EntityFramework – contains query of composite key

This is a nasty problem for which I don’t know any elegant solution.

Suppose you have these key combinations, and you only want to select the marked ones (*).

Id1  Id2
---  ---
1    2 *
1    3
1    6
2    2 *
2    3 *
... (many more)

How to do this is a way that Entity Framework is happy? Let’s look at some possible solutions and see if they’re any good.

Solution 1: Join (or Contains) with pairs

The best solution would be to create a list of the pairs you want, for instance Tuples, (List<Tuple<int,int>>) and join the database data with this list:

from entity in db.Table // db is a DbContext
join pair in Tuples on new { entity.Id1, entity.Id2 }
                equals new { Id1 = pair.Item1, Id2 = pair.Item2 }
select entity

In LINQ to objects this would be perfect, but, too bad, EF will throw an exception like

Unable to create a constant value of type ‘System.Tuple`2 (…) Only primitive types or enumeration types are supported in this context.

which is a rather clumsy way to tell you that it can’t translate this statement into SQL, because Tuples is not a list of primitive values (like int or string).¹. For the same reason a similar statement using Contains (or any other LINQ statement) would fail.

Solution 2: In-memory

Of course we could turn the problem into simple LINQ to objects like so:

from entity in db.Table.AsEnumerable() // fetch db.Table into memory first
join pair Tuples on new { entity.Id1, entity.Id2 }
             equals new { Id1 = pair.Item1, Id2 = pair.Item2 }
select entity

Needless to say that this is not a good solution. db.Table could contain millions of records.

Solution 3: Two Contains statements

So let’s offer EF two lists of primitive values, [1,2] for Id1 and [2,3] for Id2. We don’t want to use join (see side note), so let’s use Contains:

from entity in db.Table
where ids1.Contains(entity.Id1) && ids2.Contains(entity.Id2)
select entity

But now the results also contains entity {1,3}! Well, of course, this entity perfectly matches the two predicates. But let’s keep in mind that we’re getting closer. In stead of pulling millions of entities into memory, we now only get four of them.

Solution 4: One Contains with computed values

Solution 3 failed because the two separate Contains statements don’t only filter the combinations of their values. What if we create a list of combinations first and try to match these combinations? We know from solution 1 that this list should contain primitive values. For instance:

var computed = ids1.Zip(ids2, (i1,i2) => i1 * i2); // [2,6]

and the LINQ statement:

from entity in db.Table
where computed.Contains(entity.Id1 * entity.Id2)
select entity

There are some problems with this approach. First, you’ll see that this also returns entity {1,6}. The combination function (a*b) does not produce values that uniquely identify a pair in the database. Now we could create a list of strings like ["Id1=1,Id2=2","Id1=2,Id2=3]" and do

from entity in db.Table
where computed.Contains("Id1=" + entity.Id1 + "," + "Id2=" + entity.Id2)
select entity

(This would work in EF6, not in earlier versions).

This is getting pretty messy. But a more important problem is that this solution is not sargable, which means: it bypasses any database indexes on Id1 and Id2 that could have been used otherwise. This will perform very very poorly.

Solution 5: Best of 2 and 3

So the only viable solution I can think of is a combination of Contains and a join in memory: First do the contains statement as in solution 3. Remember, it got us very close to what we wanted. Then refine the query result by joining the result as an in-memory list:

var rawSelection = from entity in db.Table
                   where ids1.Contains(entity.Id1) && ids2.Contains(entity.Id2)
                   select entity;

var refined = from entity in rawSelection.AsEnumerable()
              join pair in Tuples on new { entity.Id1, entity.Id2 }
                              equals new { Id1 = pair.Item1, Id2 = pair.Item2 }
              select entity;

It’s not elegant, messy all the same maybe, but so far it’s the only scalable² solution to this problem I found, and applied in my own code.

Solution 6: Build a query with OR clauses

Using a Predicate builder like Linqkit or alternatives, you can build a query that contains an OR clause for each element in the list of combinations. This could be a viable option for really short lists. With a couple of hundreds of elements, the query will start performing very poorly. So I don’t consider this a good solution unless you can be 100% sure that there will always be a small number of elements. One elaboration of this option can be found here.

---

¹As a funny side note, EF does create a SQL statement when you join a primitive list, like so

from entity in db.Table // db is a DbContext
join i in MyIntegers on entity.Id1 equals i
select entity

But the generated SQL is, well, absurd. A real-life example where MyIntegers contains only 5(!) integers looks like this:

SELECT 
    [Extent1].[CmpId] AS [CmpId], 
    [Extent1].[Name] AS [Name], 
    FROM  [dbo].[Company] AS [Extent1]
    INNER JOIN  (SELECT 
        [UnionAll3].[C1] AS [C1]
        FROM  (SELECT 
            [UnionAll2].[C1] AS [C1]
            FROM  (SELECT 
                [UnionAll1].[C1] AS [C1]
                FROM  (SELECT 
                    1 AS [C1]
                    FROM  ( SELECT 1 AS X ) AS [SingleRowTable1]
                UNION ALL
                    SELECT 
                    2 AS [C1]
                    FROM  ( SELECT 1 AS X ) AS [SingleRowTable2]) AS [UnionAll1]
            UNION ALL
                SELECT 
                3 AS [C1]
                FROM  ( SELECT 1 AS X ) AS [SingleRowTable3]) AS [UnionAll2]
        UNION ALL
            SELECT 
            4 AS [C1]
            FROM  ( SELECT 1 AS X ) AS [SingleRowTable4]) AS [UnionAll3]
    UNION ALL
        SELECT 
        5 AS [C1]
        FROM  ( SELECT 1 AS X ) AS [SingleRowTable5]) AS [UnionAll4] ON [Extent1].[CmpId] = [UnionAll4].[C1]

There are n-1 UNIONs. Of course that’s not scalable at all.

Later addition:
Somewhere along the road to EF version 6.1.3 this has been greatly improved. The UNIONs have become simpler and they are no longer nested. Previously the query would give up with less than 50 elements in the local sequence (SQL exception: Some part of your SQL statement is nested too deeply.) The non-nested UNION allow local sequences up to a couple of thousands(!) of elements. It’s still slow though with “many” elements.

²As far as the Contains statement is scalable: Scalable Contains method for LINQ against a SQL backend