實體框架和Top and OrderBy的OData性能不佳

Question

我正在使用OData 5.8.0和EntityFramework 6.1.3，查詢：

&$filter=fieldA eq 'ABCDEFG'&$skip=0&$top=10&$orderby=fieldB desc

結果是：

SELECT TOP (10) 
    [Project1].[FieldA] AS [FieldA], 
    [Project1].[FieldB] AS [FieldB], 
    FROM ( SELECT [Project1].[FieldA] AS [FieldA], [Project1].[FieldB] AS [FieldB], row_number() OVER (ORDER BY [Project1].[FieldB] DESC, [Project1].[FieldA] ASC) AS [row_number]
        FROM ( SELECT 
            [Extent1].[FieldA] AS [FieldA], 
            [Extent1].[FieldB] AS [FieldB], 
            FROM [dbo].[table] AS [Extent1]
            WHERE ([Extent1].[FieldA] = 'ABCDEFG') OR (([Extent1].[FieldA] IS NULL) AND ('ABCDEFG' IS NULL))
        )  AS [Project1]
    )  AS [Project1]
    WHERE [Project1].[row_number] > 0
    ORDER BY [Project1].[FieldB] DESC, [Project1].[FieldA] ASC

對於大量的字段A，大約需要20秒才能在DB上運行。

如果我使用相同的LINQ：

var newList = table.Where(f => f.fieldA == 'ABCDEFG').OrderByDescending(f => f.fieldB).Take(10).Skip(0).ToList();

結果是：

SELECT 
    [Limit1].[FieldA] AS [FieldA], 
    [Limit1].[FieldB] AS [FieldB]
    FROM ( SELECT [Limit1].[FieldA] AS [FieldA], [Limit1].[FieldB] AS [FieldB], row_number() OVER (ORDER BY [Limit1].[FieldB] DESC) AS [row_number]
        FROM ( SELECT TOP (10) [Project1].[FieldA] AS [FieldA], [Project1].[FieldB] AS [FieldB]
            FROM ( SELECT 
                [Extent1].[FieldA] AS [FieldA], 
                [Extent1].[FieldB] AS [FieldB]
                FROM [dbo].[table] AS [Extent1]
                WHERE ([Extent1].[FieldA] = 'ABCDEFG') OR (([Extent1].[FieldA] IS NULL) AND ('ABCDEFG' IS NULL))
            )  AS [Project1]
            ORDER BY [Project1].[FieldB] DESC
        )  AS [Limit1]
    )  AS [Limit1]
    WHERE [Limit1].[row_number] > 0
    ORDER BY [Limit1].[FieldB] DES

運行需要120毫秒。

如何強制OData使用相同的表達式（即不在外部語句中使用TOP）？

Answer 1

我發現這個問題是OData不是很聰明，並且以錯誤的順序應用查詢選項。 下面的代碼首先應用orderBy，然后應用頂部：

private static IQueryable<Item> ApplyOptimizedOdataOptions(IQueryable<Item> origQuery,  ODataQueryOptions<Item> options)
{
    var defaultOdataQuerySettings = new ODataQuerySettings();
    if (options.Top != null && options.OrderBy != null)
    {
        // We can optimze this query. Apply the OrderBy first, then Top.
        IQueryable results = options.OrderBy.ApplyTo(origQuery, defaultOdataQuerySettings);
        results = options.Top.ApplyTo(results, defaultOdataQuerySettings);
        results = options.ApplyTo(results, defaultOdataQuerySettings, AllowedQueryOptions.Top | AllowedQueryOptions.OrderBy);

        return results as IQueryable<Item>;
    }

    return options.ApplyTo(origQuery, defaultOdataQuerySettings) as IQueryable<Item>;
}

如果我運行通過使用此IQueryable生成的結果SQL語句：

SET STATISTICS TIME ON;
// Run SQL here
SET STATISTICS TIME OFF;

重新排列這些語句會導致：

SQL Server執行時間：CPU時間= 0毫秒，經過的時間= 1毫秒。

與不重新訂購相比：

SQL Server執行時間：CPU時間= 1213 ms，經過的時間= 20112ms。

速度提升約20,000倍。

Answer 2

實際上，orderby應該比top早應用，以便top穩定並創建場景。運行sql時，這是相同的邏輯，總是應該有一個top帶有orderby，或者按默認順序排列，在這種情況下，您應該只使用top，得到結果后再訂購。