Announcing AWS Lambda support for .NET Core 3.1

This post is courtesy of Norm Johanson, Senior Software Development Engineer, AWS SDKs and Tools.

From today, you can develop AWS Lambda functions using .NET Core 3.1. You can deploy to Lambda by setting the runtime parameter value to dotnetcore3.1. Version 1.17.0.0 AWS Toolkit for Visual Studio and version 4.0.0 of the .NET Core Global Tool Amazon.Lambda.Tools are also available today. These make it easy to build and deploy your .NET Core 3.1 Lambda functions.

New features of .NET Core 3.1

.NET Core 3.1 brings many new runtime features to Lambda including C# 8.0 and F# 4.7 support, .NET Standard 2.1 support, new JSON serializer, and a new ReadyToRun feature for ahead-of-time compilation. There are also new versions of the .NET Lambda tooling and libraries. These include the Amazon.Lambda.AspNetCoreServer package, which allows you to run ASP.NET Core 3.1
projects as Lambda functions.

New Lambda JSON serializer

.NET Core Lambda functions support JSON serialization of input and return parameters. Use this feature by registering a serializer in your Lambda code. Typically, this is done using an assembly attribute like this which registers the JsonSerializer class from the Amazon.Lambda.Serialization.Json NuGet package as the serializer:

[assembly:LambdaSerializer(typeof(Amazon.Lambda.Serialization.Json.JsonSerializer))]

Amazon.Lambda.Serialization.Json uses the popular NuGet package Newtonsoft.Json for serialization. Newtonsoft.Json is a powerful serializer with many built-in features. This makes it a large assembly to add to your .NET Core Lambda functions.

Starting with .NET Core 3.0, a new JSON serializer called System.Text.Json is built into the .NET Core framework. This serializer is focused on the core features of serialization and built for performance. To take advantage of this new serializer, use the new NuGet package Amazon.Lambda.Serialization.SystemTextJson. Testing with this new serializer shows significant
improvements to Lambda cold start latency. The new Lambda blueprints available in Visual Studio or dotnet new via Amazon.Lambda.Templates default to this new serializer using the following assembly attribute.

[assembly:LambdaSerializer(typeof(Amazon.Lambda.Serialization.SystemTextJson.DefaultLambdaJsonSerializer))]

Most .NET Lambda event packages work with either of the AWS serializer packages. For performance and simplicity reasons, the new versions of Amazon.Lambda.APIGatewayEvents and Amazon.Lambda.AspNetCoreServer only use the newer, faster Amazon.Lambda.Serialization.SystemTextJson for JSON serialization when targeting .NET Core 3.1.

ReadyToRun for better cold start performance

.NET Core 3.0 introduces a new build concept called ReadyToRun, which is also available in .NET Core 3.1. ReadyToRun performs much of the work of the just-in-time compiler used by the .NET runtime. If your project contains large amounts of code or large dependencies like the AWS SDK for .NET, this feature can significantly reduce cold start latency. It has less effect on small
functions using only the .NET Core base library.

To use ReadyToRun for Lambda, you must package your .NET Lambda function on Linux. You can use a Linux environment like an EC2 Linux instance, or CodeBuild, which also recently added .NET Core 3.1 support. Once you are in a Linux environment using .NET Lambda tooling, enable ReadyToRun by setting the –msbuild-parameters switch:

"/p:PublishReadyToRun=true --self-contained false"

For example, to deploy a function with ReadyToRun enabled using the .NET Core Global Tool for Lambda Amazon.Lambda.Tools, use:

dotnet lambda deploy-function R2RExample --msbuild-parameters "/p:PublishReadyToRun=true --self-contained false"

To avoid setting in the command line, set msbuild-parameters as a property in the aws-lambda-tools-defaults.json file.

Updated AWS Mock .NET Lambda Test Tool

With .NET Core 2.1, AWS released the AWS .NET Core Mock Lambda Test Tool. This makes it easy to debug .NET Core Lambda functions. If you are using the AWS Toolkit for Visual Studio, the toolkit automatically installs or updates the test tool, and configures your launchSettings.json file. With the toolkit, you can use F5 debugging when the project opens.

For .NET Core 3.1, this tool offers new features. First, the way the tool loads .NET Lambda code internally is redesigned. Previously, the assemblies in customer code may collide with the test tool’s assemblies. Now the Lambda code is loaded in a separate AssemblyLoadContext, preventing this collision.

The test tool is an ASP.NET Core application that loads and executes the Lambda code. This allows the debugger that is currently attached to the test tool to debug the loaded Lambda code. Pressing F5 opens the web interface, allowing you to select the function, payload, and other parameters. Once everything is set, choose execute to run the code inside the test tool’s process. To improve the debug turnaround cycle, there is a new switch: –no-ui. This skips the web interface after code changes, making it faster to debug your code.

My work flow for this tool is to use the web interface for the initial debug session, then save the request JSON. After the initial debug session, I edit the launchSettings.json file, which looks like this, setting up the port for the web interface:


{
  "profiles": {
    "Mock Lambda Test Tool": {
      "commandName": "Executable",
      "commandLineArgs": "--port 5050",
      "workingDirectory": ".\\bin\\$(Configuration)\\netcoreapp3.1",
      "executablePath": "C:\\Users\\%USERNAME%\\.dotnet\\tools\\dotnet-lambda-test-tool-3.1.exe"
    }
  }
}

I update this to:


{
  "profiles": {
    "Mock Lambda Test Tool": {
      "commandName": "Executable",
      "commandLineArgs": "—no-ui --payload SavedRequest",
      "workingDirectory": ".\\bin\\$(Configuration)\\netcoreapp3.1",
      "executablePath": "C:\\Users\\%USERNAME%\\.dotnet\\tools\\dotnet-lambda-test-tool-3.1.exe"
    }
  }
}

This uses the saved request from the web interface as the input payload, instead of the web interface.

For more information about this feature, see the new Documentation tab in the test tool after launching. Here is a demonstration of my debug workflow:

Amazon Linux 2

.NET Core 3.1, like (Ruby 2.7, Python 3.8, Node.js 10 and 12, and Java 11) is based on an Amazon Linux 2 execution environment. Amazon Linux 2 provides a secure, stable, and high-performance execution environment to develop and run cloud and enterprise applications.

Migrate to .NET Core 3.1

To migrate existing .NET Core 2.1 Lambda functions to the new 3.1 runtime, follow the steps below:

Open the csproj or fsproj file.
- Set the TargetFramework element to netcoreapp3.1.
Open the aws-lambda-tools-defaults.json file.
- If it exists, set the function-runtime field to dotnetcore3.1
- If it exists, set the framework field to netcoreapp3.1. If you remove the field, the value is inferred from the project file.
If it exists, open the serverless.template file.
- For any AWS::Lambda::Function or AWS::Servereless::Function, set the Runtime property to dotnetcore3.1
Update all Amazon.Lambda.* NuGet package references to the latest versions.

To use the new JSON serializer, follow these steps:

Remove the NuGet package reference to Amazon.Lambda.Serialization.Json.
Add the NuGet package reference to Amazon.Lambda.Serialization.SystemTextJson.
In your code, where the LambdaSerializer attribute registers the JSON serializer, change the parameter to Amazon.Lambda.Serialization.SystemTextJson.DefaultLambdaJsonSerializer.

Conclusion

There is a blueprint in Visual Studio for detecting labels for images uploaded in S3:

By converting this blueprint to .NET Core 3.1 and using the new JSON serializer and ReadyToRun features, the cold start time is reduced by 40% when using 256 MB of memory. Performance improvements vary, so be sure to try these new features in your Lambda functions.

Start building .NET Core 3.1 Lambda functions with the latest versions of the AWS Toolkit for Visual Studio or the .NET Core Global Tool Amazon.Lambda.Tools. If you are not using .NET Core Lambda tooling, specify dotnetcore3.1 as the runtime value in your preferred tool to deploy Lambda functions.

We would like to hear your feedback for AWS .NET Lambda support. Contact the AWS .NET Team for Lambda questions through our .NET Lambda GitHub repository.

AWS Compute Blog