Docker中的.NET核心高内存使用率(AWS ECS Fargate)

.NET Core high memory usage in Docker (AWS ECS Fargate)(Docker中的.NET核心高内存使用率(AWS ECS Fargate))

问题描述

以下是对该应用程序的描述。我有一个运行Hangfire的控制台应用程序。有几个工作是重复进行的。尤其是一个作业，它要解析几个大型的XML文件。我将此控制台应用程序部署为AWS ECS Fargate容器。

每次运行重复作业时，容器内存使用量不断增加，直到应用程序变得无响应或AWS终止容器。这是ECS上的一个16 GB的容器(大)，所以它应该能够处理我扔给它的任何东西。我可以断定，一旦内存使用量达到约13 GB，它就注定要崩溃。

关键是我在本地运行此应用程序，它的内存使用率从来不会很高。我将从我的工作站和下面的容器中发布几个示例日志。

感谢您的真知灼见！

分析XML：

• 使用XmlSerializer放入JSON格式的对象
• 使用XDocument只需浏览文档
• 使用XSD文件生成类，然后使用XmlSerializer填充

这三种方法都在本地运行，但在服务器上使用大量内存。这些都是不变的XML文件，我在本地和容器中运行完全相同的文件。

仅供参考：我已经关闭了容器上的服务器垃圾收集。这不管用。

编辑2019年3月5日

下面的日志来自以下代码。它不具备应用程序的全部功能，而应用程序相当复杂。此代码位于注册了Hangfire的方法中，每天运行一次。它从一个FTP站点提取XML文件，然后基本上运行下面的代码来处理这些文件。每个文件的大小不超过10 MB。

XmlSerializer serializer = new XmlSerializer(Model);
foreach (var file in files)
{
    Console.WriteLine($"Processing file: {file.Name}");
    Model model;
    using (XmlReader reader = XmlReader.Create(file.FullName, new XmlReaderSettings { CheckCharacters = false }))
    {
        model = (Model)serializer.Deserialize(reader);
    }
    SaveToDatabase(model);
    Console.WriteLine($"Application Memory Usage - Working: {Process.GetCurrentProcess()?.WorkingSet64 ?? 0}, Private: {Process.GetCurrentProcess()?.PrivateMemorySize64 ?? 0}");
}

工作站日志：

[13:52:58 INF] Processing file: price_feed_supplemental_002981.xml
[13:52:58 INF] Application Memory Usage - Working: 1.4 GB, Private: 1.6 GB
[13:52:58 INF] Processing file: price_feed_002982.xml
[13:53:01 INF] Application Memory Usage - Working: 1.4 GB, Private: 1.6 GB
[13:53:01 INF] Processing file: price_feed_supplemental_002982.xml
[13:53:01 INF] Application Memory Usage - Working: 1.4 GB, Private: 1.6 GB
[13:53:01 INF] Processing file: price_feed_002983.xml
[13:53:01 INF] Application Memory Usage - Working: 1.4 GB, Private: 1.6 GB
[13:53:01 INF] Processing file: price_feed_supplemental_002983.xml
[13:53:02 INF] Application Memory Usage - Working: 1.4 GB, Private: 1.6 GB
[13:53:02 INF] Processing file: price_feed_002984.xml

容器日志：

[19:05:35 INF] Application Memory Usage - Working: 1.9 GB, Private: 0.0 bytes
[19:05:35 INF] Processing file: price_feed_supplemental_002987.xml
[19:05:35 INF] Application Memory Usage - Working: 1.9 GB, Private: 0.0 bytes
[19:05:35 INF] Processing file: price_feed_supplemental_002988.xml
[19:05:35 INF] Application Memory Usage - Working: 1.9 GB, Private: 0.0 bytes
[19:05:35 INF] Processing file: price_feed_002988.xml
[19:05:40 INF] Application Memory Usage - Working: 1.9 GB, Private: 0.0 bytes
[19:05:40 INF] Processing file: price_feed_002989.xml
[19:05:43 INF] Application Memory Usage - Working: 2.1 GB, Private: 0.0 bytes
[19:05:43 INF] Processing file: price_feed_supplemental_002989.xml
[19:05:43 INF] Application Memory Usage - Working: 2.1 GB, Private: 0.0 bytes
[19:05:43 INF] Processing file: price_feed_supplemental_002990.xml
[19:05:43 INF] Application Memory Usage - Working: 2.1 GB, Private: 0.0 bytes
[19:05:43 INF] Processing file: price_feed_002990.xml
[19:05:51 INF] Application Memory Usage - Working: 2.6 GB, Private: 0.0 bytes
[19:05:51 INF] Processing file: price_feed_002991.xml
[19:05:57 INF] Application Memory Usage - Working: 2.2 GB, Private: 0.0 bytes
[19:05:57 INF] Processing file: price_feed_supplemental_002991.xml
[19:06:00 INF] Application Memory Usage - Working: 2.2 GB, Private: 0.0 bytes
[19:06:00 INF] Processing file: price_feed_002992.xml
[19:06:19 INF] Application Memory Usage - Working: 2.1 GB, Private: 0.0 bytes
[19:06:19 INF] Processing file: price_feed_supplemental_002992.xml
[19:06:21 INF] Application Memory Usage - Working: 2.3 GB, Private: 0.0 bytes
[19:06:21 INF] Uncompressing file: usa_LTSF_20190228_002993_001_INC.zip
[19:06:22 INF] Source file: price_feed_002993_1.xml
[19:06:22 INF] Source file: price_feed_supplemental_002993_8.xml
[19:06:23 INF] Source file: price_feed_supplemental_002993_11.xml
[19:06:23 INF] Source file: price_feed_002993_2.xml
[19:06:23 INF] Source file: price_feed_supplemental_002993_5.xml
[19:06:23 INF] Source file: price_feed_supplemental_002993_4.xml
[19:06:24 INF] Source file: price_feed_supplemental_002993_17.xml
[19:06:24 INF] Source file: price_feed_supplemental_002993_15.xml
[19:06:24 INF] Source file: price_feed_supplemental_002993_3.xml
[19:06:25 INF] Source file: price_feed_supplemental_002993_1.xml
[19:06:25 INF] Source file: price_feed_supplemental_002993_9.xml
[19:06:25 INF] Source file: price_feed_supplemental_002993_7.xml
[19:06:26 INF] Source file: price_feed_supplemental_002993_6.xml
[19:06:26 INF] Source file: price_feed_supplemental_002993_14.xml
[19:06:26 INF] Source file: price_feed_supplemental_002993_13.xml
[19:06:27 INF] Source file: price_feed_supplemental_002993_16.xml
[19:06:27 INF] Source file: price_feed_supplemental_002993_12.xml
[19:06:27 INF] Source file: price_feed_supplemental_002993_2.xml
[19:06:28 INF] Source file: price_feed_supplemental_002993_10.xml
[19:06:28 INF] Processing file: price_feed_002984.xml
[19:06:28 INF] Application Memory Usage - Working: 2.5 GB, Private: 0.0 bytes
[19:06:28 INF] Processing file: price_feed_supplemental_002984.xml
[19:06:28 INF] Application Memory Usage - Working: 2.5 GB, Private: 0.0 bytes
[19:06:28 INF] Processing file: price_feed_supplemental_002985.xml
[19:06:28 INF] Application Memory Usage - Working: 2.5 GB, Private: 0.0 bytes
[19:06:28 INF] Processing file: price_feed_002985.xml
[19:06:28 INF] Application Memory Usage - Working: 2.5 GB, Private: 0.0 bytes
[19:06:28 INF] Processing file: price_feed_supplemental_002986.xml
[19:06:28 INF] Application Memory Usage - Working: 2.5 GB, Private: 0.0 bytes
[19:06:28 INF] Processing file: price_feed_002986.xml
[19:06:28 INF] Application Memory Usage - Working: 2.5 GB, Private: 0.0 bytes
[19:06:28 INF] Processing file: price_feed_002987.xml
[19:06:30 INF] Application Memory Usage - Working: 2.6 GB, Private: 0.0 bytes
[19:06:30 INF] Processing file: price_feed_supplemental_002987.xml
[19:06:30 INF] Application Memory Usage - Working: 2.6 GB, Private: 0.0 bytes
[19:06:30 INF] Processing file: price_feed_supplemental_002988.xml
[19:06:30 INF] Application Memory Usage - Working: 2.6 GB, Private: 0.0 bytes
[19:06:30 INF] Processing file: price_feed_002988.xml
[19:06:35 INF] Application Memory Usage - Working: 2.9 GB, Private: 0.0 bytes
[19:06:35 INF] Processing file: price_feed_002989.xml
[19:06:38 INF] Application Memory Usage - Working: 3.1 GB, Private: 0.0 bytes
[19:06:38 INF] Processing file: price_feed_supplemental_002989.xml
[19:06:38 INF] Application Memory Usage - Working: 3.1 GB, Private: 0.0 bytes
[19:06:38 INF] Processing file: price_feed_supplemental_002990.xml
[19:06:38 INF] Application Memory Usage - Working: 3.1 GB, Private: 0.0 bytes
[19:06:38 INF] Processing file: price_feed_002990.xml
[19:06:47 INF] Application Memory Usage - Working: 3.7 GB, Private: 0.0 bytes
[19:06:47 INF] Processing file: price_feed_002991.xml
[19:06:53 INF] Application Memory Usage - Working: 4.1 GB, Private: 0.0 bytes
[19:06:53 INF] Processing file: price_feed_supplemental_002991.xml
[19:06:55 INF] Application Memory Usage - Working: 4.3 GB, Private: 0.0 bytes
[19:06:55 INF] Processing file: price_feed_002992.xml
[19:07:05 INF] Application Memory Usage - Working: 4.9 GB, Private: 0.0 bytes
[19:07:05 INF] Processing file: price_feed_supplemental_002992.xml
[19:07:07 INF] Application Memory Usage - Working: 5.0 GB, Private: 0.0 bytes
[19:07:07 INF] Processing file: price_feed_002993_1.xml
[19:09:40 INF] 1 servers were removed due to timeout
[19:10:49 INF] Application Memory Usage - Working: 10.0 GB, Private: 0.0 bytes
[19:10:49 INF] Processing file: price_feed_supplemental_002993_8.xml
[19:12:10 INF] Application Memory Usage - Working: 12.5 GB, Private: 0.0 bytes
[19:12:10 INF] Processing file: price_feed_supplemental_002993_11.xml
[19:13:08 INF] Application Memory Usage - Working: 13.5 GB, Private: 0.0 bytes
[19:13:08 INF] Processing file: price_feed_002993_2.xml
[19:14:06 INF] Application Memory Usage - Working: 13.8 GB, Private: 0.0 bytes
[19:14:06 INF] Processing file: price_feed_supplemental_002993_5.xml

推荐答案

通过将<ServerGarbageCollection>false</ServerGarbageCollection>添加到Web应用程序的csproj文件来禁用服务器垃圾收集(强制工作站垃圾收集)似乎会使GC更急切，从而帮助保持堆小，但非托管内存仍在缓慢增加。它最终是一个内存泄漏--静态方法通过引用将其一个参数传递回调用方，静态方法不是由GC管理的，因此每次调用此方法并提供更多要枚举的字符串时，通过引用传递的对象变得越来越大。

我删除了按引用传递，该方法现在正常返回对象(通过return)，但我仍然必须在此静态方法中调用GC.Collect()以防止约500MB-1 GB的瞬时峰值。正常情况下，即使在生产环境中这也没有问题，因为GC确实会迅速将内存使用率降至正常水平，但是，如果这些内存峰值之一导致.NET应用程序超过内存配额，将终止ECS实例。

不用说，我们正在使用O(N^2)个枚举中的一些大字符串，我怀疑在很好地重写该方法之后可以删除对GC.Collect()的调用。在您的例子中，您可能在某个地方遗漏了using，或者您正在使用事件(不太可能)，或者可能像我一样，您有一个通过引用传递某些内容的静态方法--或者不是--而那个不受GC约束的静态方法正在慢慢地泄漏。

对于禁用的&q；服务器&垃圾收集，这似乎只有在Fargate/ECS任务仅在一个vCPU上运行时才有效。有关服务器与工作站GC的详细信息，请参阅here。

Visual Studio 2022的内置性能测量工具还可以，但VS偶尔会崩溃(我的工作计算机也弱得要死)，所以我最终使用了JetBrains的dotMemory，它有更好的界面和更好的快照比较功能。

这篇关于Docker中的.NET核心高内存使用率(AWS ECS Fargate)的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持编程学习网！