IEnumerable 扩展以批量拉取结果

IEnumerable extension to pull results in batches

我正在使用 Entity Framework 并且经常 运行 遇到我想要遍历大量记录的问题。我的问题是,如果我一次把它们全部拉出来,我就有超时的风险;如果我一次拉一个,实际上每条记录都将是一个单独的查询,并且需要很长时间。

我想实现一个 Linq 扩展,它可以批量提取结果,但仍然可以用作 IEnumerable。我会给它一组键(很可能是我正在提取的任何记录的主要 ID)、批处理大小(简单对象较大,复杂对象较小),以及定义如何应用的 Func一组记录类型 T 的一组键。我会这样称呼它:

//get the list of items to pull--in this case, a set of order numbers
List<int> orderNumbers = GetOrderNumbers();

//set the batch size
int batchSize = 100;

//loop through the set using BatchedSelector extension. Note the selection
//function at the end which allows me to 
foreach (var order in dbContext.Orders.BatchedSelector(repairNumbers, batchSize, (o, k) => k.Contains(o.OrderNumber)))
{
    //do things
}

这是我的解决方案草稿:

    /// <summary>
    /// A Linq extension that fetches IEnumerable results in batches, aggregating queries
    /// to improve EF performance. Operates transparently to application and acts like any
    /// other IEnumerable.
    /// </summary>
    /// <typeparam name="T">Header record type</typeparam>
    /// <param name="source">Full set of records</param>
    /// <param name="keys">The set of keys that represent specific records to pull</param>
    /// <param name="selector">Function that filters the result set to only those which match the key set</param>
    /// /// <param name="maxBatchSize">Maximum number of records to pull in one query</param>
    /// <returns></returns>
    public static IEnumerable<T> BatchedSelector<T>(this IEnumerable<T> source, IEnumerable<int> keys, Func<T, IEnumerable<int>, bool> selector, int maxBatchSize)
    {
        //the index of the next key (or set of keys) to process--we start at 0 of course
        int currentKeyIndex = 0;             

        //to provide some resiliance, we will allow the batch size to decrease if we encounter errors
        int currentBatchSize = maxBatchSize;
        int batchDecreaseAmount = Math.Max(1, maxBatchSize / 10); //10%, but at least 1

        //other starting variables; a list to hold results and the associated batch of keys
        List<T> resultList = null;
        IEnumerable<int> keyBatch = null;

        //while there are still keys remaining, grab the next set of keys
        while ((keyBatch = keys.Skip(currentKeyIndex).Take(currentBatchSize)).Count() > 0)
        {
            //try to fetch the results
            try
            {
                resultList = source.Where(o => selector(o, keyBatch)).ToList();  // <-- this is where errors occur
                currentKeyIndex += maxBatchSize;  //increment key index to mark these keys as processed
            }
            catch
            {
                //decrease the batch size for our retry
                currentBatchSize -= batchDecreaseAmount;

                //if we've run out of batch overhead, throw the error
                if (currentBatchSize <= 0) throw;

                //otherwise, restart the loop
                continue;
            }

            //since we've successfully gotten the set of keys, yield the results
            foreach (var match in resultList) yield return match;
        }

        //the loop is over; we're done
        yield break;
    }

出于某种原因,"where" 子句无效。我已验证正确的密钥在 keyBatch 中,但预期的 WHERE OrderNumber IN (k1, k2, k3, kn) 行不存在。就好像我根本没有 where 语句。

我最好的猜测是我需要构建表达式并对其进行编译,但我不确定这是否是问题所在,我也不确定如何修复它。会喜欢任何输入。谢谢!

WhereSkipTake等所有这些方法都是扩展方法,不是IEnumerable<T>的成员。因为所有这些方法实际上有 2 个版本,一个用于 IEnumerable<>,一个用于 IQueryable<>

可枚举扩展

  • Where(Func<TSource, bool> predicate)
  • Select(Func<TSource, TResult> selector)

可查询扩展

  • Where(Expression<Func<TSource, bool>> predicate)
  • Select(Expression<Func<TSource, TResult>> predicate)

如您所见,不同之处在于 Queryable 扩展采用 Expression<> 而不是直接委托。这些表达式允许 EF 将您的代码转换为 SQL.

由于您在 BatchedSelector() 方法中将 variables/parameters 声明为 IEnumerable<> 您正在使用 Enumerable class 中的扩展,并且此扩展已执行在记忆中。

一个常见的错误是认为由于多态性,DbSet (IQueryable<>) 无论您将它用作 IEnumerable<>,查询都会被翻译成 SQL,这仅适用于适当的成员,但不适用于扩展方法。

可以修复您的代码,将 IEnumerable<> variables/parameters 更改为 IQueryable<>

您可以详细了解 IEnumerableIQueryable here 之间的区别。

首先谢谢大家Arturo。你让我走上了这个解决方案的正确轨道。我认为这是一个 Linq->Entity 问题,但这些问题对我来说仍然远非凭直觉来解决。

其次,我大量借鉴了Shimmy's answer to this question。谢谢希米!

首先,我更新了方法以支持整数以外的键类型,因为为什么不呢。所以方法签名现在是(注意对 IQueryable 源的更改):

public static IEnumerable<T> BatchedSelector<T, TKey>(this IQueryable<T> source, Expression<Func<T, TKey>> selector, IEnumerable<TKey> keys, int maxBatchSize)

除了产生错误的行之外,该方法基本保持不变,现在被替换为:

resultList = source.WhereIn(selector, keyBatch).ToList();

WhereIn 是主要从 Shimmy 借来的 Linq 扩展:

    public static IQueryable<T> WhereIn<T, TKey>(this IQueryable<T> source, Expression<Func<T, TKey>> selector, IEnumerable<TKey> keyCollection)
    {
        if (selector == null) throw new ArgumentNullException("Null selector");
        if (keyCollection == null) throw new ArgumentNullException("Null collection");

        //if no items in collection, no results
        if (!keyCollection.Any()) return source.Where(t => false);

        //assemble expression
        var p = selector.Parameters.Single();
        var equals = keyCollection.Select(value => (Expression)Expression.Equal(selector.Body, Expression.Constant(value, typeof(TKey))));
        var body = equals.Aggregate((accumulate, equal) => Expression.Or(accumulate, equal));

        //return expression
        return source.Where(Expression.Lambda<Func<T, bool>>(body, p));
    }

这让我学到了一些很酷的东西:如果你输入一个包含一堆常量比较的 where 子句,它将被转换为一个 SQL In 语句!整洁!

通过这些更改,该方法可以快速轻松地产生结果。