SQL Server

What is deferred compilation?

When I talked about row estimations for table variables, I mentioned ‘deferred compile’, but didn’t give a whole lot of details. What, then, is a deferred compilation? Let’s start with how batches work normally.

T-SQL is an interpreted language. While we talk about compiles, they’re not compilations in the sense of what happens to C++. There’s no conversion of the script to a machine language or intermediate language which is used from that point onwards. Every time a batch executes, it has to be parsed, bound and have an execution plan generated or fetched from cache.

When a batch is parsed, the entire batch is parsed.

SET STATISTICS TIME ON
GO

SELECT * FROM dbo.Clients

SELECT * FROM dbo.StarSystems ss 
INNER JOIN dbo.Stations s ON s.StarSystemID = ss.StarSystemID

SELECT GETDATE()
GO

Ignoring the SET STATISTICS, we’ve got there one batch with two queries in it. Two queries, two execution times but only one parse and compile time. The entire batch was compiled and then the batch was executed.

If we set up an XE session to track compiles, it shows similar

The XE session shows that the compilations for both queries were completed before either started execution.

But what happens if we reference a table that doesn’t exist?

SET STATISTICS TIME ON
GO

SELECT * FROM dbo.Clients

SELECT * FROM dbo.StarSystems_DoesNotExist ss 
INNER JOIN dbo.Stations s ON s.StarSystemID = ss.StarSystemID

SELECT GETDATE()
GO

The first of the queries is still compiled, but not the second. The second can’t be compiled because it references an object that doesn’t exist. The first query then executes, and right before when the second query would execute, it’s sent back to be bound and optimised. In this case, the object still doesn’t exist and so we get an error.

Msg 208, Level 16, State 1, Line 6
Invalid object name 'dbo.StarSystems_DoesNotExist'
.

But if the object was created between the start of the batch and the query that uses it, we get a still different result

SET STATISTICS TIME ON
GO

SELECT * FROM dbo.Clients

SELECT * INTO #StarSystems FROM dbo.StarSystems

SELECT * FROM #StarSystems ss 
INNER JOIN dbo.Stations s ON s.StarSystemID = ss.StarSystemID

SELECT GETDATE()
GO

This time we have three queries. Two of them get an execution plan generated when the batch starts, but the third can’t, because the table it references doesn’t exist. Instead, the statement starts executing but can’t execute because there’s no plan. It gets sent back to the optimiser to be compiled, then the query executes.

This is a deferred compile (also called deferred resolution). A compile that does not happen when the batch starts, but is rather deferred until the point that the query itself executes, usually because the table does not exist at the point the batch starts..

On table variable row estimations

At first glance, the question of how many rows are estimated from a table variable is easy.

But, is it really that simple? Well, not really. To dig into the why, first we need to identify why table variables estimate 1 row. The obvious answer is because they don’t have statistics. However…

ALTER DATABASE CURRENT SET AUTO_CREATE_STATISTICS OFF
GO

CREATE TABLE Test (SomeCol INT);

INSERT INTO Test (SomeCol)
VALUES (1),(22),(37),(45),(55),(67),(72),(86),(91)

SELECT SomeCol FROM Test

SELECT * FROM sys.stats WHERE object_id = OBJECT_ID('Test')

DROP TABLE dbo.Test

That table has no statistics, but it still estimates rows correctly.

So it’s not just the absence of statistics. Hmmm… Where else is there a difference with a table variable?

It has to do with when the plans are generated. The XE event results are from an event tracking statement start and end and the post-compilation event for the plan. For the query using the table variable, the entire batch is compiled before the execution starts. For the permanent table, there are multiple compilation events.

And this is because of something called ‘deferred compile’. For the table variable, the entire batch is compiled at the start, at a time where the table variable does not exist, and because there are no statistics, no recompile is triggered after the insert. Hence, there cannot be any row estimation other than 1 row, because the table did not exist when the estimate was made.

For the permanent table, the compilation of the query that uses the table is deferred until the query starts, not when the batch starts. Hence the plan for the query is generated after the table exists, after it’s been populated. That’s the difference here.

Now, there’s still no statistics, and so there’s no way to get data distribution, but that’s not the only way to get information on the rows in the table. The Storage Engine knows how many rows are in the table, though data distribution isn’t known.

Hence, with a table variable we can expect to see an estimated row count other than 1 any time the table variable exists before the query that uses it is compiled.

That will happen when the table variable is a table-type parameter, when the query using it has the RECOMPILE option, and when SQL 2019’s deferred compile for table variables is in play.

CREATE OR ALTER PROCEDURE TestRowEstimations @Input TestTableType READONLY AS

SELECT SomeCol FROM @Input;

DECLARE @Test TABLE (SomeCol INT);

INSERT INTO @Test (SomeCol)
VALUES (1),(22),(37),(45),(55),(67),(72),(86),(91);

SELECT SomeCol FROM @Test;

SELECT SomeCol FROM @Test OPTION (RECOMPILE);
GO
Table-valued parameter
Normal select on compatibility mode 140
Normal select on compatibility mode 150
Select with OPTION(RECOMPILE)

Pluralsight Free April

It’s a few days into April, but not too late I hope to mention that Pluralsight is offering their entire library, free to new accounts, for the month of April. Sign up on their promotion page to take advantage of this offer.

And, I do have a few courses up there for anyone interested:

And there’s an upcoming course on index maintenance as well that I hope will be published shortly.

Books of 2019

I got sloppy with my book tracking this year, and completely stopped marking what I read on Goodreads part way through the year, hence no full list of what I read.

The high points of the books read however have to be the Expanse series (Babylon’s Ashes and Persepolis Rising were read this year). This series just keeps getting better, larger stakes, more problems. Looking forward to the final book.

Another series that I read that I’ve been thoroughly enjoying is the Starship Mage series. Last year saw the publication of “Sword of Mars“, where we find some answers, get a few massive battles and a really tense cliffhanger. Hope the next book comes out soon.

And the last book that I want to call attention to is “All those Explosions were Someone Else’s Fault“, which is a hilarious and interesting take on the superhero/supervillian genre. I highly recommend it.

I will try and do a better job this year of tracking my reading. My goal is again 75 books in a year, and I’m somewhat confident that I can make that number.

A new way of getting the actual execution plan

Getting the actual execution plan, that is the plan with run-time statistics for a query from an application has always been a little difficult. It’s fine if you can get the query running in Management Studio and reproducing the behaviour from the app, but that can be difficult.

There’s the query_post_execution_showplan event in Extended Events, but that’s a pretty heavy event and not something that I’d like to run on a busy server.

No more! SQL 2019 adds a new plan-related function to get the last actual plan for a query: sys.dm_exec_query_plan_stats.

The function is not available by default, Last_Query_Plan_Stats database scoped configuration has to be set  to allow it to run, and it’s going to add some overhead, how much is still to be determined.

ALTER DATABASE SCOPED CONFIGURATION SET LAST_QUERY_PLAN_STATS = ON

It’s a function which takes a parameter of a plan handle or a sql handle. Hence it can be used alone, or it can be on the right-hand side of an apply from any table or DMV that has a plan handle or sql handle in it. As an example it can be used with QueryStore.

WITH hist
AS (SELECT q.query_id, 
           q.query_hash,
           MAX(rs.max_duration)  AS MaxDuration
    FROM 
        sys.query_store_query q INNER JOIN sys.query_store_plan p ON q.query_id = p.query_id
        INNER JOIN sys.query_store_runtime_stats rs ON p.plan_id = rs.plan_id
        INNER JOIN sys.query_store_runtime_stats_interval rsi ON rs.runtime_stats_interval_id = rsi.runtime_stats_interval_id
    WHERE start_time < DATEADD(HOUR, -1, GETDATE())
    GROUP BY q.query_id, query_hash),
recent
AS (SELECT q.query_id, 
           q.query_hash,
           MAX(rs.max_duration)  AS MaxDuration
    FROM 
        sys.query_store_query q INNER JOIN sys.query_store_plan p ON q.query_id = p.query_id
        INNER JOIN sys.query_store_runtime_stats rs ON p.plan_id = rs.plan_id
        INNER JOIN sys.query_store_runtime_stats_interval rsi ON rs.runtime_stats_interval_id = rsi.runtime_stats_interval_id
    WHERE start_time > DATEADD(HOUR, -1, GETDATE())
    GROUP BY q.query_id, query_hash),
regressed_queries 
AS (
    SELECT hist.query_id, 
            hist.query_hash
        FROM hist INNER JOIN recent ON hist.query_id = recent.query_id
        WHERE recent.MaxDuration > 1.2*hist.MaxDuration
    )
SELECT st.text, OBJECT_NAME(st.objectid) AS ObjectName, qs.last_execution_time, qps.query_plan
    FROM sys.dm_exec_query_stats qs 
        CROSS APPLY sys.dm_exec_sql_text (qs.sql_handle) st
        OUTER APPLY sys.dm_exec_query_plan_stats(qs.plan_handle) qps
    WHERE query_hash IN (SELECT query_hash FROM regressed_queries)

The above query checks query store for any query that has regressed in duration in the last hour (defined as max duration > 120% of previous max duration) and pulls the last actual plan for that query out.

And a look at that plan tells me that I have a bad parameter sniffing problem, a problem that might have been missed or mis-diagnosed with only the estimated plan available.

In-line scalar functions in SQL Server 2019

Yes, yes, yes, finally!

It’s hardly a secret that I’m not a fan of scalar user-defined functions. I refer to them as ‘developer pit-traps’ due to the amount of times I’ve seen developers absolutely wreck their database performance by over-using them (or using them at all).

The main problem with them is that they haven’t been in-line, meaning the function gets evaluated on every single row, and the overhead from doing so is usually terrible.

One of the improvements in SQL Server 2019 is that scalar user-defined functions now are in-line. Not all of them, there are conditions that have to be met. Most scalar UDFs that I’ve seem in client systems will meet them, the not referencing table variables will probably be the main limiting factor.

The full requirements are laid out in the documentation: https://docs.microsoft.com/en-us/sql/relational-databases/user-defined-functions/scalar-udf-inlining

I’m going to use the same function that I used when I evaluated natively-compiled functions (https://sqlinthewild.co.za/index.php/2016/01/12/natively-compiled-user-defined-functions/), and run it against a table with 860k rows in it, both in compat mode 140 (SQL Server 2017) and compat mode 150 (SQL Server 2019)

CREATE FUNCTION dbo.DateOnly (@Input DATETIME)
  RETURNS DATETIME
AS
BEGIN
  RETURN DATEADD(dd, DATEDIFF (dd, 0, @Input), 0);
END
GO

As in the earlier post, I’ll use extended events to catch the performance characteristics.

First, something to compare against. The query, without functions, is:

SELECT DATEADD(dd, DATEDIFF (dd, 0, TransactionDate), 0) FROM Transactions

This takes, on average,  343ms to run, and 320ms of CPU time.

The results of the first test are impressive.

Compat ModeDuration (ms)CPU (ms)
14010 6668594
150356353

I keep having people ask about SCHEMABINDING, so same test again, with the function recreated WITH SCHEMABINDING

Compat ModeDuration (ms)CPU (ms)
14054483818
150325320

Better, but still over an order of magnitude slower than the query without the function in SQL 2017 and earlier.

Last test, what about something with data access? I’ll switch to my Shipments and ShipmentDetails tables for this. The base query without the function is:

SELECT s.ShipmentID, 
    (SELECT SUM(Mass) AS TotalMass FROM ShipmentDetails sd WHERE sd.ShipmentID = s.ShipmentID) TotalShipmentMass
FROM Shipments s;

I’m writing it with a subquery instead of a join to keep it as similar as possible to the version with the function. It should be the same as if I had used a join though. That query takes, on average, 200ms, with 145ms CPU time.

There are 26240 rows in the Shipments table, and on average 34 detail rows per shipment. The function is:

CREATE FUNCTION dbo.ShipmentMass(@ShipmentID INT)
RETURNS NUMERIC(10,2)
AS
BEGIN
    DECLARE @ShipmentMass NUMERIC(10,2);
    SELECT @ShipmentMass = SUM(Mass) FROM ShipmentDetails sd WHERE sd.ShipmentID = @ShipmentID;

    RETURN @ShipmentMass;

END

And the results are:

Compat ModeDuration (ms)CPU (ms)
140961 211 (16 minutes)959 547
15032803272

The test under compat mode 140 had to be run overnight. 9 hours to run the query 25 times… And people wonder why I complain about scalar user-defined functions in systems.

Under compat mode 150 with the inline function it’s way better (3 seconds vs 16 minutes for a single execution), but it’s still over an order of magnitude slower than the same query with the subquery. I’ll test this again after RTM, but for the moment it look like my guidance for functions for SQL 2019 going forward is going to be that scalar functions that don’t access data are fine, but scalar functions that do should still be replaced by inline table-valued functions or no function at all, wherever possible.

No, this is not a bug in T-SQL

(or, Column scope and binding order in subqueries)

I keep seeing this in all sorts of places. People getting an unexpected result when working with a subquery, typically an IN subquery, and assuming that they’ve found a bug in SQL Server.

It’s a bug alright, in that developer’s code though.

Let’s see if anyone can spot the mistake.

We’ll start with a table of orders.

CREATE TABLE Orders (
  OrderID INT IDENTITY PRIMARY KEY,
  ClientID INT,
  OrderNumber VARCHAR(20)
)

There would be more to it in a real system, but this will do for a demo. We’re doing some archiving of old orders, of inactive clients. The IDs of those inactive clients have been put into a temp table

CREATE TABLE #TempClients (
ClientD INT
);

And, to check before running the actual delete, we run the following:

SELECT * FROM dbo.Orders
WHERE ClientID IN (SELECT ClientID FROM #TempClients)

And it returns the entire Orders table. The IN appears to have been completely ignored. At least the query was checked before doing the delete, that’s saved an unpleasant conversation with the DBA if nothing else.

Anyone spotted the mistake yet?

It’s a fairly simple one, not easy to see in passing, but if I test the subquery alone it should become obvious.

The column name in the temp table is missing an I, probably just a typo, but it has some rather pronounced effects.

The obvious next question is why the select with the subquery in it didn’t fail, after all, the query asks for ClientID from #TempClients, and there’s no such column. However there is a ClientID column available in that query, and it’s in the Orders table. And that’s a valid column for the subquery, because column binding order, when we have subqueries, is first to tables within the subquery, and then, if no match is found, to tables in the outer query.

It has to work this way, otherwise correlated subqueries would not be possible. For example:

SELECT c.LegalName,
c.HypernetAddress
FROM dbo.Clients AS c
WHERE EXISTS (SELECT 1 FROM dbo.Shipments s WHERE s.HasLivestock = 1 AND c.ClientID = s.ClientID)

In that example, c.ClientID explicitly references the Client table in the outer query. If I left off the c., the column would be bound to the ClientID column in the Shipments table.

Going back to our original example…

SELECT * FROM dbo.Orders
WHERE ClientID IN (SELECT ClientID FROM #TempClients)

When the query is parsed and bound, the ClientID column mentioned in the subquery does not match any column from any table within the subquery, and hence it’s checked against tables in the outer query, and it does match a column in the orders table. Hence the query essentially becomes

SELECT * FROM dbo.Orders
WHERE ClientID IN (SELECT dbo.Orders.ClientID FROM #TempClients)

Which is essentially equivalent to

SELECT * FROM dbo.Orders
WHERE 1=1

This is one reason why all columns should always, always, always, be qualified with their tables (or table aliases), especially when there are subqueries involved, as doing so would have completely prevented this problem.

SELECT * FROM dbo.Orders o
WHERE o.ClientID IN (SELECT tc.ClientID FROM #TempClients tc)

With the column in the subquery only allowed to be bound to columns within the #TempClients table, the query throws the expected column not found error.

And we’re no longer in danger of deleting everything from the orders table, as we would have if that subquery had been part of a delete and not a select.

Comparing plans in Management Studio

Previously I looked at using Query Store to compare execution plans, but it’s not the only way that two execution plans can be compared. The other method requires a saved execution plan and the Management Studio execution plan viewer.

Let’s start by assuming I have a saved execution plan for a query and I want to compare it to the execution plan that the same query currently has. First step is to run the query with actual execution plan on. Right-click on the execution plan and select ‘Compare Showplan’

ComparePlans2

Pick a saved execution plan. It doesn’t have to be for the same query, but the comparison will be of little use if the two plans being compared are not from the same query.

ComparePlansFindPlan

And then we get the same comparison screen as we saw last time with the comparison via Query Store. Similar portions of the plan are marked by coloured blocks, and the properties window shows which properties differ between the two plans.

ComparePlansDetail2

Comparing plans in Query Store

One feature that was added in the 2016 version of SSMS that hasn’t received a lot of attention, is the ability to compare execution plans.

There’s two ways of doing this, from Query Store and from saved files.

Let’s start with Query Store, and I’m going to use a demo database that I’ve been working on for a few months – Interstellar Transport (IST). I’ve got a stored procedure in there that has a terrible parameter sniffing problem (intentionally). I’m going to run it a few times with one parameter value, then run it a few more times with another parameter value, remove the plan from cache and repeat the executions in the reverse order.

With that done, the query should show up in the ‘Queries with High Variance’ report (SQL 2017)

image

image

The query has the two expected plans, and they are quite different from each other.

Plan1

Plan2

I can click on the points on the graph individually to see the plans, but comparing the plans in that way is difficult and requires that I make notes somewhere else. What I can do instead is select two different points on the graph and chose the ‘compare plans’ option.

image

This brings up a window where the two plans are displayed one above the other, and areas in the plan which are similar are highlighted.

image

Select an operator and pull up the properties, and the properties of the operator from both plans are shown, with the differences highlighted.

image

This isn’t the only way to compare query plans. The next post will show how it can be done without using Query Store at all.

Memory Grant Feedback and data skew

The new adaptive query processing features in SQL Server 2017 are useful for fixing performance problems that were previously very hard to fix. They’re not perfect though, and one of the problems with memory grant feedback in particular is that it’s sensitive to data skew.

Before I get into why, let’s look at what adaptive memory grant does in the first place.

Queries request memory for operations like sorts, hash joins, hash aggregates and a few other operators. This is not TempDB space (ideally), it’s just memory. The amount requested is based on the optimiser’s guesses as to the size of the data that will be hashed/sorted, and that’s based off statistics and parameter values. Hence, there’s a chance for the guess to be wrong, and when it is, we get things like this:

PlanWithSpill

SpillDetails

When spill happen, the intermediate resultsets (or parts of them) do get written to TempDB. And read back. And potentially written and read back again, and maybe a few more times. This can be horribly slow.

Of course, there’s a chance that the estimate will be wrong in the other direction. Too large. It’s not as obviously bad, but it can limit the throughput of the system. Instead of the query running really slowly, it may have to wait before it runs at all, waiting for the memory to be granted. (RESOURCE_SEMAPHORE).

These were really hard problems to fix. There isn’t a query hint to request more or less memory than the estimates would allocate (though you can specify, as a percentage of the resource pool, the max and min memory to be allocated), so fixes had to be creative, typically tricking the optimiser into thinking there were more or fewer rows than there really were, or that the rows were wider (there are some lovely tricks that can be done with CROSS APPLY for example)

Adaptive memory grants don’t do anything to correct the optimiser’s mis-estimates. What they do, is allow the query processor to learn from the mistakes. If a query’s memory grant is significantly over or under what is needed, then a note is made of that, somewhere in memory, and the next time the query runs, the memory grant is adjusted to a value based on what the previous execution needed.

So, if we run the example from above a second time, making absolutely no changes in the process, the spills are gone.

SpillGone

This is great, unless you have a particular pattern in your workload, where one query will sometimes have a small number of rows flowing through it, and sometimes a large number. This is not a problem specific to Memory Grant Feedback. It’s been around for a long time, we call it bad parameter sniffing in many cases.

So let’s try a test of running the same query multiple times, alternating between parameter values that return small row counts and parameter values that return large row counts. The plan is the same in all cases, it’s a reused cached plan, and it’s one that’s not bad for the larger row counts (hash join, hash aggregate), so we don’t have the typical bad parameter sniffing problem, but the memory grant will oscillate, being based upon the previous query’s execution. I’m going to execute the stored procedure 200 times.

And I should mention that this is an extreme case. I specifically constructed a scenario where the memory grant required by one execution would be completely inappropriate for the next one. This is not (I hope) something that would happen in the real world.

I monitored what was happening with Extended Events, with the memory_grant_updated_by_feedback and memory_grant_feedback_loop_disabled events.

The results were kinda as expected.

image

And then something interesting happened. I didn’t clear the cache or anything, this was as the procedures executed in a loop.

image

After 8 executions, each with a memory grant update, both the execution count and the count of updates to the grant reset to 1.

This happened again 8 executions later

image

And again 8 executions later

image

Then, finally, after 32 executions, the update is disabled.

image

The procedure then went on to execute a further 168 times, with the same memory grant each time, equal to the last updated value.

So what can we conclude from this?

Firstly, there seems to be a re-evaluation of the memory grant feedback process every 8 modifications, deciding whether to continue adjusting. Second, it will stop adjusting memory grants at some point, though the conditions aren’t documented and I can’t tell from the test I ran what the conditions are. Since they’re not documented, they will probably change in future CUs/versions without notice.

Once the feedback cycle stops, the last memory grant value is what will be used for that query until its plan is removed from cache, at which point the adjustment cycle starts over from scratch.

If you’re working with a system that has this kind of query, with wide differences in optimal memory grant, I would suggest not relying on memory grant feedback, and changing the code so that the grant needed is more constant. This may require splitting procedures up, optimise hints or other fixes for bad parameter sniffing.

I suggest that because the feedback works great for ‘dialling in’ a good value for needed memory grant, but not for cases where the optimal grant is constantly changing. The 200 executions above took 4 minutes total without memory grant feedback, but 12 minutes with memory grant feedback.

It’s a great solution when the original estimate doesn’t match what the query needs, but it’s sub-optimal for queries with constantly changing memory needs. Procedures with widely changing memory needs should be fixed with other methods, including but not limited to multiple procedures, dynamic SQL, plan forcing, or other query hints.