem I’ll revisit later when I’m tuning queries that have high physical reads.

Notice you can also see two issues on this table that I’ll need to look into.  First, I have a nonclustered index that’s larger than my clustered index, something is wrong here.  Perhaps I should have put the fill factor on here to show that wasn’t the issue.  The real answer is that I’m working on a table that somehow got missed on the index maintenance, and I now know that this is possible and needs to be fixed.  Another issue is that there is a filtered index on here that has about 0.1% fewer rows than the unfiltered indexes, which means that you’re getting no benefits while alienating queries that may have used this otherwise.

Unused NC2

And one last example.  The 2nd and 3rd indexes both have a single key field on Col2 and just vary on the included columns.  This is a simple task of combining the two indexes by making one of them include Col1, Col2, Col6, Col7, Col8, and Col9, then dropping the other.  That saves us about 200 MB on disk, and avoids us having duplicated data in the cache.  Ask a SAN or backup admin and they’ll tell you the disk savings is great, but ask an unrealistically informed user and you’ll realize that cache savings is what makes a difference dropping your physical reads, upping your PLE, and making your app runs faster.

Then, there’s another simple task on this table of dropping the final index, which is nothing more than 112 MB of dead weight that was being updated an average of 30 times a day.  Not much to say here, they obviously didn’t use this as much as they thought they would.

Finally, you’re left with the info that User_Lookups is almost as high as the total number of seeks and scans on the nonclustered indexes.  This means that almost every query that hits the nonclustered indexes need to do a key lookup on the clustered index.  Knowing that, if I was going to get to that level of tuning, I would be looking into the queries that hit this table and determine if I want to add more included columns to actually cover the query or admit that I’ll never cover it efficiently and remove some included columns.  My answer here will vary partially depending on how many records are involved – key lookups are ok for 10 rows, but not so much for 1,000,000 rows.

Duplicated Index

None of these are hypothetical situations I forced AdventureWorks to do.  This is a real, production database, and I wrote two queries against it.  One to help me find these problems so I can put a change request in to fix them, and a second query that does the exact same thing while obfuscating the names to protect my employment.  So when you see the duplicated indexes or the sizes I’m working with, you can’t say that would never happen in production….it just did.

You reduced the load, now add to it

Now that you have fewer indexes in your database, it’s time to add to them. There are several ways to figure out what indexes you want to add, and I prefer to do everything manually. That means there’s no DTA (Database Tuning Advisor) here, it’s all scripts with proof that what I’m doing is making a difference.

Where are you finding the queries to tune

There are a couple approaches to finding the queries in most need of attention. Query Stats will give you the queries that take the most resources during your peak times, which will reduce the load on your server the quickest. Traces on duration will give you the queries that take the longest along with sample calls with all the parameters you need. Blocking monitors will give you the queries other queries are waiting on the most. Then, one of my favorites, job shadowing gives you what the pain points are for the users.

The first thing I look at is blocking, since that silently sits in the background making queries that could run extremely fast sit there and wait for a single slow query. If there is a lot of consistent blocking then take the head blockers and start tuning them. This won’t eliminate the fact that they’re grabbing locks, but it will cause those locks to be released quicker. Locks by themselves are not an issue and shouldn’t be eliminated, but the impact the locks have on other queries are an issue that is best to be handled this way. If you don’t have many locks that go too long, you can skip this and revisit it later. I say “too long” instead of giving you a number because it varies a lot; 5 seconds is too long for a web app while a minute isn’t much for a data warehouse.

Once that is out of the way, I’ll start diving into the most expensive queries on the server. They may not be the ones with the longest average run-time, but they add up to the most on the server. If you can cut the reads from 100,000 to 1,000 on a proc called every second then you’re doing more for the server than you would by tuning an hourly query that does 10,000,000 reads. Also, focusing on physical reads will help raise your PLE if that’s a concern on your server, which it usually is. Don’t stick on this for too long, you’re just trying to get the easy tuning out of the way and hopefully change the look of your Top-10 most expensive queries list a bit.

Now I move on to looking at the traces to see if any huge queries are running that are easy tuning opportunities. Again, skim through it, pick up the low hanging fruit and move on. Right now you’re still in the phase of reducing the load on the server, but you may not have hit what the users see yet.

Another place to look is in the cache to find out what indexes are taking up the most space there. If you know your data well enough then you should know what to expect from your cache. What you see there will probably surprise you, and you’ll probably see an Index_id of 0 or 1 (Heap or Clustered Index) sitting there taking 10 times what you would expect it to do. These are cases where there are clustered index scans taking place that you need to find. I was recently able to clear up 12 GB of cache on a server doing a full index scan on a 12 GB table where a 10 MB filtered index was able to do the job, which made PLE almost double.

So, now that you have the server running a little (or a lot) better in general, I get into the work that never ends. Get a hold of the people using the app, either by yourself or asking customer service to arrange it, and find out what they hate the most about the app’s performance. What you need here is for someone to write a list of when anything took a while, what screen they were on, and approximately how long it took. Don’t go requesting anyone to use stopwatches, you’re looking for “a couple seconds”, “about a minute”, or “I click OK then go get a cup of coffee.” Your goal here is to be able to correlate their pain points to something captured in your traces, giving you the procs they’re using along with real-life example calls to dive into.

Something you may find odd is that I’m diving into the part that affects the end users the most last, but I do it on purpose. The other pieces are things I skim through and do what I can in a week or so, but they’ll make the bigger impact on the server in general along with resolving some of the customer’s complaints in the process. In the process, you’re learning more about the server and how everything goes together so that when you start talking to your customers you can make a more immediate impact on their complaints.

After getting through list once I let my ADD take over, and man do I have ADD. The customers have an idea of what a DBA can do for them now to the point that they may even be contacting you directly when something annoys them, developers realize a SQL Server specialist can be a good person to lean on, and your server’s biggest pain point hurts a lot less. Once you get to this point, jump around on this list on your servers and take up a query or two when you feel like it.

Related articles

Query Stats

by

DBAs rarely use the full potential of sys.dm_exec_query_stats.  It’s common to see the queries for looking at the most expensive queries according to any of the stats contained within the current cache, which is great to see.  However, if you grab snapshots of this information the proper way then you can get more detail that will really make a difference in your job.  Capturing this every 4 hours is probably enough, and here’s what it’s going to do for you.

First, using it the way most DBAs do and look at the most expensive queries according to a specific metric.  It works the same for them all, so we’ll just say we’re interested in physical reads in this case because our PLE isn’t what we’re hoping it could be.  You’ve probably seen the queries that directly query sys.dm_exec_query_stats to give you this similar to the one on MSDN for this DMV.  That will give you everything that’s in the cache right now, which could contain queries that recompile every hour in combination with other queries that have remained in the cache since the server was rebooted two weeks ago.  This means that when you’re querying to see what you can do to improve your PLE during business hours you’re getting anything that happens to be there now, not what ran in the time window you’re looking to improve.

Now look at it from the point of view that you capture this information every 4 hours.  You can get the total number of physical reads grouped by sql_handle where the interval_end time is in the last month, on weekdays, between 9:00 AM and 5:00 PM.  This is both a more complete and more accurate picture of what you care about, and you’re more likely to be focusing on the queries you should be tuning from the start.

Once you do start tuning, you want to know what effect you really had, and not just rely on the couple sets of test parameters you used in your NonProd environments.  A couple days after your changes made it to production, it’s a good idea to make sure you had the desired effect.  Query the data you have saved off for the last three days and for the same days last week.

What I went over so far is just realizing how this takes the traditional use of this to a new level, but there’s more functionality here now that it’s being saved off.  Now this is extremely useful in incident response.  Before you could see if anything was hogging resources as seen in my Recently Recompiled Resource Hogs post, but now you take it a step further by being able to verify if what’s running long now has always run that way or if something is different.  Also, with having the plan_handles saved off you can even see if it’s getting a different execution plan, which will help you determine if this was a change in data or if it’s a bad execution plan.

Not to shoot myself in the foot, but this isn’t perfect.  If you capture this data every 4 hours starting at 12:00 and it recompiles at 3:59 then you’ll only capture one minute for that interval.  Also, if you’re dealing with a query that can’t be cached then it won’t be found here at all; this is why you’re also running traces to find the big queries and watching sys.dm_os_performance_counters to know when there are a lot of compiles and recompiles.

To start capturing this, here are the tables you’ll need to capture everything.

SET ANSI_NULLS ON
GO

SET QUOTED_IDENTIFIER ON
GO

SET ANSI_PADDING ON
GO

/*
IF OBJECT_ID('QueryStats') IS NOT NULL BEGIN
    DROP TABLE QueryStats
END

IF OBJECT_ID('QueryStatsLast') IS NOT NULL BEGIN
    DROP TABLE QueryStatsLast
END

IF OBJECT_ID('QueryText') IS NOT NULL BEGIN
    DROP TABLE QueryText
END
--*/

CREATE TABLE [dbo].[QueryStats](
    [interval_start] [smalldatetime] NOT NULL,
    [interval_end] [smalldatetime] NOT NULL,
    [sql_handle] [varbinary](64) NOT NULL,
    [plan_handle] [varbinary](64) NOT NULL,
    [statement_start_offset] [int] NOT NULL,
    [statement_end_offset] [int] NOT NULL,
    [execution_count] [int] NOT NULL,
    [total_elapsed_time_ms] [int] NOT NULL,
    [min_elapsed_time_ms] [int] NOT NULL,
    [max_elapsed_time_ms] [int] NOT NULL,
    [total_worker_time_ms] [int] NOT NULL,
    [min_worker_time_ms] [int] NOT NULL,
    [max_worker_time_ms] [int] NOT NULL,
    [total_logical_reads] [int] NOT NULL,
    [min_logical_reads] [int] NOT NULL,
    [max_logical_reads] [int] NOT NULL,
    [total_physical_reads] [int] NOT NULL,
    [min_physical_reads] [int] NOT NULL,
    [max_physical_reads] [int] NOT NULL,
    [total_logical_writes] [int] NOT NULL,
    [min_logical_writes] [int] NOT NULL,
    [max_logical_writes] [int] NOT NULL,
    [creation_time] [smalldatetime] NOT NULL,
    [last_execution_time] [smalldatetime] NOT NULL,
) ON [PRIMARY]
GO

--Wide is bad, if you have a lot of nonclustered indexes.  
--This is a high-writes, low-reads table, so a single clustered index is probably all I want.
--With Interval_End being the first key field, they're inserted in order and eliminate mass fragmentation
  -- as well as bad write performance.
CREATE UNIQUE CLUSTERED INDEX IX_QueryStats_intervalend_sqlhandle_statementstartoffset_planhandle_U_C ON QueryStats
(
    interval_end
    , sql_handle
    , statement_start_offset
    , plan_handle
)

CREATE TABLE [dbo].[QueryStatsLast](
    [sql_handle] [varbinary](64) NOT NULL,
    [plan_handle] [varbinary](64) NOT NULL,
    [statement_start_offset] [int] NOT NULL,
    [statement_end_offset] [int] NOT NULL,
    [objtype] [nvarchar](20) NOT NULL,
    [execution_count] [bigint] NOT NULL,
    [total_elapsed_time_ms] [bigint] NOT NULL,
    [min_elapsed_time_ms] [bigint] NOT NULL,
    [max_elapsed_time_ms] [bigint] NOT NULL,
    [total_worker_time_ms] [bigint] NOT NULL,
    [min_worker_time_ms] [bigint] NOT NULL,
    [max_worker_time_ms] [bigint] NOT NULL,
    [total_logical_reads] [bigint] NOT NULL,
    [min_logical_reads] [bigint] NOT NULL,
    [max_logical_reads] [bigint] NOT NULL,
    [total_physical_reads] [bigint] NOT NULL,
    [min_physical_reads] [bigint] NOT NULL,
    [max_physical_reads] [bigint] NOT NULL,
    [total_logical_writes] [bigint] NOT NULL,
    [min_logical_writes] [bigint] NOT NULL,
    [max_logical_writes] [bigint] NOT NULL,
    [creation_time] [datetime] NOT NULL,
    [last_execution_time] [datetime] NOT NULL,
    [DateAdded] [datetime] NOT NULL
) ON [PRIMARY]

CREATE UNIQUE CLUSTERED INDEX [IX_QueryStatsLast_sqlhandle_planhandle_statementstartoffset_U_C] ON [dbo].[QueryStatsLast] 
(
    [sql_handle] ASC,
    [plan_handle] ASC,
    [statement_start_offset] ASC
)WITH (PAD_INDEX  = OFF, STATISTICS_NORECOMPUTE  = OFF, SORT_IN_TEMPDB = OFF, IGNORE_DUP_KEY = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS  = ON, ALLOW_PAGE_LOCKS  = ON) ON [PRIMARY]
GO

CREATE TABLE [dbo].[QueryText](
    [sql_handle] [varbinary](64) NOT NULL,
    [QueryText] [nvarchar](max) NOT NULL,
    [DatabaseName] [nvarchar](128) NULL,
    [objtype] [nvarchar](20) NULL
) ON [PRIMARY]

CREATE UNIQUE CLUSTERED INDEX [IX_QueryText_sqlhandle_U_C] ON [dbo].[QueryText] 
(
    [sql_handle] ASC
)WITH (PAD_INDEX  = OFF, STATISTICS_NORECOMPUTE  = OFF, SORT_IN_TEMPDB = OFF, IGNORE_DUP_KEY = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS  = ON, ALLOW_PAGE_LOCKS  = ON) ON [PRIMARY]
GO

Now that the tables are created, here’s how I populate them.  Set this up to run once every 4 hours to keep the data small during typical running, and kick it up to once an hour or more if needed to pinpoint an issue if you need it.

DECLARE @interval_start smalldatetime
    , @interval_end smalldatetime

IF OBJECT_ID('tempdb..#QS') IS NOT NULL BEGIN
    DROP TABLE #QS 
END 

CREATE TABLE [dbo].[#QS](
    [sql_handle] [varbinary](64) NOT NULL,
    [plan_handle] [varbinary](64) NOT NULL,
    [statement_start_offset] [int] NOT NULL,
    [statement_end_offset] [int] NOT NULL,
    [objtype] [nvarchar](20) NOT NULL,
    [execution_count] [bigint] NOT NULL,
    [total_elapsed_time_ms] [bigint] NOT NULL,
    [min_elapsed_time_ms] [bigint] NOT NULL,
    [max_elapsed_time_ms] [bigint] NOT NULL,
    [total_worker_time_ms] [bigint] NOT NULL,
    [min_worker_time_ms] [bigint] NOT NULL,
    [max_worker_time_ms] [bigint] NOT NULL,
    [total_logical_reads] [bigint] NOT NULL,
    [min_logical_reads] [bigint] NOT NULL,
    [max_logical_reads] [bigint] NOT NULL,
    [total_physical_reads] [bigint] NOT NULL,
    [min_physical_reads] [bigint] NOT NULL,
    [max_physical_reads] [bigint] NOT NULL,
    [total_logical_writes] [bigint] NOT NULL,
    [min_logical_writes] [bigint] NOT NULL,
    [max_logical_writes] [bigint] NOT NULL,
    [creation_time] [datetime] NOT NULL,
    [last_execution_time] [datetime] NOT NULL,
    [DateAdded] [datetime] NOT NULL
) 
/*
--This sounded like a great idea, but it just slowed it down several seconds.
CREATE UNIQUE CLUSTERED INDEX TempQS_Cluster ON #QS  
(
    sql_handle
    , plan_handle
    , statement_start_offset
)
*/

--The Cached Plans Object Type is in here in case you want to treat ad-hoc or prepared statements differently
INSERT INTO #QS 
SELECT qs.sql_handle
    , qs.plan_handle
    , qs.statement_start_offset
    , qs.statement_end_offset
    , cp.objtype
    , qs.execution_count 
    , total_elapsed_time_ms = qs.total_elapsed_time/1000
    , min_elapsed_time_ms = qs.min_elapsed_time/1000
    , max_elapsed_time_ms = qs.max_elapsed_time/1000 
    , total_worker_time_ms = qs.total_worker_time/1000 
    , min_worker_time_ms = qs.min_worker_time/1000
    , max_worker_time_ms = qs.max_worker_time/1000
    , qs.total_logical_reads
    , qs.min_logical_reads
    , qs.max_logical_reads
    , qs.total_physical_reads
    , qs.min_physical_reads 
    , qs.max_physical_reads
    , qs.total_logical_writes 
    , qs.min_logical_writes 
    , qs.max_logical_writes
    , qs.creation_time
    , qs.last_execution_time
    , DateAdded = getDate()
FROM sys.dm_exec_query_stats AS qs
    INNER JOIN sys.dm_exec_cached_plans cp on qs.plan_handle = cp.plan_handle
--WHERE cp.objtype NOT IN ('Adhoc')

INSERT INTO QueryText (sql_handle, QueryText, DatabaseName, objtype)
SELECT QS.sql_handle
    , QueryText = qt.text
    , DatabaseName = DB_NAME(max(qt.dbid))
    , max(QS.objtype)
FROM (SELECT #QS.sql_handle
        , #QS.objtype  
        FROM #QS
            LEFT JOIN QueryText QST ON #QS.sql_handle = QST.sql_handle
        WHERE QST.sql_handle IS NULL) QS
    CROSS APPLY sys.dm_exec_sql_text(QS.sql_handle) qt
GROUP BY QS.sql_handle
    , qt.text

--All the values are the same in each table
SELECT TOP 1 @interval_start = dateadded FROM QueryStatsLast
SELECT TOP 1 @interval_end = dateadded FROM #QS

IF @interval_start IS NULL BEGIN
    SELECT @interval_start = create_date
    FROM sys.databases
    WHERE name = 'tempdb'
END

INSERT INTO QueryStats (interval_start
    , interval_end
    , sql_handle
    , plan_handle
    , statement_start_offset
    , statement_end_offset
    , execution_count 
    , total_elapsed_time_ms
    , min_elapsed_time_ms 
    , max_elapsed_time_ms
    , total_worker_time_ms  
    , min_worker_time_ms 
    , max_worker_time_ms 
    , total_logical_reads
    , min_logical_reads
    , max_logical_reads
    , total_physical_reads
    , min_physical_reads 
    , max_physical_reads
    , total_logical_writes 
    , min_logical_writes 
    , max_logical_writes
    , creation_time
    , last_execution_time)
SELECT @interval_start
    , @interval_end
    , qs.sql_handle
    , qs.plan_handle
    , qs.statement_start_offset
    , qs.statement_end_offset
    , qs.execution_count - ISNULL(qsl.execution_count, 0)
    , qs.total_elapsed_time_ms - ISNULL(qsl.total_elapsed_time_ms, 0)
    , qs.min_elapsed_time_ms 
    , qs.max_elapsed_time_ms 
    , qs.total_worker_time_ms - ISNULL(qsl.total_worker_time_ms, 0)
    , qs.min_worker_time_ms
    , qs.max_worker_time_ms 
    , qs.total_logical_reads - ISNULL(qsl.total_logical_reads, 0)
    , qs.min_logical_reads 
    , qs.max_logical_reads 
    , qs.total_physical_reads - ISNULL(qsl.total_physical_reads, 0)
    , qs.min_physical_reads 
    , qs.max_physical_reads 
    , qs.total_logical_writes - ISNULL(qsl.total_logical_writes, 0)
    , qs.min_logical_writes 
    , qs.max_logical_writes 
    , qs.creation_time
    , qs.last_execution_time
FROM #QS qs
    LEFT OUTER JOIN QueryStatsLast qsl ON qs.sql_handle = qsl.sql_handle
        AND qs.plan_handle = qsl.plan_handle
        AND qs.statement_start_offset = qsl.statement_start_offset
        AND qs.creation_time = qsl.creation_time 
WHERE qs.execution_count - ISNULL(qsl.execution_count, 0) > 0 --Only keep records for when it was executed in that interval

TRUNCATE TABLE QueryStatsLast

INSERT INTO QueryStatsLast
SELECT sql_handle
    , plan_handle
    , statement_start_offset
    , statement_end_offset
    , objtype
    , execution_count 
    , total_elapsed_time_ms
    , min_elapsed_time_ms 
    , max_elapsed_time_ms
    , total_worker_time_ms  
    , min_worker_time_ms 
    , max_worker_time_ms 
    , total_logical_reads
    , min_logical_reads
    , max_logical_reads
    , total_physical_reads
    , min_physical_reads 
    , max_physical_reads
    , total_logical_writes 
    , min_logical_writes 
    , max_logical_writes
    , creation_time
    , last_execution_time
    , DateAdded
FROM #QS
Related articles

Blocking – Capturing and Monitoring

by

If a query is taking longer to run than normal, there’s a good chance it’s being blocked by something else. This is especially true when you’re doing something rediculously simple and SQL Server just sits there thinking. Symptoms of blocking problems include a trace of SQL:BatchCompleted and RPC:Completed with durations over 10 seconds comes back with results using less than a second of CPU time and very few reads and writes. Also, if you’re watching your wait stats, then you’ll start to see more waits that start with LCK_ than normal.  Not the mention the overly obvious blocking found in your Running Processes

This query will show you the blocking currently occurring on your server.

SELECT tl.resource_type
    , database_name = DB_NAME(tl.resource_database_id)
    , assoc_entity_id = tl.resource_associated_entity_id
    , lock_req = tl.request_mode
    , waiter_sid = tl.request_session_id
    , wait_duration = wt.wait_duration_ms
    , wt.wait_type
    , waiter_batch = wait_st.text
    , waiter_stmt = substring(wait_st.text,er.statement_start_offset/2 + 1,
                abs(case when er.statement_end_offset = -1
                then len(convert(nvarchar(max), wait_st.text)) * 2
                else er.statement_end_offset end - er.statement_start_offset)/2 + 1)
    , waiter_host = es.host_name
    , waiter_user = es.login_name
    , blocker_sid = wt.blocking_session_id
    , blocker_stmt = block_st.text 
    , blocker_host = block_es.host_name
    , blocker_user = block_es.login_name
FROM sys.dm_tran_locks tl (nolock)
    INNER JOIN sys.dm_os_waiting_tasks wt (nolock) ON tl.lock_owner_address = wt.resource_address
    INNER JOIN sys.dm_os_tasks ot (nolock) ON tl.request_session_id = ot.session_id AND tl.request_request_id = ot.request_id AND tl.request_exec_context_id = ot.exec_context_id
    INNER JOIN sys.dm_exec_requests er (nolock) ON tl.request_session_id = er.session_id AND tl.request_request_id = er.request_id
    INNER JOIN sys.dm_exec_sessions es (nolock) ON tl.request_session_id = es.session_id
    LEFT JOIN sys.dm_exec_requests block_er (nolock) ON wt.blocking_session_id = block_er.session_id
    LEFT JOIN sys.dm_exec_sessions block_es (nolock) ON wt.blocking_session_id = block_es.session_id 
    CROSS APPLY sys.dm_exec_sql_text(er.sql_handle) wait_st
    OUTER APPLY sys.dm_exec_sql_text(block_er.sql_handle) block_st

However, if you have time to watch your servers all the time and keep running that query then you’re overstaffed in a world that leans towards being understaffed. For way too many reasons to list here, throw the results of this into a table every minute. If someone says a query ran long and you see the server was waiting on locks, look here. If you want to be proactive and look for ways you can improve server performance, look here.

CREATE TABLE Blocking (
    BlockingID BigInt Identity(1,1) NOT NULL
    , resource_type NVarChar(60)
    , database_name SysName
    , assoc_entity_id BigInt
    , lock_req NVarChar(60)
    , wait_spid Int
    , wait_duration_ms Int
    , wait_type NVarChar(60)
    , wait_batch NVarChar(max)
    , wait_stmt NVarChar(max)
    , wait_host SysName
    , wait_user SysName
    , block_spid Int
    , block_stmt NVarChar(max)
    , block_host SysName
    , block_user SysName
    , DateAdded datetime NOT NULL DEFAULT (GetDate())
)
GO

CREATE UNIQUE CLUSTERED INDEX IX_Blocking_DateAdded_BlockingID_U_C ON Blocking
(
    DateAdded
    , BlockingID
) WITH (Fillfactor = 95)
GO

So, now you know what your blocking problems are. So, how do you fix them? That’s a deeper dive than I’ll be doing now, but here’s they key words to look into:

  • NOLOCK hint
  • Query tuning
  • Process timing
  • Eliminate cursors

Every situation is going to be unique, and I can’t pretend like I can give you all the answers. Finding that there is an issue can be difficult enough, and those are the answers I can help with.

Related articles