SQL Server 2016 - excessive memory grant warning on poor performing query The Next CEO of...

What happens if you roll doubles 3 times then land on "Go to jail?"

Workaholic Formal/Informal

How to add tiny 0.5A 120V load to very remote split phase 240v 3 wire well house

Is HostGator storing my password in plaintext?

Help understanding this unsettling image of Titan, Epimetheus, and Saturn's rings?

How to avoid supervisors with prejudiced views?

How to transpose the 1st and -1th levels of arbitrarily nested array?

Novel about a guy who is possessed by the divine essence and the world ends?

How do I go from 300 unfinished/half written blog posts, to published posts?

Skipping indices in a product

Does it take more energy to get to Venus or to Mars?

Solving Integral Equation by Converting to Differential Equations

"and that skill is always a class skill for you" - does "always" have any meaning in Pathfinder?

Why do remote companies require working in the US?

Is it possible to search for a directory/file combination?

How fast would a person need to move to trick the eye?

calculus parametric curve length

Can you replace a racial trait cantrip when leveling up?

What is ( CFMCC ) on ILS approach chart?

How to solve a differential equation with a term to a power?

Rotate a column

What connection does MS Office have to Netscape Navigator?

Is there a difference between "Fahrstuhl" and "Aufzug"

How to count occurrences of text in a file?



SQL Server 2016 - excessive memory grant warning on poor performing query



The Next CEO of Stack OverflowFix for slow SQL_INLINE_TABLE_VALUED_FUNCTIONLarge memory grant requestsPoor performing Query -Tsql execution plan - estimated number of rows =1 Paste the PlanMSSQL - Query had to wait for memory grantRow estimates always too lowBad performance using “NOT IN”Warning about memory “Excessive Grant” in the query plan - how to find out what is causing it?Optimizing table valued function SQL ServerWhen does SQL Server warn about an Excessive Memory Grant?Warning in Execution Plan












2















I have a relatively large database of 550GB on a SQL Server 2016 EE instance which has a max memory limit of 112GB of the total 128GB RAM available to the OS. The database is at the latest compatibility level of 130. Developers have complained of the below query which executes within an acceptable time to them of 30 seconds when executed in isolation, but when they run their processes at scale the same query is executed multiple times concurrently across several threads and this is when they have observed that the execution time suffers and performance/throughput drops. The problematic T-SQL is:



select distinct dg.entityId, et.EntityName, dg.Version
from DataGathering dg with(nolock)
inner join entity e with(nolock)
on e.EntityId = dg.EntityId
inner join entitytype et with(nolock)
on et.EntityTypeID = e.EntityTypeID
and et.EntityName = 'Account_Third_Party_Details'
inner join entitymapping em with(nolock)
on em.ChildEntityId = dg.EntityId
and em.ParentEntityId = -1
where dg.EntityId = dg.RootId

union all

select distinct dg1.EntityId, et.EntityName, dg1.version
from datagathering dg1 with(nolock)
inner join entity e with(nolock)
on e.EntityId = dg1.EntityId
inner join entitytype et with(nolock)
on et.EntityTypeID = e.EntityTypeID
and et.EntityName = 'TIN_Details'
where dg1.EntityId = dg1.RootId
and dg1.EntityId not in (
select distinct ChildEntityId
from entitymapping
where ChildEntityId = dg1.EntityId
and ParentEntityId = -1)


The actual execution plan shows the below memory grant warning:



enter image description here



The graphical execution plan can be found here:



https://www.brentozar.com/pastetheplan/?id=r18ZtCidN



Below are the row counts and sizes of the tables touched by this query. The most expensive operator is an index scan of a non-clustered index on the DataGathering table which makes sense given the size of the table compared to the others. I understand why/how the memory grant is required which I believe is due to how the query is written which requires multiple sorts and hash operators. What I need advice/guidance on is how to avoid the memory grants, T-SQL and re-factoring code is not my strong point, is there a way to re-write this query so that it is more performant? If I can tune the query to run faster in isolation then hopefully the benefits would transfer to when it is run at scale which is when the performance starts to suffer. Happy to provide any more information and hoping to learn something from this!



enter image description here



After updating statistics on 3 of the tables:



UPDATE STATISTICS Entity WITH FULLSCAN; 
UPDATE STATISTICS EntityMapping WITH FULLSCAN;
UPDATE STATISTICS EntityType WITH FULLSCAN;


...the execution plan has improved some:



https://www.brentozar.com/pastetheplan/?id=rkVmdkh_4



Unfortunately, the "Excessive Grant" warning is still there.










share|improve this question

























  • Thanks for your input. I'll run update statistics with fullscan against the four tables listed in my post and let you know if that makes any difference and if the execution plan changes. It will take some time since the DataGathering table is large! I was hoping to focus my efforts on re-writing that hideous query though. So are you saying that removing the distinct keyword throughout the entire query and replacing union all with union is logically the same and will return the same data?

    – Fza
    5 hours ago






  • 2





    "So are you saying that removing the distinct keyword throughout the entire query and replacing union all with union is logically the same and will return the same data?" - No, that is not logically equivalent. Your current query removes duplicates from each of the individual sets (with DISTINCT), and then combines those sets with UNION ALL - allowing duplicates between the two sets. Kin's suggestion eliminates all duplicate rows, even those between the two sets, so results could be different.

    – Josh Darnell
    4 hours ago


















2















I have a relatively large database of 550GB on a SQL Server 2016 EE instance which has a max memory limit of 112GB of the total 128GB RAM available to the OS. The database is at the latest compatibility level of 130. Developers have complained of the below query which executes within an acceptable time to them of 30 seconds when executed in isolation, but when they run their processes at scale the same query is executed multiple times concurrently across several threads and this is when they have observed that the execution time suffers and performance/throughput drops. The problematic T-SQL is:



select distinct dg.entityId, et.EntityName, dg.Version
from DataGathering dg with(nolock)
inner join entity e with(nolock)
on e.EntityId = dg.EntityId
inner join entitytype et with(nolock)
on et.EntityTypeID = e.EntityTypeID
and et.EntityName = 'Account_Third_Party_Details'
inner join entitymapping em with(nolock)
on em.ChildEntityId = dg.EntityId
and em.ParentEntityId = -1
where dg.EntityId = dg.RootId

union all

select distinct dg1.EntityId, et.EntityName, dg1.version
from datagathering dg1 with(nolock)
inner join entity e with(nolock)
on e.EntityId = dg1.EntityId
inner join entitytype et with(nolock)
on et.EntityTypeID = e.EntityTypeID
and et.EntityName = 'TIN_Details'
where dg1.EntityId = dg1.RootId
and dg1.EntityId not in (
select distinct ChildEntityId
from entitymapping
where ChildEntityId = dg1.EntityId
and ParentEntityId = -1)


The actual execution plan shows the below memory grant warning:



enter image description here



The graphical execution plan can be found here:



https://www.brentozar.com/pastetheplan/?id=r18ZtCidN



Below are the row counts and sizes of the tables touched by this query. The most expensive operator is an index scan of a non-clustered index on the DataGathering table which makes sense given the size of the table compared to the others. I understand why/how the memory grant is required which I believe is due to how the query is written which requires multiple sorts and hash operators. What I need advice/guidance on is how to avoid the memory grants, T-SQL and re-factoring code is not my strong point, is there a way to re-write this query so that it is more performant? If I can tune the query to run faster in isolation then hopefully the benefits would transfer to when it is run at scale which is when the performance starts to suffer. Happy to provide any more information and hoping to learn something from this!



enter image description here



After updating statistics on 3 of the tables:



UPDATE STATISTICS Entity WITH FULLSCAN; 
UPDATE STATISTICS EntityMapping WITH FULLSCAN;
UPDATE STATISTICS EntityType WITH FULLSCAN;


...the execution plan has improved some:



https://www.brentozar.com/pastetheplan/?id=rkVmdkh_4



Unfortunately, the "Excessive Grant" warning is still there.










share|improve this question

























  • Thanks for your input. I'll run update statistics with fullscan against the four tables listed in my post and let you know if that makes any difference and if the execution plan changes. It will take some time since the DataGathering table is large! I was hoping to focus my efforts on re-writing that hideous query though. So are you saying that removing the distinct keyword throughout the entire query and replacing union all with union is logically the same and will return the same data?

    – Fza
    5 hours ago






  • 2





    "So are you saying that removing the distinct keyword throughout the entire query and replacing union all with union is logically the same and will return the same data?" - No, that is not logically equivalent. Your current query removes duplicates from each of the individual sets (with DISTINCT), and then combines those sets with UNION ALL - allowing duplicates between the two sets. Kin's suggestion eliminates all duplicate rows, even those between the two sets, so results could be different.

    – Josh Darnell
    4 hours ago
















2












2








2


1






I have a relatively large database of 550GB on a SQL Server 2016 EE instance which has a max memory limit of 112GB of the total 128GB RAM available to the OS. The database is at the latest compatibility level of 130. Developers have complained of the below query which executes within an acceptable time to them of 30 seconds when executed in isolation, but when they run their processes at scale the same query is executed multiple times concurrently across several threads and this is when they have observed that the execution time suffers and performance/throughput drops. The problematic T-SQL is:



select distinct dg.entityId, et.EntityName, dg.Version
from DataGathering dg with(nolock)
inner join entity e with(nolock)
on e.EntityId = dg.EntityId
inner join entitytype et with(nolock)
on et.EntityTypeID = e.EntityTypeID
and et.EntityName = 'Account_Third_Party_Details'
inner join entitymapping em with(nolock)
on em.ChildEntityId = dg.EntityId
and em.ParentEntityId = -1
where dg.EntityId = dg.RootId

union all

select distinct dg1.EntityId, et.EntityName, dg1.version
from datagathering dg1 with(nolock)
inner join entity e with(nolock)
on e.EntityId = dg1.EntityId
inner join entitytype et with(nolock)
on et.EntityTypeID = e.EntityTypeID
and et.EntityName = 'TIN_Details'
where dg1.EntityId = dg1.RootId
and dg1.EntityId not in (
select distinct ChildEntityId
from entitymapping
where ChildEntityId = dg1.EntityId
and ParentEntityId = -1)


The actual execution plan shows the below memory grant warning:



enter image description here



The graphical execution plan can be found here:



https://www.brentozar.com/pastetheplan/?id=r18ZtCidN



Below are the row counts and sizes of the tables touched by this query. The most expensive operator is an index scan of a non-clustered index on the DataGathering table which makes sense given the size of the table compared to the others. I understand why/how the memory grant is required which I believe is due to how the query is written which requires multiple sorts and hash operators. What I need advice/guidance on is how to avoid the memory grants, T-SQL and re-factoring code is not my strong point, is there a way to re-write this query so that it is more performant? If I can tune the query to run faster in isolation then hopefully the benefits would transfer to when it is run at scale which is when the performance starts to suffer. Happy to provide any more information and hoping to learn something from this!



enter image description here



After updating statistics on 3 of the tables:



UPDATE STATISTICS Entity WITH FULLSCAN; 
UPDATE STATISTICS EntityMapping WITH FULLSCAN;
UPDATE STATISTICS EntityType WITH FULLSCAN;


...the execution plan has improved some:



https://www.brentozar.com/pastetheplan/?id=rkVmdkh_4



Unfortunately, the "Excessive Grant" warning is still there.










share|improve this question
















I have a relatively large database of 550GB on a SQL Server 2016 EE instance which has a max memory limit of 112GB of the total 128GB RAM available to the OS. The database is at the latest compatibility level of 130. Developers have complained of the below query which executes within an acceptable time to them of 30 seconds when executed in isolation, but when they run their processes at scale the same query is executed multiple times concurrently across several threads and this is when they have observed that the execution time suffers and performance/throughput drops. The problematic T-SQL is:



select distinct dg.entityId, et.EntityName, dg.Version
from DataGathering dg with(nolock)
inner join entity e with(nolock)
on e.EntityId = dg.EntityId
inner join entitytype et with(nolock)
on et.EntityTypeID = e.EntityTypeID
and et.EntityName = 'Account_Third_Party_Details'
inner join entitymapping em with(nolock)
on em.ChildEntityId = dg.EntityId
and em.ParentEntityId = -1
where dg.EntityId = dg.RootId

union all

select distinct dg1.EntityId, et.EntityName, dg1.version
from datagathering dg1 with(nolock)
inner join entity e with(nolock)
on e.EntityId = dg1.EntityId
inner join entitytype et with(nolock)
on et.EntityTypeID = e.EntityTypeID
and et.EntityName = 'TIN_Details'
where dg1.EntityId = dg1.RootId
and dg1.EntityId not in (
select distinct ChildEntityId
from entitymapping
where ChildEntityId = dg1.EntityId
and ParentEntityId = -1)


The actual execution plan shows the below memory grant warning:



enter image description here



The graphical execution plan can be found here:



https://www.brentozar.com/pastetheplan/?id=r18ZtCidN



Below are the row counts and sizes of the tables touched by this query. The most expensive operator is an index scan of a non-clustered index on the DataGathering table which makes sense given the size of the table compared to the others. I understand why/how the memory grant is required which I believe is due to how the query is written which requires multiple sorts and hash operators. What I need advice/guidance on is how to avoid the memory grants, T-SQL and re-factoring code is not my strong point, is there a way to re-write this query so that it is more performant? If I can tune the query to run faster in isolation then hopefully the benefits would transfer to when it is run at scale which is when the performance starts to suffer. Happy to provide any more information and hoping to learn something from this!



enter image description here



After updating statistics on 3 of the tables:



UPDATE STATISTICS Entity WITH FULLSCAN; 
UPDATE STATISTICS EntityMapping WITH FULLSCAN;
UPDATE STATISTICS EntityType WITH FULLSCAN;


...the execution plan has improved some:



https://www.brentozar.com/pastetheplan/?id=rkVmdkh_4



Unfortunately, the "Excessive Grant" warning is still there.







sql-server t-sql query-performance sql-server-2016 memory-grant






share|improve this question















share|improve this question













share|improve this question




share|improve this question








edited 4 hours ago









Josh Darnell

7,37522241




7,37522241










asked 5 hours ago









FzaFza

3761414




3761414













  • Thanks for your input. I'll run update statistics with fullscan against the four tables listed in my post and let you know if that makes any difference and if the execution plan changes. It will take some time since the DataGathering table is large! I was hoping to focus my efforts on re-writing that hideous query though. So are you saying that removing the distinct keyword throughout the entire query and replacing union all with union is logically the same and will return the same data?

    – Fza
    5 hours ago






  • 2





    "So are you saying that removing the distinct keyword throughout the entire query and replacing union all with union is logically the same and will return the same data?" - No, that is not logically equivalent. Your current query removes duplicates from each of the individual sets (with DISTINCT), and then combines those sets with UNION ALL - allowing duplicates between the two sets. Kin's suggestion eliminates all duplicate rows, even those between the two sets, so results could be different.

    – Josh Darnell
    4 hours ago





















  • Thanks for your input. I'll run update statistics with fullscan against the four tables listed in my post and let you know if that makes any difference and if the execution plan changes. It will take some time since the DataGathering table is large! I was hoping to focus my efforts on re-writing that hideous query though. So are you saying that removing the distinct keyword throughout the entire query and replacing union all with union is logically the same and will return the same data?

    – Fza
    5 hours ago






  • 2





    "So are you saying that removing the distinct keyword throughout the entire query and replacing union all with union is logically the same and will return the same data?" - No, that is not logically equivalent. Your current query removes duplicates from each of the individual sets (with DISTINCT), and then combines those sets with UNION ALL - allowing duplicates between the two sets. Kin's suggestion eliminates all duplicate rows, even those between the two sets, so results could be different.

    – Josh Darnell
    4 hours ago



















Thanks for your input. I'll run update statistics with fullscan against the four tables listed in my post and let you know if that makes any difference and if the execution plan changes. It will take some time since the DataGathering table is large! I was hoping to focus my efforts on re-writing that hideous query though. So are you saying that removing the distinct keyword throughout the entire query and replacing union all with union is logically the same and will return the same data?

– Fza
5 hours ago





Thanks for your input. I'll run update statistics with fullscan against the four tables listed in my post and let you know if that makes any difference and if the execution plan changes. It will take some time since the DataGathering table is large! I was hoping to focus my efforts on re-writing that hideous query though. So are you saying that removing the distinct keyword throughout the entire query and replacing union all with union is logically the same and will return the same data?

– Fza
5 hours ago




2




2





"So are you saying that removing the distinct keyword throughout the entire query and replacing union all with union is logically the same and will return the same data?" - No, that is not logically equivalent. Your current query removes duplicates from each of the individual sets (with DISTINCT), and then combines those sets with UNION ALL - allowing duplicates between the two sets. Kin's suggestion eliminates all duplicate rows, even those between the two sets, so results could be different.

– Josh Darnell
4 hours ago







"So are you saying that removing the distinct keyword throughout the entire query and replacing union all with union is logically the same and will return the same data?" - No, that is not logically equivalent. Your current query removes duplicates from each of the individual sets (with DISTINCT), and then combines those sets with UNION ALL - allowing duplicates between the two sets. Kin's suggestion eliminates all duplicate rows, even those between the two sets, so results could be different.

– Josh Darnell
4 hours ago












1 Answer
1






active

oldest

votes


















4














This might not help with the memory grant situation (hopefully the additional stats updates will help some with that), but I noticed that parallelism is being inhibited in this query. Check out this part of the plan:



screenshot of plan explorer window



Since there's only one row on the outer side of the nested loops join, all 900k rows are being funneled onto one thread. So despite this query running at DOP 8, this portion of the plan is completely serial. That includes the sort. Here's the XML for that sort:



screenshot of plan XML showing unbalanced parallelism



If at all possible, consider avoiding the join to EntityType, and instead just grabbing that Id and filtering the Entity table with it. This will allow it to just be a predicate on an index scan of the Entity table, hopefully allowing parallelism and speeding up the execution.



Something like this:



DECLARE @tinDetailsId int;

SELECT @tinDetailsId = et.EntityTypeID
FROM entitytype et
WHERE et.EntityName = 'TIN_Details';


Which you could then reference in the bottom half of the query, eliminating the join:



select distinct dg1.EntityId, et.EntityName, dg1.version
from datagathering dg1 with(nolock)
inner join entity e with(nolock)
on e.EntityId = dg1.EntityId
where dg1.EntityId = dg1.RootId
and e.EntityTypeID = @tinDetailsId
and dg1.EntityId not in (
select distinct ChildEntityId
from entitymapping
where ChildEntityId = dg1.EntityId
and ParentEntityId = -1)


You would want to do the same thing with EntityName "Account_Third_Party_Details" in the top part of the query, as it has the same problem - with twice as many rows.



PS: Totally unrelated to the topic at hand, I noticed that you have nolock hints on all the tables in this query. Make sure that you are aware of the implications of this. Check out this nifty blog posts on the topic:



Bad habits : Putting NOLOCK everywhere by Aaron Bertrand
The Read Uncommitted Isolation Level by Paul White






share|improve this answer


























    Your Answer








    StackExchange.ready(function() {
    var channelOptions = {
    tags: "".split(" "),
    id: "182"
    };
    initTagRenderer("".split(" "), "".split(" "), channelOptions);

    StackExchange.using("externalEditor", function() {
    // Have to fire editor after snippets, if snippets enabled
    if (StackExchange.settings.snippets.snippetsEnabled) {
    StackExchange.using("snippets", function() {
    createEditor();
    });
    }
    else {
    createEditor();
    }
    });

    function createEditor() {
    StackExchange.prepareEditor({
    heartbeatType: 'answer',
    autoActivateHeartbeat: false,
    convertImagesToLinks: false,
    noModals: true,
    showLowRepImageUploadWarning: true,
    reputationToPostImages: null,
    bindNavPrevention: true,
    postfix: "",
    imageUploader: {
    brandingHtml: "Powered by u003ca class="icon-imgur-white" href="https://imgur.com/"u003eu003c/au003e",
    contentPolicyHtml: "User contributions licensed under u003ca href="https://creativecommons.org/licenses/by-sa/3.0/"u003ecc by-sa 3.0 with attribution requiredu003c/au003e u003ca href="https://stackoverflow.com/legal/content-policy"u003e(content policy)u003c/au003e",
    allowUrls: true
    },
    onDemand: true,
    discardSelector: ".discard-answer"
    ,immediatelyShowMarkdownHelp:true
    });


    }
    });














    draft saved

    draft discarded


















    StackExchange.ready(
    function () {
    StackExchange.openid.initPostLogin('.new-post-login', 'https%3a%2f%2fdba.stackexchange.com%2fquestions%2f233536%2fsql-server-2016-excessive-memory-grant-warning-on-poor-performing-query%23new-answer', 'question_page');
    }
    );

    Post as a guest















    Required, but never shown

























    1 Answer
    1






    active

    oldest

    votes








    1 Answer
    1






    active

    oldest

    votes









    active

    oldest

    votes






    active

    oldest

    votes









    4














    This might not help with the memory grant situation (hopefully the additional stats updates will help some with that), but I noticed that parallelism is being inhibited in this query. Check out this part of the plan:



    screenshot of plan explorer window



    Since there's only one row on the outer side of the nested loops join, all 900k rows are being funneled onto one thread. So despite this query running at DOP 8, this portion of the plan is completely serial. That includes the sort. Here's the XML for that sort:



    screenshot of plan XML showing unbalanced parallelism



    If at all possible, consider avoiding the join to EntityType, and instead just grabbing that Id and filtering the Entity table with it. This will allow it to just be a predicate on an index scan of the Entity table, hopefully allowing parallelism and speeding up the execution.



    Something like this:



    DECLARE @tinDetailsId int;

    SELECT @tinDetailsId = et.EntityTypeID
    FROM entitytype et
    WHERE et.EntityName = 'TIN_Details';


    Which you could then reference in the bottom half of the query, eliminating the join:



    select distinct dg1.EntityId, et.EntityName, dg1.version
    from datagathering dg1 with(nolock)
    inner join entity e with(nolock)
    on e.EntityId = dg1.EntityId
    where dg1.EntityId = dg1.RootId
    and e.EntityTypeID = @tinDetailsId
    and dg1.EntityId not in (
    select distinct ChildEntityId
    from entitymapping
    where ChildEntityId = dg1.EntityId
    and ParentEntityId = -1)


    You would want to do the same thing with EntityName "Account_Third_Party_Details" in the top part of the query, as it has the same problem - with twice as many rows.



    PS: Totally unrelated to the topic at hand, I noticed that you have nolock hints on all the tables in this query. Make sure that you are aware of the implications of this. Check out this nifty blog posts on the topic:



    Bad habits : Putting NOLOCK everywhere by Aaron Bertrand
    The Read Uncommitted Isolation Level by Paul White






    share|improve this answer






























      4














      This might not help with the memory grant situation (hopefully the additional stats updates will help some with that), but I noticed that parallelism is being inhibited in this query. Check out this part of the plan:



      screenshot of plan explorer window



      Since there's only one row on the outer side of the nested loops join, all 900k rows are being funneled onto one thread. So despite this query running at DOP 8, this portion of the plan is completely serial. That includes the sort. Here's the XML for that sort:



      screenshot of plan XML showing unbalanced parallelism



      If at all possible, consider avoiding the join to EntityType, and instead just grabbing that Id and filtering the Entity table with it. This will allow it to just be a predicate on an index scan of the Entity table, hopefully allowing parallelism and speeding up the execution.



      Something like this:



      DECLARE @tinDetailsId int;

      SELECT @tinDetailsId = et.EntityTypeID
      FROM entitytype et
      WHERE et.EntityName = 'TIN_Details';


      Which you could then reference in the bottom half of the query, eliminating the join:



      select distinct dg1.EntityId, et.EntityName, dg1.version
      from datagathering dg1 with(nolock)
      inner join entity e with(nolock)
      on e.EntityId = dg1.EntityId
      where dg1.EntityId = dg1.RootId
      and e.EntityTypeID = @tinDetailsId
      and dg1.EntityId not in (
      select distinct ChildEntityId
      from entitymapping
      where ChildEntityId = dg1.EntityId
      and ParentEntityId = -1)


      You would want to do the same thing with EntityName "Account_Third_Party_Details" in the top part of the query, as it has the same problem - with twice as many rows.



      PS: Totally unrelated to the topic at hand, I noticed that you have nolock hints on all the tables in this query. Make sure that you are aware of the implications of this. Check out this nifty blog posts on the topic:



      Bad habits : Putting NOLOCK everywhere by Aaron Bertrand
      The Read Uncommitted Isolation Level by Paul White






      share|improve this answer




























        4












        4








        4







        This might not help with the memory grant situation (hopefully the additional stats updates will help some with that), but I noticed that parallelism is being inhibited in this query. Check out this part of the plan:



        screenshot of plan explorer window



        Since there's only one row on the outer side of the nested loops join, all 900k rows are being funneled onto one thread. So despite this query running at DOP 8, this portion of the plan is completely serial. That includes the sort. Here's the XML for that sort:



        screenshot of plan XML showing unbalanced parallelism



        If at all possible, consider avoiding the join to EntityType, and instead just grabbing that Id and filtering the Entity table with it. This will allow it to just be a predicate on an index scan of the Entity table, hopefully allowing parallelism and speeding up the execution.



        Something like this:



        DECLARE @tinDetailsId int;

        SELECT @tinDetailsId = et.EntityTypeID
        FROM entitytype et
        WHERE et.EntityName = 'TIN_Details';


        Which you could then reference in the bottom half of the query, eliminating the join:



        select distinct dg1.EntityId, et.EntityName, dg1.version
        from datagathering dg1 with(nolock)
        inner join entity e with(nolock)
        on e.EntityId = dg1.EntityId
        where dg1.EntityId = dg1.RootId
        and e.EntityTypeID = @tinDetailsId
        and dg1.EntityId not in (
        select distinct ChildEntityId
        from entitymapping
        where ChildEntityId = dg1.EntityId
        and ParentEntityId = -1)


        You would want to do the same thing with EntityName "Account_Third_Party_Details" in the top part of the query, as it has the same problem - with twice as many rows.



        PS: Totally unrelated to the topic at hand, I noticed that you have nolock hints on all the tables in this query. Make sure that you are aware of the implications of this. Check out this nifty blog posts on the topic:



        Bad habits : Putting NOLOCK everywhere by Aaron Bertrand
        The Read Uncommitted Isolation Level by Paul White






        share|improve this answer















        This might not help with the memory grant situation (hopefully the additional stats updates will help some with that), but I noticed that parallelism is being inhibited in this query. Check out this part of the plan:



        screenshot of plan explorer window



        Since there's only one row on the outer side of the nested loops join, all 900k rows are being funneled onto one thread. So despite this query running at DOP 8, this portion of the plan is completely serial. That includes the sort. Here's the XML for that sort:



        screenshot of plan XML showing unbalanced parallelism



        If at all possible, consider avoiding the join to EntityType, and instead just grabbing that Id and filtering the Entity table with it. This will allow it to just be a predicate on an index scan of the Entity table, hopefully allowing parallelism and speeding up the execution.



        Something like this:



        DECLARE @tinDetailsId int;

        SELECT @tinDetailsId = et.EntityTypeID
        FROM entitytype et
        WHERE et.EntityName = 'TIN_Details';


        Which you could then reference in the bottom half of the query, eliminating the join:



        select distinct dg1.EntityId, et.EntityName, dg1.version
        from datagathering dg1 with(nolock)
        inner join entity e with(nolock)
        on e.EntityId = dg1.EntityId
        where dg1.EntityId = dg1.RootId
        and e.EntityTypeID = @tinDetailsId
        and dg1.EntityId not in (
        select distinct ChildEntityId
        from entitymapping
        where ChildEntityId = dg1.EntityId
        and ParentEntityId = -1)


        You would want to do the same thing with EntityName "Account_Third_Party_Details" in the top part of the query, as it has the same problem - with twice as many rows.



        PS: Totally unrelated to the topic at hand, I noticed that you have nolock hints on all the tables in this query. Make sure that you are aware of the implications of this. Check out this nifty blog posts on the topic:



        Bad habits : Putting NOLOCK everywhere by Aaron Bertrand
        The Read Uncommitted Isolation Level by Paul White







        share|improve this answer














        share|improve this answer



        share|improve this answer








        edited 3 hours ago

























        answered 3 hours ago









        Josh DarnellJosh Darnell

        7,37522241




        7,37522241






























            draft saved

            draft discarded




















































            Thanks for contributing an answer to Database Administrators Stack Exchange!


            • Please be sure to answer the question. Provide details and share your research!

            But avoid



            • Asking for help, clarification, or responding to other answers.

            • Making statements based on opinion; back them up with references or personal experience.


            To learn more, see our tips on writing great answers.




            draft saved


            draft discarded














            StackExchange.ready(
            function () {
            StackExchange.openid.initPostLogin('.new-post-login', 'https%3a%2f%2fdba.stackexchange.com%2fquestions%2f233536%2fsql-server-2016-excessive-memory-grant-warning-on-poor-performing-query%23new-answer', 'question_page');
            }
            );

            Post as a guest















            Required, but never shown





















































            Required, but never shown














            Required, but never shown












            Required, but never shown







            Required, but never shown

































            Required, but never shown














            Required, but never shown












            Required, but never shown







            Required, but never shown







            Popular posts from this blog

            Can't compile dgruyter and caption packagesLaTeX templates/packages for writing a patent specificationLatex...

            Schneeberg (Smreczany) Bibliografia | Menu...

            Hans Bellmer Spis treści Życiorys | Upamiętnienie | Przypisy | Bibliografia | Linki zewnętrzne |...