Swift AWS Lambda Runtime

Many modern systems have client components like iOS, macOS or watchOS applications as well as server components that those clients interact with. Serverless functions are often the easiest and most efficient way for client application developers to extend their applications into the cloud.

Serverless functions are increasingly becoming a popular choice for running event-driven or otherwise ad-hoc compute tasks in the cloud. They power mission critical microservices and data intensive workloads. In many cases, serverless functions allow developers to more easily scale and control compute costs given their on-demand nature.

When using serverless functions, attention must be given to resource utilization as it directly impacts the costs of the system. This is where Swift shines! With its low memory footprint, deterministic performance, and quick start time, Swift is a fantastic match for the serverless functions architecture.

Combine this with Swift's developer friendliness, expressiveness, and emphasis on safety, and we have a solution that is great for developers at all skill levels, scalable, and cost effective.

Swift AWS Lambda Runtime was designed to make building Lambda functions in Swift simple and safe. The library is an implementation of the AWS Lambda Runtime API and uses an embedded asynchronous HTTP Client based on SwiftNIO that is fine-tuned for performance in the AWS Runtime context. The library provides a multi-tier API that allows building a range of Lambda functions: From quick and simple closures to complex, performance-sensitive event handlers.

Project status

This is the beginning of a community-driven open-source project actively seeking contributions. While the core API is considered stable, the API may still evolve as we get closer to a 1.0 version. There are several areas which need additional attention, including but not limited to:

Further performance tuning
Additional trigger events
Additional documentation and best practices
Additional examples

Getting started

If you have never used AWS Lambda or Docker before, check out this getting started guide which helps you with every step from zero to a running Lambda.

First, create a SwiftPM project and pull Swift AWS Lambda Runtime as dependency into your project

// swift-tools-version:5.2

import PackageDescription

let package = Package(
    name: "my-lambda",
    products: [
        .executable(name: "MyLambda", targets: ["MyLambda"]),
    ],
    dependencies: [
        .package(url: "https://github.com/swift-server/swift-aws-lambda-runtime.git", from: "0.1.0"),
    ],
    targets: [
        .target(name: "MyLambda", dependencies: [
          .product(name: "AWSLambdaRuntime", package: "swift-aws-lambda-runtime"),
        ]),
    ]
)

Next, create a main.swift and implement your Lambda.

Using Closures

The simplest way to use AWSLambdaRuntime is to pass in a closure, for example:

// Import the module
import AWSLambdaRuntime

// in this example we are receiving and responding with strings
Lambda.run { (context, payload: String, callback) in
  callback(.success("Hello, \(payload)"))
}

More commonly, the payload would be a JSON, which is modeled using Codable, for example:

// Import the module
import AWSLambdaRuntime

// Request, uses Codable for transparent JSON encoding
private struct Request: Codable {
  let name: String
}

// Response, uses Codable for transparent JSON encoding
private struct Response: Codable {
  let message: String
}

// In this example we are receiving and responding with `Codable`.
Lambda.run { (context, request: Request, callback) in
  callback(.success(Response(message: "Hello, \(request.name)")))
}

Since most Lambda functions are triggered by events originating in the AWS platform like SNS, SQS or APIGateway, the package also includes a AWSLambdaEvents module that provides implementations for most common AWS event types further simplifying writing Lambda functions. For example, handling an SQS message:

// Import the modules
import AWSLambdaRuntime
Import AWSLambdaEvents

// In this example we are receiving an SQS Message, with no response (Void).
Lambda.run { (context, message: SQS.Message, callback) in
  ...
  callback(.success(Void()))
}

Modeling Lambda functions as Closures is both simple and safe. Swift AWS Lambda Runtime will ensure that the user-provided code is offloaded from the network processing thread such that even if the code becomes slow to respond or gets hang, the underlying process can continue to function. This safety comes at a small performance penalty from context switching between threads. In many cases, the simplicity and safety of using the Closure based API is often preferred over the complexity of the performance-oriented API.

Using EventLoopLambdaHandler

Performance sensitive Lambda functions may choose to use a more complex API which allows user code to run on the same thread as the networking handlers. Swift AWS Lambda Runtime uses SwiftNIO as its underlying networking engine which means the APIs are based on SwiftNIO concurrency primitives like the EventLoop and EventLoopFuture. For example:

// Import the modules
import AWSLambdaRuntime
import AWSLambdaEvents
import NIO

// Our Lambda handler, conforms to EventLoopLambdaHandler
struct Handler: EventLoopLambdaHandler {
    typealias In = SNS.Message // Request type
    typealias Out = Void // Response type

    // In this example we are receiving an SNS Message, with no response (Void).
    func handle(context: Lambda.Context, payload: In) -> EventLoopFuture<Out> {
        ...
        context.eventLoop.makeSucceededFuture(Void())
    }
}

Lambda.run(Handler())

Beyond the small cognitive complexity of using the EventLoopFuture based APIs, note these APIs should be used with extra care. An EventLoopLambdaHandler will execute the user code on the same EventLoop (thread) as the library, making processing faster but requiring the user code to never call blocking APIs as it might prevent the underlying process from functioning.

Deploying to AWS Lambda

To deploy Lambda functions to AWS Lambda, you need to compile the code for Amazon Linux which is the OS used on AWS Lambda microVMs, package it as a Zip file, and upload to AWS.

AWS offers several tools to interact and deploy Lambda functions to AWS Lambda including SAM and the AWS CLI. The Examples Directory includes complete sample build and deployment scripts that utilize these tools.

Note the examples mentioned above use dynamic linking, therefore bundle the required Swift libraries in the Zip package along side the executable. You may choose to link the Lambda function statically (using -static-stdlib) which could improve performance but requires additional linker flags.

To build the Lambda function for Amazon Linux, use the Docker image published by Swift.org on Swift toolchains and Docker images for Amazon Linux 2, as demonstrated in the examples.

Architecture

The library defines three protocols for the implementation of a Lambda Handler. From low-level to more convenient:

ByteBufferLambdaHandler

An EventLoopFuture based processing protocol for a Lambda that takes a ByteBuffer and returns a ByteBuffer? asynchronously.

ByteBufferLambdaHandler is the lowest level protocol designed to power the higher level EventLoopLambdaHandler and LambdaHandler based APIs. Users are not expected to use this protocol, though some performance sensitive applications that operate at the ByteBuffer level or have special serialization needs may choose to do so.

public protocol ByteBufferLambdaHandler {
    /// The Lambda handling method
    /// Concrete Lambda handlers implement this method to provide the Lambda functionality.
    ///
    /// - parameters:
    ///  - context: Runtime `Context`.
    ///  - payload: The event or request payload encoded as `ByteBuffer`.
    ///
    /// - Returns: An `EventLoopFuture` to report the result of the Lambda back to the runtime engine.
    /// The `EventLoopFuture` should be completed with either a response encoded as `ByteBuffer` or an `Error`
    func handle(context: Lambda.Context, payload: ByteBuffer) -> EventLoopFuture<ByteBuffer?>
}

EventLoopLambdaHandler

EventLoopLambdaHandler is a strongly typed, EventLoopFuture based asynchronous processing protocol for a Lambda that takes a user defined In and returns a user defined Out.

EventLoopLambdaHandler extends ByteBufferLambdaHandler, providing ByteBuffer -> In decoding and Out -> ByteBuffer? encoding for Codable and String.

EventLoopLambdaHandler executes the user provided Lambda on the same EventLoop as the core runtime engine, making the processing fast but requires more care from the implementation to never block the EventLoop. It it designed for performance sensitive applications that use Codable or String based Lambda functions.

public protocol EventLoopLambdaHandler: ByteBufferLambdaHandler {
    associatedtype In
    associatedtype Out

    /// The Lambda handling method
    /// Concrete Lambda handlers implement this method to provide the Lambda functionality.
    ///
    /// - parameters:
    ///  - context: Runtime `Context`.
    ///  - payload: Payload of type `In` representing the event or request.
    ///
    /// - Returns: An `EventLoopFuture` to report the result of the Lambda back to the runtime engine.
    /// The `EventLoopFuture` should be completed with either a response of type `Out` or an `Error`
    func handle(context: Lambda.Context, payload: In) -> EventLoopFuture<Out>

    /// Encode a response of type `Out` to `ByteBuffer`
    /// Concrete Lambda handlers implement this method to provide coding functionality.
    /// - parameters:
    ///  - allocator: A `ByteBufferAllocator` to help allocate the `ByteBuffer`.
    ///  - value: Response of type `Out`.
    ///
    /// - Returns: A `ByteBuffer` with the encoded version of the `value`.
    func encode(allocator: ByteBufferAllocator, value: Out) throws -> ByteBuffer?

    /// Decode a`ByteBuffer` to a request or event of type `In`
    /// Concrete Lambda handlers implement this method to provide coding functionality.
    ///
    /// - parameters:
    ///  - buffer: The `ByteBuffer` to decode.
    ///
    /// - Returns: A request or event of type `In`.
    func decode(buffer: ByteBuffer) throws -> In
}

LambdaHandler

LambdaHandler is a strongly typed, completion handler based asynchronous processing protocol for a Lambda that takes a user defined In and returns a user defined Out.

LambdaHandler extends ByteBufferLambdaHandler, performing ByteBuffer -> In decoding and Out -> ByteBuffer encoding for Codable and String.

LambdaHandler offloads the user provided Lambda execution to a DispatchQueue making processing safer but slower.

public protocol LambdaHandler: EventLoopLambdaHandler {
    /// Defines to which `DispatchQueue` the Lambda execution is offloaded to.
    var offloadQueue: DispatchQueue { get }

    /// The Lambda handling method
    /// Concrete Lambda handlers implement this method to provide the Lambda functionality.
    ///
    /// - parameters:
    ///  - context: Runtime `Context`.
    ///  - payload: Payload of type `In` representing the event or request.
    ///  - callback: Completion handler to report the result of the Lambda back to the runtime engine.
    ///  The completion handler expects a `Result` with either a response of type `Out` or an `Error`
    func handle(context: Lambda.Context, payload: In, callback: @escaping (Result<Out, Error>) -> Void)
}

Closures

In addition to protocol-based Lambda, the library provides support for Closure-based ones, as demonstrated in the overview section above. Closure-based Lambdas are based on the LambdaHandler protocol which mean they are safer. For most use cases, Closure-based Lambda is a great fit and users are encouraged to use them.

The library includes implementations for Codable and String based Lambda. Since AWS Lambda is primarily JSON based, this covers the most common use cases.

public typealias CodableClosure<In: Decodable, Out: Encodable> = (Lambda.Context, In, @escaping (Result<Out, Error>) -> Void) -> Void

public typealias StringClosure = (Lambda.Context, String, @escaping (Result<String, Error>) -> Void) -> Void

This design allows for addition payload types as well, and such Lambda implementation can extend one of the above protocols and provided their own ByteBuffer -> In decoding and Out -> ByteBuffer encoding.

Context

When calling the user provided Lambda function, the library provides a Context class that provides metadata about the execution context, as well as utilities for logging and allocating buffers.

public final class Context {
    /// The request ID, which identifies the request that triggered the function invocation.
    public let requestID: String

    /// The AWS X-Ray tracing header.
    public let traceID: String

    /// The ARN of the Lambda function, version, or alias that's specified in the invocation.
    public let invokedFunctionARN: String

    /// The timestamp that the function times out
    public let deadline: DispatchWallTime

    /// For invocations from the AWS Mobile SDK, data about the Amazon Cognito identity provider.
    public let cognitoIdentity: String?

    /// For invocations from the AWS Mobile SDK, data about the client application and device.
    public let clientContext: String?

    /// `Logger` to log with
    ///
    /// - note: The `LogLevel` can be configured using the `LOG_LEVEL` environment variable.
    public let logger: Logger

    /// The `EventLoop` the Lambda is executed on. Use this to schedule work with.
    /// This is useful when implementing the `EventLoopLambdaHandler` protocol.
    ///
    /// - note: The `EventLoop` is shared with the Lambda runtime engine and should be handled with extra care.
    ///  Most importantly the `EventLoop` must never be blocked.
    public let eventLoop: EventLoop

    /// `ByteBufferAllocator` to allocate `ByteBuffer`
    /// This is useful when implementing `EventLoopLambdaHandler`
    public let allocator: ByteBufferAllocator
}

Configuration

The library’s behavior can be fine tuned using environment variables based configuration. The library supported the following environment variables:

LOG_LEVEL: Define the logging level as defined by SwiftLog. Set to INFO by default.
MAX_REQUESTS: Max cycles the library should handle before exiting. Set to none by default.
STOP_SIGNAL: Signal to capture for termination. Set to TERM by default.
REQUEST_TIMEOUT: Max time to wait for responses to come back from the AWS Runtime engine. Set to none by default.

AWS Lambda Runtime Engine Integration

The library is designed to integrate with AWS Lambda Runtime Engine via the AWS Lambda Runtime API which was introduced as part of AWS Lambda Custom Runtimes in 2018. The latter is an HTTP server that exposes three main RESTful endpoint:

/runtime/invocation/next
/runtime/invocation/response
/runtime/invocation/error

A single Lambda execution workflow is made of the following steps:

The library calls AWS Lambda Runtime Engine /next endpoint to retrieve the next invocation request.
The library parses the response HTTP headers and populate the Context object.
The library reads the /next response body and attempt to decode it. Typically it decodes to user provided In type which extends Decodable, but users may choose to write Lambda functions that receive the input as String or ByteBuffer which require less, or no decoding.
The library hands off the Context and In payload to the user provided handler. In the case of LambdaHandler based handler this is done on a dedicated DispatchQueue, providing isolation between user's and the library's code.
User provided handler processes the request asynchronously, invoking a callback or returning a future upon completion, which returns a Result type with the Out or Error populated.
In case of error, the library posts to AWS Lambda Runtime Engine /error endpoint to provide the error details, which will show up on AWS Lambda logs.
In case of success, the library will attempt to encode the response. Typically it encodes from user provided Out type which extends Encodable, but users may choose to write Lambda functions that return a String or ByteBuffer, which require less, or no encoding. The library then posts the response to AWS Lambda Runtime Engine /response endpoint to provide the response to the callee.

The library encapsulates the workflow via the internal LambdaRuntimeClient and LambdaRunner structs respectively.

Lifecycle Management

AWS Lambda Runtime Engine controls the Application lifecycle and in the happy case never terminates the application, only suspends it's execution when no work is avaialble.

As such, the library main entry point is designed to run forever in a blocking fashion, performing the workflow described above in an endless loop.

That loop is broken if/when an internal error occurs, such as a failure to communicate with AWS Lambda Runtime Engine API, or under other unexpected conditions.

By default, the library also registers a Signal handler that traps INT and TERM , which are typical Signals used in modern deployment platforms to communicate shutdown request.

Integration with AWS Platform Events

AWS Lambda functions can be invoked directly from the AWS Lambda console UI, AWS Lambda API, AWS SDKs, AWS CLI, and AWS toolkits. More commonly, they are invoked as a reaction to an events coming from the AWS platform. To make it easier to integrate with AWS platform events, the library includes an AWSLambdaEvents target which provides abstractions for many commonly used events. Additional events can be easily modeled when needed following the same patterns set by AWSLambdaEvents. Integration points with the AWS Platform include:

Note: Each one of the integration points mentioned above includes a set of Codable structs that mirror AWS' data model for these APIs.

Performance

Lambda functions performance is usually measured across two axes:

Cold start times: The time it takes for a Lambda function to startup, ask for an invocation and process the first invocation.
Warm invocation times: The time it takes for a Lambda function to process an invocation after the Lambda has been invoked at least once.

Larger packages size (Zip file uploaded to AWS Lambda) negatively impact the cold start time, since AWS needs to download and unpack the package before starting the process.

Swift provides great Unicode support via ICU. Therefore, Swift-based Lambda functions include the ICU libraries which tend to be large. This impacts the download time mentioned above and an area for further optimization. Some of the alternatives worth exploring are using the system ICU that comes with Amazon Linux (albeit older than the one Swift ships with) or working to remove the ICU dependency altogether. We welcome ideas and contributions to this end.

wiomoc/swift-aws-lambda-runtime