实体框架LINQ到实体的联接查询超时

Question

I am executing the following LINQ to Entities query but it is stuck and does not return response until timeout. I executed the same query on SQL Server and it return 92000 in 3 sec.

            var query = (from r in WinCtx.PartsRoutings
                     join s in WinCtx.Tab_Processes on r.ProcessName equals s.ProcessName
                     join p in WinCtx.Tab_Parts on r.CustPartNum equals p.CustPartNum
                     select new { r}).ToList();

SQL Generated:

SELECT [ I omitted columns]
    FROM   [dbo].[PartsRouting] AS [Extent1]
INNER JOIN [dbo].[Tab_Processes] AS [Extent2] ON ([Extent1].[ProcessName] = [Extent2].[ProcessName]) OR (([Extent1].[ProcessName] IS NULL) AND ([Extent2].[ProcessName] IS NULL))
INNER JOIN [dbo].[Tab_Parts] AS [Extent3] ON ([Extent1].[CustPartNum] = [Extent3].[CustPartNum]) OR (([Extent1].[CustPartNum] IS NULL) AND ([Extent3].[CustPartNum] IS NULL))

PartsRouting Table has 100,000+ records, Parts = 15000+, Processes = 200.

I tried too many things found online but nothing worked for me as to how I can achieve the result with same performance of SQL.

Answer 1

Based on the comments, looks like the issue is caused by the additional OR with IS NULL conditions in joins generated by the EF SQL translator. They were added in EF in order to emulate the C# == operator semantics which are different from SQL = for NULL values.

You can start by turning that EF behavior off through UseDatabaseNullSemantics property (it's false by default):

WinCtx.Configuration.UseDatabaseNullSemantics = true;

Unfortunately that's not enough, because it fixes the normal comparison operators, but they simply forgot to do the same for join conditions.

In case you are using joins just for filtering (as it seems), you can replace them with LINQ Any conditions which translates to SQL EXISTS and nowadays database query optimizers are treating it the same way as if it was an inner join:

var query = (from r in WinCtx.PartsRoutings
             where WinCtx.Tab_Processes.Any(s => r.ProcessName == s.ProcessName)
             where WinCtx.Tab_Parts.Any(p => r.CustPartNum == p.CustPartNum)
             select new { r }).ToList();

You might also consider using just select r since creating anonymous type with single property just introdeces additional memory overhead with no advantages.

Update: Looking at the latest comment, you do need fields from joined tables (that's why it's important to not omit relevant parts of the query in question). In such case, you could try the alternative join syntax with where clauses:

WinCtx.Configuration.UseDatabaseNullSemantics = true;
var query = (from r in WinCtx.PartsRoutings
             from s in WinCtx.Tab_Processes where r.ProcessName == s.ProcessName
             from p in WinCtx.Tab_Parts where r.CustPartNum == p.CustPartNum
             select new { r, s.Foo, p.Bar }).ToList();