表加入Entity Framework的性能问题

Question

连接两个表导致选择时间从330秒增加到40秒.要加入的表仅包含ID和文本.加入两张桌子时,我没想到选择时间会增加8倍.我的JOIN有什么问题,或者这是正常的SQL Server行为吗？

主表填充了3500万条记录,以查看在达到10 GB的SQL Server Express限制时它的执行情况.在字段LogTimeStamp和字段LogType上创建了另一个索引.

连接表的内容:

var queryList = messages
    .Join(types,
    type => type.LogType,
    typeText => typeText.LogType, (msg, msgType) => new
    {
        msg.LogID,
        msg.LogTimeStamp,
        msg.LogUser,
        msg.LogType,
        msgType.LogTypeName,
        msg.LogMessage
    })
    .Where(t => System.Data.Entity.DbFunctions.TruncateTime(t.LogTimeStamp) >= fromDate)
    .Where(t => System.Data.Entity.DbFunctions.TruncateTime(t.LogTimeStamp) <= toDate)
    .Where(t => t.LogType != 4)
    .OrderBy(m => m.LogID)
    .ToList();

结果SQL

SELECT 
  1 AS [C1], 
  [Extent1].[LogID] AS [LogID], 
  [Extent1].[LogTimeStamp] AS [LogTimeStamp], 
  [Extent1].[LogUser] AS [LogUser], 
  [Extent1].[LogType] AS [LogType], 
  [Extent2].[LogTypeName] AS [LogTypeName], 
  [Extent1].[LogMessage] AS [LogMessage]
  FROM  [dbo].[AuditTrailMessages] AS [Extent1]
  INNER JOIN [dbo].[AuditTrailLogTypes] AS [Extent2] ON [Extent1].[LogType] = [Extent2].[LogType]
WHERE ((convert (datetime2, convert(varchar(255), [Extent1].[LogTimeStamp], 102) ,  102)) >= @p__linq__0) 
  AND ((convert (datetime2, convert(varchar(255), [Extent1].[LogTimeStamp], 102) ,  102)) <= @p__linq__1) 
  AND ( NOT ((4 =  CAST( [Extent1].[LogType] AS int)) AND ( CAST( [Extent1].[LogType] AS int) IS NOT NULL)))  

相比

var queryList = messages
    .Where(t => System.Data.Entity.DbFunctions.TruncateTime(t.LogTimeStamp) >= fromDate)
    .Where(t => System.Data.Entity.DbFunctions.TruncateTime(t.LogTimeStamp) <= toDate)
    .Where(t => t.LogType != 4)
    .OrderBy(m => m.LogID)
    .ToList();

结果SQL

SELECT 
  [Extent1].[LogID] AS [LogID], 
  [Extent1].[LogTimeStamp] AS [LogTimeStamp], 
  [Extent1].[LogUser] AS [LogUser], 
  [Extent1].[LogMessage] AS [LogMessage], 
  [Extent1].[LogType] AS [LogType]
  FROM [dbo].[AuditTrailMessages] AS [Extent1]
  WHERE ((convert (datetime2, convert(varchar(255), [Extent1].[LogTimeStamp], 102) ,  102)) >= @p__linq__0) 
  AND ((convert (datetime2, convert(varchar(255), [Extent1].[LogTimeStamp], 102) ,  102)) <= @p__linq__1) 
  AND ( NOT ((4 =  CAST( [Extent1].[LogType] AS int)) AND ( CAST( [Extent1].[LogType] AS int) IS NOT NULL)))

Answer 1

你看到了所有这些:

((convert(datetime2,convert(varchar(255),[Extent1].[LogTimeStamp],102)

他们很坏.特别糟糕.他们基本上说"不要使用索引,进行全表扫描".

你可以做到:

t => System.Data.Entity.DbFunctions.TruncateTime(t.LogTimeStamp)> = fromDate

这不好.这不是必需的.日期中的任何时间戳都大于或等于每个定义的日期,并且小于下一个日期.

所以:

t => System.Data.Entity.DbFunctions.TruncateTime(t.LogTimeStamp)> = fromDate

变成

t => t.LogTimeStamp> = fromDate

和

t => System.Data.Entity.DbFunctions.TruncateTime(t.LogTimeStamp)<= toDate

变成

t => t.LogTimeStamp <toDate.AddDays(1)

.Where(t => t.LogType!= 4)

看起来像一个类型不匹配 - 让我猜,它不是数据库中的int.然后使用Equals方法.这是EF中的已知错误.说它应该没关系 - 此时你应该归结为很多条目,你的问题可能是日期时间比较的低效代码.

永远不要在比较的现场做一个功能.决不.它们会终止任何索引使用(除非有一个完全具有此功能的索引).始终重写查询以使所有函数都处于常量.

不是EF问题 - 一般的SQL初学者错误.