我正在寻找插入实体框架的最快方法。

我之所以问这个问题,是因为您有一个活动的TransactionScope,并且插入量很大(4000+)。它可能会持续超过10分钟(事务的默认超时),这将导致事务不完整。


当前回答

TL;博士我知道这是一个老帖子,但我已经实施了一个解决方案,从其中一个提议开始,扩展它并解决其中的一些问题;此外,我还阅读了所提出的其他解决方案,与这些方案相比,我似乎提出了一种更适合原始问题中提出的要求的解决方案。

在这个解决方案中,我扩展了Slauma的方法,我认为它非常适合原始问题中提出的情况,即使用实体框架和事务范围对数据库进行昂贵的写入操作。

在Slauma的解决方案中,这只是一个草稿,只是用来了解​​EF的速度与实施批量插入的策略-存在以下问题:

交易超时(默认情况下,1分钟可通过代码延长至最多10分钟);复制宽度等于事务结束时使用的提交大小的第一个数据块(这个问题很奇怪,可以通过变通方法解决)。

我还报告了一个例子,其中包括几个从属实体的上下文插入,从而扩展了Slauma提出的案例研究。

我能够验证的性能是10K记录/分钟,在数据库中插入200K宽的记录块,每个记录块大约1KB。速度是恒定的,性能没有下降,测试需要大约20分钟才能成功运行。

详细的解决方案

主持在示例存储库类中插入的批量插入操作的方法:

abstract class SomeRepository { 

    protected MyDbContext myDbContextRef;

    public void ImportData<TChild, TFather>(List<TChild> entities, TFather entityFather)
            where TChild : class, IEntityChild
            where TFather : class, IEntityFather
    {

        using (var scope = MyDbContext.CreateTransactionScope())
        {

            MyDbContext context = null;
            try
            {
                context = new MyDbContext(myDbContextRef.ConnectionString);

                context.Configuration.AutoDetectChangesEnabled = false;

                entityFather.BulkInsertResult = false;
                var fileEntity = context.Set<TFather>().Add(entityFather);
                context.SaveChanges();

                int count = 0;

                //avoids an issue with recreating context: EF duplicates the first commit block of data at the end of transaction!!
                context = MyDbContext.AddToContext<TChild>(context, null, 0, 1, true);

                foreach (var entityToInsert in entities)
                {
                    ++count;
                    entityToInsert.EntityFatherRefId = fileEntity.Id;
                    context = MyDbContext.AddToContext<TChild>(context, entityToInsert, count, 100, true);
                }

                entityFather.BulkInsertResult = true;
                context.Set<TFather>().Add(fileEntity);
                context.Entry<TFather>(fileEntity).State = EntityState.Modified;

                context.SaveChanges();
            }
            finally
            {
                if (context != null)
                    context.Dispose();
            }

            scope.Complete();
        }

    }

}

仅用于示例目的的接口:

public interface IEntityChild {

    //some properties ...

    int EntityFatherRefId { get; set; }

}

public interface IEntityFather {

    int Id { get; set; }
    bool BulkInsertResult { get; set; }
}

db上下文中,我将解决方案的各个元素实现为静态方法:

public class MyDbContext : DbContext
{

    public string ConnectionString { get; set; }


    public MyDbContext(string nameOrConnectionString)
    : base(nameOrConnectionString)
    {
        Database.SetInitializer<MyDbContext>(null);
        ConnectionString = Database.Connection.ConnectionString;
    }


    /// <summary>
    /// Creates a TransactionScope raising timeout transaction to 30 minutes
    /// </summary>
    /// <param name="_isolationLevel"></param>
    /// <param name="timeout"></param>
    /// <remarks>
    /// It is possible to set isolation-level and timeout to different values. Pay close attention managing these 2 transactions working parameters.
    /// <para>Default TransactionScope values for isolation-level and timeout are the following:</para>
    /// <para>Default isolation-level is "Serializable"</para>
    /// <para>Default timeout ranges between 1 minute (default value if not specified a timeout) to max 10 minute (if not changed by code or updating max-timeout machine.config value)</para>
    /// </remarks>
    public static TransactionScope CreateTransactionScope(IsolationLevel _isolationLevel = IsolationLevel.Serializable, TimeSpan? timeout = null)
    {
        SetTransactionManagerField("_cachedMaxTimeout", true);
        SetTransactionManagerField("_maximumTimeout", timeout ?? TimeSpan.FromMinutes(30));

        var transactionOptions = new TransactionOptions();
        transactionOptions.IsolationLevel = _isolationLevel;
        transactionOptions.Timeout = TransactionManager.MaximumTimeout;
        return new TransactionScope(TransactionScopeOption.Required, transactionOptions);
    }

    private static void SetTransactionManagerField(string fieldName, object value)
    {
        typeof(TransactionManager).GetField(fieldName, BindingFlags.NonPublic | BindingFlags.Static).SetValue(null, value);
    }


    /// <summary>
    /// Adds a generic entity to a given context allowing commit on large block of data and improving performance to support db bulk-insert operations based on Entity Framework
    /// </summary>
    /// <typeparam name="T"></typeparam>
    /// <param name="context"></param>
    /// <param name="entity"></param>
    /// <param name="count"></param>
    /// <param name="commitCount">defines the block of data size</param>
    /// <param name="recreateContext"></param>
    /// <returns></returns>
    public static MyDbContext AddToContext<T>(MyDbContext context, T entity, int count, int commitCount, bool recreateContext) where T : class
    {
        if (entity != null)
            context.Set<T>().Add(entity);

        if (count % commitCount == 0)
        {
            context.SaveChanges();
            if (recreateContext)
            {
                var contextConnectionString = context.ConnectionString;
                context.Dispose();
                context = new MyDbContext(contextConnectionString);
                context.Configuration.AutoDetectChangesEnabled = false;
            }
        }

        return context;
    }
}

其他回答

如果您添加的实体()依赖于上下文中的其他预加载实体(例如导航财产),则Dispose()上下文会产生问题

我使用类似的概念来保持我的上下文较小,以实现相同的性能

但我只是分离已经SaveChanges()的实体,而不是Dispose()上下文并重新创建

public void AddAndSave<TEntity>(List<TEntity> entities) where TEntity : class {

const int CommitCount = 1000; //set your own best performance number here
int currentCount = 0;

while (currentCount < entities.Count())
{
    //make sure it don't commit more than the entities you have
    int commitCount = CommitCount;
    if ((entities.Count - currentCount) < commitCount)
        commitCount = entities.Count - currentCount;

    //e.g. Add entities [ i = 0 to 999, 1000 to 1999, ... , n to n+999... ] to conext
    for (int i = currentCount; i < (currentCount + commitCount); i++)        
        _context.Entry(entities[i]).State = System.Data.EntityState.Added;
        //same as calling _context.Set<TEntity>().Add(entities[i]);       

    //commit entities[n to n+999] to database
    _context.SaveChanges();

    //detach all entities in the context that committed to database
    //so it won't overload the context
    for (int i = currentCount; i < (currentCount + commitCount); i++)
        _context.Entry(entities[i]).State = System.Data.EntityState.Detached;

    currentCount += commitCount;
} }

如果需要,用try-catch和TrasactionScope()将其包装起来,为了保持代码干净,没有在这里显示它们

SqlBulkCopy速度极快

这是我的实现:

// at some point in my calling code, I will call:
var myDataTable = CreateMyDataTable();
myDataTable.Rows.Add(Guid.NewGuid,tableHeaderId,theName,theValue); // e.g. - need this call for each row to insert

var efConnectionString = ConfigurationManager.ConnectionStrings["MyWebConfigEfConnection"].ConnectionString;
var efConnectionStringBuilder = new EntityConnectionStringBuilder(efConnectionString);
var connectionString = efConnectionStringBuilder.ProviderConnectionString;
BulkInsert(connectionString, myDataTable);

private DataTable CreateMyDataTable()
{
    var myDataTable = new DataTable { TableName = "MyTable"};
// this table has an identity column - don't need to specify that
    myDataTable.Columns.Add("MyTableRecordGuid", typeof(Guid));
    myDataTable.Columns.Add("MyTableHeaderId", typeof(int));
    myDataTable.Columns.Add("ColumnName", typeof(string));
    myDataTable.Columns.Add("ColumnValue", typeof(string));
    return myDataTable;
}

private void BulkInsert(string connectionString, DataTable dataTable)
{
    using (var connection = new SqlConnection(connectionString))
    {
        connection.Open();
        SqlTransaction transaction = null;
        try
        {
            transaction = connection.BeginTransaction();

            using (var sqlBulkCopy = new SqlBulkCopy(connection, SqlBulkCopyOptions.TableLock, transaction))
            {
                sqlBulkCopy.DestinationTableName = dataTable.TableName;
                foreach (DataColumn column in dataTable.Columns) {
                    sqlBulkCopy.ColumnMappings.Add(column.ColumnName, column.ColumnName);
                }

                sqlBulkCopy.WriteToServer(dataTable);
            }
            transaction.Commit();
        }
        catch (Exception)
        {
            transaction?.Rollback();
            throw;
        }
    }
}

使用此技术可以提高实体框架中插入记录的速度。这里我使用一个简单的存储过程来插入记录。为了执行这个存储过程,我使用实体框架的.FromSql()方法来执行Raw SQL。

存储过程代码:

CREATE PROCEDURE TestProc
@FirstParam VARCHAR(50),
@SecondParam VARCHAR(50)

AS
  Insert into SomeTable(Name, Address) values(@FirstParam, @SecondParam) 
GO

接下来,循环遍历所有4000条记录,并添加执行存储的

该过程每100次循环一次。

为此,我创建了一个字符串查询来执行这个过程,并继续将每一组记录附加到它。

然后检查循环是否以100的倍数运行,在这种情况下,使用.FromSql()执行它。

所以对于4000条记录,我只需要执行以下步骤4000/100=40次。

检查以下代码:

string execQuery = "";
var context = new MyContext();
for (int i = 0; i < 4000; i++)
{
    execQuery += "EXEC TestProc @FirstParam = 'First'" + i + "'', @SecondParam = 'Second'" + i + "''";

    if (i % 100 == 0)
    {
        context.Student.FromSql(execQuery);
        execQuery = "";
    }
}

我将推荐这篇关于如何使用EF进行批量插入的文章。

实体框架和慢速批量INSERT

他探索了这些领域并比较了绩效:

默认EF(57分钟完成添加30000条记录)替换为ADO.NET代码(对于相同的30000,25秒)上下文膨胀-通过为每个工作单元使用一个新的上下文来保持活动的上下文图较小(相同的30000个插入需要33秒)大列表-关闭AutoDetectChangesEnabled(将时间缩短至约20秒)批处理(最短16秒)DbTable.AddRange()-(性能在12范围内)

秘密是插入到相同的空白暂存表中。插件快速发光。然后在主大表中运行一个单独的插入。然后截断临时表,为下一批做好准备。

ie.

insert into some_staging_table using Entity Framework.

-- Single insert into main table (this could be a tiny stored proc call)
insert into some_main_already_large_table (columns...)
   select (columns...) from some_staging_table
truncate table some_staging_table