hal

A runtime environment for Haskell applications running on AWS Lambda.

Flexible

This library uniquely supports different types of AWS Lambda Handlers for your needs/comfort with advanced Haskell. Instead of exposing a single function that constructs a Lambda, this library exposes many.

For lambdas that are pure and safe, then pureRuntime is ideal. It accepts a handler with the signature (FromJSON a, ToJSON b) => a -> b. This runtime guarantees that side-effects cannot occur.

For advanced use cases mRuntime unlocks the full power of Monad Transformers. It accepts handlers with the signature (MonadCatch m, MonadIO m, FromJSON event, ToJSON result) => (event -> m result) This enables users to add caching logic or expose complex environments.

With numerous options in between these two, developers can choose the right balance of flexibility vs simplicity.

Performant

Measuring lambda performance is tricky, so investigation and optimization is ongoing. Current indications show a warm execution overhead of only ~20% more than the official Rust Runtime (a much lower level language).

Robust

While testing continues, we have executed over 30k test events without error caused by the runtime. Naive approaches lead to error rates well over 10%.

Table of Contents

• Supported Platforms / GHC Versions
• Quick Start
• Local Testing

Supported Platforms / GHC Versions

We currently support this library under the same environment that AWS Lambda supports.

Our CI currently targets the latest three LTS Stackage Versions, the latest three minor versions of GHC under Cabal (e.g. 8.6.x, 8.4.x, and 8.2.x), and GHC-head / Stackage nightly builds.

If you haven’t already, adding docker: { enable: true } to your stack.yaml file will ensure that you’re building a binary that can run in AWS Lambda.

Quick Start

This quick start assumes you have the following tools installed:

• Stack
• Docker
• aws-cli

Add hal to your stack.yaml’s extra-deps and enable Docker integration so that your binary is automatically compiled in a compatible environment for AWS. Also add hal to your project’s dependency list (either project-name.cabal or package.yaml)

#...
extra-deps:
  - hal-${DESIRED_VERSION}
# ...
docker:
  enable: true
# ...

Then, define your types and handler:

{-# LANGUAGE DeriveGeneric  #-}
{-# LANGUAGE NamedFieldPuns #-}

module Main where

import AWS.Lambda.Runtime (pureRuntime)
import Data.Aeson         (FromJSON, ToJSON)
import GHC.Generics       (Generic)

data IdEvent  = IdEvent { input   :: String } deriving Generic
instance FromJSON IdEvent

data IdResult = IdResult { output :: String } deriving Generic
instance ToJSON IdResult

handler :: IdEvent -> IdResult
handler IdEvent { input } = IdResult { output = input }

main :: IO ()
main = pureRuntime handler

Your binary should be called bootstrap in order for the custom runtime to execute properly:

# Example snippet of package.yaml
# ...
executables:
  bootstrap:
    source-dirs: src
    main: Main.hs  # e.g. {project root}/src/Main.hs
# ...

You’ll need to either build on a compatible linux host or inside a compatible docker container (or some other mechanism like nix). Note that current Stack LTS images are not compatible. If you see an error message that contains “version ‘GLIBC_X.XX’ not found” when running (hosted or locally), then your build environment is not compatible.

Enable stack’s docker integration and define an optional image within stack.yaml:

# file: stack.yaml
docker:
  enabled: true
  # If omitted, this defaults to fpco/stack-build:lts-${YOUR_LTS_VERSION}
  image: ${BUILD_IMAGE}

Don’t forget to define your CloudFormation stack:

# file: template.yaml
AWSTemplateFormatVersion: '2010-09-09'
Transform: 'AWS::Serverless-2016-10-31'
Description: Test for the Haskell Runtime.
Resources:
  HelloWorldApp:
    Type: 'AWS::Serverless::Function'
    Properties:
      Handler: NOT_USED
      Runtime: provided
      # CodeUri is a relative path from the directory that this CloudFormation
      # file is defined.
      CodeUri: .stack-work/docker/_home/.local/bin/
      Description: My Haskell runtime.
      MemorySize: 128
      Timeout: 3

Finally, build, upload and test your lambda!

# Build the binary, make sure your executable is named `bootstrap`
stack build --copy-bins

# Create your function package
aws cloudformation package \
  --template-file template.yaml \
  --s3-bucket your-existing-bucket > \
  deployment_stack.yaml

# Deploy your function
aws cloudformation deploy \
  --stack-name "hello-world-haskell" \
  --region us-west-2 \
  --capabilities CAPABILITY_IAM \
  --template-file deployment_stack.yaml

# Take it for a spin!
aws lambda invoke \
  --function-name your-function-name \
  --region us-west-2 \
  --payload '{"input": "foo"}' \
  output.txt

Local Testing

Dependencies

• Stack
• Docker

Build

docker pull fpco/stack-build:lts-{version} # First build only, find the latest version in stack.yaml
stack build --copy-bins

Execute w/ Docker

echo '{ "accountId": "byebye" }' | docker run -i --rm \
    -e DOCKER_LAMBDA_USE_STDIN=1 \
    -v ${PWD}/.stack-work/docker/_home/.local/bin/:/var/task \
    lambci/lambda:provided

Execute w/ SAM Local

Note that hal currently only supports aws-sam-cli on versions <1.0.

echo '{ "accountId": "byebye" }' | sam local invoke --region us-east-1