I recently removed 3 joins from a linq query against several Entities and used associations between the tables instead.
It seems that the queries now take longer to execute.
I made the changes hoping for some performance improvements.
Is this consistent with what others have observed?
Is there a logical reason for this?
Is there a good rule of thumb about which strategy to use?
Linq query with joins:
from n in db.Names
.Where(n.ID = "myID")
**join d in db.InstitutionDemographics on n.ID equals d.ID
join m in db.MemberAdhocIds on n.ID equals m.ID
join ba in db.NameAddresses on n.BillingAddressNumber equals ba.AddressNumber
join sa in db.NameAddresses on n.ShippingAddressNumber equals sa.AddressNumber
join ma in db.NameAddresses on n.MailingAddressNumber equals ma.AddressNumber
join mas in db.MemberAdhocServices on n.ID equals mas.InstitutionID**
join pn in db.PrimaryContactNames on n.ID equals pn.ID into PrimeName
from pn in PrimeName.DefaultIfEmpty()
join cd in currentDuesResults on n.ID equals cd.ID into DuesJoin
from cd in DuesJoin.DefaultIfEmpty()
join bd in billingDateResults on n.ID equals bd.ID
into BillingDate
from bd in BillingDate.DefaultIfEmpty()
The ** marked section is what was replaced with associations in the EDMX file. The relationships are 1 to1 and constraints are based on the IDs just like the joins.
I have to also mention that there are no FK relationships on the SQL tables. (legacy stuff that I can not change)
** New information
I've captured the SQL generated by the query using join -vs- associations and run it from SSMS. The performance is almost identical. So it appears that the sql generated is not the issue but something else in how the linq code is working.
Related
I'm trying to rewrite SQL procedure to Linq, it all went well and works fine, as long as it works on small data set. I couldn't really find answer to this anywhere. Thing is, I have 3 joins in the query, 2 are left joins and 1 is inner join, they all join to each other/like a tree. Below you can see SQL procedure:
SELECT ...
FROM sprawa s (NOLOCK)
LEFT JOIN strona st (NOLOCK) on s.ident = st.id_sprawy
INNER JOIN stan_szczegoly ss (NOLOCK) on s.kod_stanu = ss.kod_stanu
LEFT JOIN broni b (NOLOCK) on b.id_strony = st.ident
What I'd like to ask you is a way to translate this to Linq. For now I have this:
var queryOne = from s in db.sprawa
join st in db.strona on s.ident equals st.id_sprawy into tmp1
from st2 in tmp1.DefaultIfEmpty()
join ss in db.stan_szczegoly on s.kod_stanu equals ss.kod_stanu
join b in db.broni on st2.ident equals b.id_strony into tmp2
from b2 in tmp2.DefaultIfEmpty()
select new { };
Seems alright, but when checked with SQL Profiler, query that is sent to database looks like that:
SELECT ... FROM [dbo].[sprawa] AS [Extent1]
LEFT OUTER JOIN [dbo].[strona] AS [Extent2]
ON [Extent1].[ident] = [Extent2].[id_sprawy]
INNER JOIN [dbo].[stan_szczegoly] AS [Extent3]
ON [Extent1].[kod_stanu] = [Extent3].[kod_stanu]
INNER JOIN [dbo].[broni] AS [Extent4]
ON ([Extent2].[ident] = [Extent4].[id_strony]) OR
(([Extent2].[ident] IS NULL) AND ([Extent4].[id_strony] IS NULL))
As you can see both SQL queries are bit different. Effect is the same, but latter works incomparably slower (less than a second to over 30 minutes). There's also a union made, but it shouldn't be the problem. If asked for I'll paste code for it.
I'd be grateful for any advice on how to better the performance of my Linq statement or how to write it in a way that is translated properly.
I guess I found the solution:
var queryOne = from s in db.sprawa
join st in db.strona on s.ident equals st.id_sprawy into tmp1
where tmp1.Any()
from st2 in tmp1.DefaultIfEmpty()
join ss in db.stan_szczegoly on s.kod_stanu equals ss.kod_stanu
join b in db.broni on st2.ident equals b.id_strony into tmp2
where tmp2.Any()
from b2 in tmp2.DefaultIfEmpty()
select new { };
In other words where table.Any() after each into table statement. It doesn't make translation any better but has sped up execution time from nearly 30minutes(!) to about 5 seconds.
This has to be used carefully though, because it MAY lead to losing some records in result set.
I have two classes: Property and PropertyValue. A property has several values where each value is a new revision.
When retrieving a set of properties I want to include the latest revision of the value for each property.
in T-SQL this can very efficiently be done like this:
SELECT
p.Id,
pv1.StringValue,
pv1.Revision
FROM dbo.PropertyValues pv1
LEFT JOIN dbo.PropertyValues pv2 ON pv1.Property_Id = pv2.Property_Id AND pv1.Revision < pv2.Revision
JOIN dbo.Properties p ON p.Id = pv1.Property_Id
WHERE pv2.Id IS NULL
ORDER BY p.Id
The "magic" in this query is to join on the lesser than condition and look for rows without a result forced by the LEFT JOIN.
How can I accomplish something similar using LINQ to EF?
The best thing I could come up with was:
from pv in context.PropertyValues
group pv by pv.Property into g
select g.OrderByDescending(p => p.Revision).FirstOrDefault()
It does produce the correct result but is about 10 times slower than the other.
Maybe this can help. Where db is the database context:
(
from pv1 in db.PropertyValues
from pv2 in db.PropertyValues.Where(a=>a.Property_Id==pv1.Property_Id && pv1.Revision<pv2.Revision).DefaultIfEmpty()
join p in db.Properties
on pv1.Property_Id equals p.Id
where pv2.Id==null
orderby p.Id
select new
{
p.Id,
pv1.StringValue,
pv1.Revision
}
);
Next to optimizing a query in Linq To Entities, you also have to be aware of the work it takes for the Entity Framework to translate your query to SQL and then map the results back to your objects.
Comparing a Linq To Entities query directly to a SQL query will always result in lower performance because the Entity Framework does a lot more work for you.
So it's also important to look at optimizing the steps the Entity Framework takes.
Things that could help:
Precompile your query
Pre-generate views
Decide for yourself when to open the database connection
Disable tracking (if appropriate)
Here you can find some documentation with performance strategies.
if you want to use multiple conditions (less than expression) in join you can do this like
from pv1 in db.PropertyValues
join pv2 in db.PropertyValues on new{pv1.Property_ID, Condition = pv1.Revision < pv2.Revision} equals new {pv2.Property_ID , Condition = true} into temp
from t in temp.DefaultIfEmpty()
join p in db.Properties
on pv1.Property_Id equals p.Id
where t.Id==null
orderby p.Id
select new
{
p.Id,
pv1.StringValue,
pv1.Revision
}
Dim ds = From a In db.Model
Join b In db.1 On a.id Equals b.ID
Join c In db.2 On a.id Equals c.ID
Join d In db.3 On a.id Equals d.ID
Join f In db.4 On a.id Equals f.ID
Select a.id, a.Ref, a.Type, a.etc
Above is my linq query. At the moment I am only getting the first row from the db returned when there are currently 60 rows. Please can you tell me where I am going wrong and how to select all records.
Thanks in advance!
UPDATE:
When I take out all the joins like so:
Dim ds = From a In db.1, b In db.2, c In db.3, d In db.4, f In db.5
Select a.id, a.Ref, a.type, b.etc, c.etc, d.etc
I get a system.outofmemory exception!
You're only going to get a row produced when all of the joins match - in other words, when there's a row from Model with an AP, an Option, a Talk and an Invoice. My guess is that there's only one of those.
LINQ does an inner join by default. If you're looking for a left outer join (i.e. where a particular row may not have an Invoice, or a Talk etc) then you need to use a group join, usually in conjunction with DefaultIfEmpty.
I'm not particularly hot on VB syntax, but this article looks like it's what you're after.
I want to achieve the following in Linq to Entities:
Get all Enquires that have no Application or the Application has a status != 4 (Completed)
select e.*
from Enquiry enq
left outer join Application app
on enq.enquiryid = app.enquiryid
where app.Status <> 4 or app.enquiryid is null
Has anyone done this before without using DefaultIfEmpty(), which is not supported by Linq to Entities?
I'm trying to add a filter to an IQueryable query like this:
IQueryable<Enquiry> query = Context.EnquirySet;
query = (from e in query
where e.Applications.DefaultIfEmpty()
.Where(app=>app.Status != 4).Count() >= 1
select e);
Thanks
Mark
In EF 4.0+, LEFT JOIN syntax is a little different and presents a crazy quirk:
var query = from c1 in db.Category
join c2 in db.Category on c1.CategoryID equals c2.ParentCategoryID
into ChildCategory
from cc in ChildCategory.DefaultIfEmpty()
select new CategoryObject
{
CategoryID = c1.CategoryID,
ChildName = cc.CategoryName
}
If you capture the execution of this query in SQL Server Profiler, you will see that it does indeed perform a LEFT OUTER JOIN. HOWEVER, if you have multiple LEFT JOIN ("Group Join") clauses in your Linq-to-Entity query, I have found that the self-join clause MAY actually execute as in INNER JOIN - EVEN IF THE ABOVE SYNTAX IS USED!
The resolution to that? As crazy and, according to MS, wrong as it sounds, I resolved this by changing the order of the join clauses. If the self-referencing LEFT JOIN clause was the 1st Linq Group Join, SQL Profiler reported an INNER JOIN. If the self-referencing LEFT JOIN clause was the LAST Linq Group Join, SQL Profiler reported an LEFT JOIN.
Do this:
IQueryable<Enquiry> query = Context.EnquirySet;
query = (from e in query
where (!e.Applications.Any())
|| e.Applications.Any(app => app.Status != 4)
select e);
I don't find LINQ's handling of the problem of what would be an "outer join" in SQL "goofy" at all. The key to understanding it is to think in terms of an object graph with nullable properties rather than a tabular result set.
Any() maps to EXISTS in SQL, so it's far more efficient than Count() in some cases.
Thanks guys for your help. I went for this option in the end but your solutions have helped broaden my knowledge.
IQueryable<Enquiry> query = Context.EnquirySet;
query = query.Except(from e in query
from a in e.Applications
where a.Status == 4
select e);
Because of Linq's goofy (read non-standard) way of handling outers, you have to use DefaultIfEmpty().
What you'll do is run your Linq-To-Entities query into two IEnumerables, then LEFT Join them using DefaultIfEmpty(). It may look something like:
IQueryable enq = Enquiry.Select();
IQueryable app = Application.Select();
var x = from e in enq
join a in app on e.enquiryid equals a.enquiryid
into ae
where e.Status != 4
from appEnq in ae.DefaultIfEmpty()
select e.*;
Just because you can't do it with Linq-To-Entities doesn't mean you can't do it with raw Linq.
(Note: before anyone downvotes me ... yes, I know there are more elegant ways to do this. I'm just trying to make it understandable. It's the concept that's important, right?)
Another thing to consider, if you directly reference any properties in your where clause from a left-joined group (using the into syntax) without checking for null, Entity Framework will still convert your LEFT JOIN into an INNER JOIN.
To avoid this, filter on the "from x in leftJoinedExtent" part of your query like so:
var y = from parent in thing
join child in subthing on parent.ID equals child.ParentID into childTemp
from childLJ in childTemp.Where(c => c.Visible == true).DefaultIfEmpty()
where parent.ID == 123
select new {
ParentID = parent.ID,
ChildID = childLJ.ID
};
ChildID in the anonymous type will be a nullable type and the query this generates will be a LEFT JOIN.
This question follows on from a question I asked yesterday about why using the join query on my Entities produced horrendously complicated SQL. It seemed that performing a query like this:
var query = from ev in genesisContext.Events
join pe in genesisContext.People_Event_Link
on ev equals pe.Event
where pe.P_ID == key
select ev;
Produced the horrible SQL that took 18 seconds to run on the database, whereas joining the entities through a where clause (sort of like pre-ANSI SQL syntax) took less than a second to run and produced the same result
var query = from pe in genesisContext.People_Event_Link
from ev in genesisContext.Events
where pe.P_ID == key && pe.Event == ev
select ev;
I've googled all over but still don't understand why the second is produces different SQL to the first. Can someone please explain the difference to me? When should I use the join keyword
This is the SQL that was produced when I used Join in my query and took 18 seconds to run:
SELECT
1 AS [C1],
[Extent1].[E_ID] AS [E_ID],
[Extent1].[E_START_DATE] AS [E_START_DATE],
[Extent1].[E_END_DATE] AS [E_END_DATE],
[Extent1].[E_COMMENTS] AS [E_COMMENTS],
[Extent1].[E_DATE_ADDED] AS [E_DATE_ADDED],
[Extent1].[E_RECORDED_BY] AS [E_RECORDED_BY],
[Extent1].[E_DATE_UPDATED] AS [E_DATE_UPDATED],
[Extent1].[E_UPDATED_BY] AS [E_UPDATED_BY],
[Extent1].[ET_ID] AS [ET_ID],
[Extent1].[L_ID] AS [L_ID]
FROM [dbo].[Events] AS [Extent1]
INNER JOIN [dbo].[People_Event_Link] AS [Extent2] ON EXISTS (SELECT
1 AS [C1]
FROM ( SELECT 1 AS X ) AS [SingleRowTable1]
LEFT OUTER JOIN (SELECT
[Extent3].[E_ID] AS [E_ID]
FROM [dbo].[Events] AS [Extent3]
WHERE [Extent2].[E_ID] = [Extent3].[E_ID] ) AS [Project1] ON 1 = 1
LEFT OUTER JOIN (SELECT
[Extent4].[E_ID] AS [E_ID]
FROM [dbo].[Events] AS [Extent4]
WHERE [Extent2].[E_ID] = [Extent4].[E_ID] ) AS [Project2] ON 1 = 1
WHERE ([Extent1].[E_ID] = [Project1].[E_ID]) OR (([Extent1].[E_ID] IS NULL) AND ([Project2].[E_ID] IS NULL))
)
WHERE [Extent2].[P_ID] = 291
This is the SQL that was produce using the ANSI Style syntax (and is fairly close to what I would write if I were writing the SQL myself):
SELECT * FROM Events AS E INNER JOIN People_Event_Link AS PE ON E.E_ID=PE.E_ID INNER JOIN PEOPLE AS P ON P.P_ID=PE.P_ID
WHERE P.P_ID = 291
Neither of the above queries are entirely "correct." In EF, it is generally correct to use the relationship properties in lieu of either of the above. For example, if you had a Person object with a one to many relationship to PhoneNumbers in a property called Person.PhoneNumbers, you could write:
var q = from p in Context.Person
from pn in p.PhoneNumbers
select pn;
The EF will build the join for you.
In terms of the question above, the reason the generated SQL is different is because the expression trees are different, even though they produce equivalent results. Expression trees are mapped to SQL, and you of course know that you can write different SQL which produces the same results but with different performance. The mapping is designed to produce decent SQL when you write a farily "conventional" EF query.
But the mapping is not so smart as to take a very unconventional query and optimize it. In your first query, you state that the objects must be equivalent. In the second, you state that the ID property must be equivalent. My sample query above says "just get the details for this one record." The EF is designed to work with the way I show, principally, but also handles scalar equivalence well.