Ayende @ Rahien

Hi!
My name is Ayende Rahien
Founder of Hibernating Rhinos LTD and RavenDB.
You can reach me by phone or email:

ayende@ayende.com

+972 52-548-6969

@

Posts: 5,947 | Comments: 44,541

filter by tags archive

ChallengePremature optimization, and all of that…


Let us look at the following pieces of code:

public void Consume(MyBooksRequest message)
{
    var user = session.Get<User>(message.UserId);
    
    bus.Reply(new MyBooksResponse
    {
        UserId = message.UserId,
        Timestamp = DateTime.Now,
        Books = user.CurrentlyReading.ToBookDtoArray()
    });
}

public void Consume(MyQueueRequest message)
{
    var user = session.Get<User>(message.UserId);

    bus.Reply(new MyQueueResponse
    {
        UserId = message.UserId,
        Timestamp = DateTime.Now,
        Queue = user.Queue.ToBookDtoArray()
    });
}

public void Consume(MyRecommendationsRequest message)
{
    var user = session.Get<User>(message.UserId);

    bus.Reply(new MyRecommendationsResponse
    {
        UserId = message.UserId,
        Timestamp = DateTime.Now,
        Recommendations = user.Recommendations.ToBookDtoArray()
    });
}

Looking at this, I see that I have a requirement to getting my books, my queues and my recommendations. It appears that getting each datum is going to result in 2 queries, the first to load the User, and the second to lazy load the actual collection that we want to return.

An almost trivial optimization would be to eliminate the lazy loading, right? That would reduce the cost from 6 queries to just 3.

However, that assumption would be wrong. The following client code:

bus.Send(
    new MyBooksRequest
    {
        UserId = userId
    },
    new MyQueueRequest
    {
        UserId = userId
    },
    new MyRecommendationsRequest
    {
        UserId = userId
    });

Produces this SQL:

-- statement #1
enlisted session in distributed transaction with isolation level: Serializable

-- statement #2
SELECT user0_.Id          as Id2_0_,
       user0_.Name        as Name2_0_,
       user0_.Street      as Street2_0_,
       user0_.Country     as Country2_0_,
       user0_.City        as City2_0_,
       user0_.ZipCode     as ZipCode2_0_,
       user0_.HouseNumber as HouseNum7_2_0_
FROM   Users user0_
WHERE  user0_.Id = 1 /* @p0 */

-- statement #3
SELECT currentlyr0_.[User] as User1_1_,
       currentlyr0_.Book   as Book1_,
       book1_.Id           as Id0_0_,
       book1_.Name         as Name0_0_,
       book1_.ImageUrl     as ImageUrl0_0_,
       book1_.Image        as Image0_0_,
       book1_.Author       as Author0_0_
FROM   UsersReadingBooks currentlyr0_
       left outer join Books book1_
         on currentlyr0_.Book = book1_.Id
WHERE  currentlyr0_.[User] = 1 /* @p0 */

-- statement #4
SELECT queue0_.[User]  as User1_1_,
       queue0_.Book    as Book1_,
       queue0_.[Index] as Index3_1_,
       book1_.Id       as Id0_0_,
       book1_.Name     as Name0_0_,
       book1_.ImageUrl as ImageUrl0_0_,
       book1_.Image    as Image0_0_,
       book1_.Author   as Author0_0_
FROM   UsersWaitingBooks queue0_
       left outer join Books book1_
         on queue0_.Book = book1_.Id
WHERE  queue0_.[User] = 1 /* @p0 */

-- statement #5
SELECT recommenda0_.[User]  as User1_1_,
       recommenda0_.Book    as Book1_,
       recommenda0_.[Index] as Index3_1_,
       book1_.Id            as Id0_0_,
       book1_.Name          as Name0_0_,
       book1_.ImageUrl      as ImageUrl0_0_,
       book1_.Image         as Image0_0_,
       book1_.Author        as Author0_0_
FROM   UsersRecommendedBooks recommenda0_
       left outer join Books book1_
         on recommenda0_.Book = book1_.Id
WHERE  recommenda0_.[User] = 1 /* @p0 */


-- statement #7
commit transaction

That seems strange, can you figure out why?

Bonus points for figuring out whatever it would be worth it to do the eager load optimization or not.

More posts in "Challenge" series:

  1. (28 Apr 2015) What is the meaning of this change?
  2. (26 Sep 2013) Spot the bug
  3. (27 May 2013) The problem of locking down tasks…
  4. (17 Oct 2011) Minimum number of round trips
  5. (23 Aug 2011) Recent Comments with Future Posts
  6. (02 Aug 2011) Modifying execution approaches
  7. (29 Apr 2011) Stop the leaks
  8. (23 Dec 2010) This code should never hit production
  9. (17 Dec 2010) Your own ThreadLocal
  10. (03 Dec 2010) Querying relative information with RavenDB
  11. (29 Jun 2010) Find the bug
  12. (23 Jun 2010) Dynamically dynamic
  13. (28 Apr 2010) What killed the application?
  14. (19 Mar 2010) What does this code do?
  15. (04 Mar 2010) Robust enumeration over external code
  16. (16 Feb 2010) Premature optimization, and all of that…
  17. (12 Feb 2010) Efficient querying
  18. (10 Feb 2010) Find the resource leak
  19. (21 Oct 2009) Can you spot the bug?
  20. (18 Oct 2009) Why is this wrong?
  21. (17 Oct 2009) Write the check in comment
  22. (15 Sep 2009) NH Prof Exporting Reports
  23. (02 Sep 2009) The lazy loaded inheritance many to one association OR/M conundrum
  24. (01 Sep 2009) Why isn’t select broken?
  25. (06 Aug 2009) Find the bug fixes
  26. (26 May 2009) Find the bug
  27. (14 May 2009) multi threaded test failure
  28. (11 May 2009) The regex that doesn’t match
  29. (24 Mar 2009) probability based selection
  30. (13 Mar 2009) C# Rewriting
  31. (18 Feb 2009) write a self extracting program
  32. (04 Sep 2008) Don't stop with the first DSL abstraction
  33. (02 Aug 2008) What is the problem?
  34. (28 Jul 2008) What does this code do?
  35. (26 Jul 2008) Find the bug fix
  36. (05 Jul 2008) Find the deadlock
  37. (03 Jul 2008) Find the bug
  38. (02 Jul 2008) What is wrong with this code
  39. (05 Jun 2008) why did the tests fail?
  40. (27 May 2008) Striving for better syntax
  41. (13 Apr 2008) calling generics without the generic type
  42. (12 Apr 2008) The directory tree
  43. (24 Mar 2008) Find the version
  44. (21 Jan 2008) Strongly typing weakly typed code
  45. (28 Jun 2007) Windsor Null Object Dependency Facility

Comments

Rafal

Looks like RSB handles message batches in a single transaction. Is it a bug or feature? BTW, serializable isolation level is almost guaranteed to cause problems in real world application.

Ayende Rahien

Rafal,

Yes, that IS a feature.

And serializable isolation is actually a good thing in some circumstances. For one, it make it really simple to think about concurrency, and if all our data is user scoped, we aren't going to deal with a lot of complexity either.

Rafal

Ah, I thought that sending an array of message is there only for performance reasons or developer convenience, didn't know about this little side effect.

Felix

Uhmm,

If the session crunching the messages is the same for all the messages in this example, then the first level cache is already an optimization. Eagerly fetch more then one collection at a time may results in a cartesian product query, that is worst than executing the 3 query to load the collection when needed.

Jan Willem B

Since one session is used, calls to session.Get <user(message.UserId) will only hit the database once.

The first time this is done, the books, queues and collections are eagerly loaded (if specified, you can't tell from the queries), but in a seperate select statements. I would not set fetchmode to join because this will lead to cartesian product. Further optimization could be to use a multicriteriaquery to send the three queries at once. But that would be an MS SQL specific optimization.

simone basso

may be grouping all those queries into the same transaction might cause some more deadlocks?

Ayende Rahien

Simone,

Can you envision a way in which two concurrent transactions (even for the same user) can deadlock with this code?

simone basso - smnbss

The problem in fact is with the type of transaction you're using

A transaction with Serializable Isolation places a range lock on the data set, preventing other users from updating or inserting rows into the data set until the transaction is complete.

Grouping those 4 queries on the same transaction will eventually put some more stress to the database if used in conjunction with a serializable transaction .

here I wrote a little bit more :)

http://smnbss.spaces.live.com/blog/cns!A117AA5E007A0648!1894.entry?&c02vws=1

Ayende Rahien

Simone,

All the data that I am accessing is local to the current user

simone basso

Sure, in your specific case it might not be a problem, but if you have other services accessing the same data with other queries it might create some problems. BTW, what is the problem then with that query? :)

Comment preview

Comments have been closed on this topic.

FUTURE POSTS

No future posts left, oh my!

RECENT SERIES

  1. RavenDB Sharding (3):
    22 May 2015 - Adding a new shard to an existing cluster, splitting the shard
  2. The RavenDB Comic Strip (2):
    20 May 2015 - Part II – a team in trouble!
  3. Challenge (45):
    28 Apr 2015 - What is the meaning of this change?
  4. Interview question (2):
    30 Mar 2015 - fix the index
  5. Excerpts from the RavenDB Performance team report (20):
    20 Feb 2015 - Optimizing Compare – The circle of life (a post-mortem)
View all series

RECENT COMMENTS

Syndication

Main feed Feed Stats
Comments feed   Comments Feed Stats