Serverless Haskell

Deploying Haskell code onto AWS Lambda as native runtime using Serverless.

There are two ways to start, either via the stack template, or directly modifying a project. You may want to use the manual approach as the template specifies a specific stack resolver as it needs to hardcode the stack.yaml file.

In either case, you will want to have Serverless installed, eg. npm install -g serverless.

Using the stack template

Create a Stack package for your code:

stack new mypackage https://raw.githubusercontent.com/seek-oss/serverless-haskell/master/serverless-haskell.hsfiles

Update the resolver in the stack.yaml file. This is hardcoded as the resolver number is not known at template interpolation time. You should pick either the latest resolver, or one you have used before and have thus prebuilt many of the core packages for.
Install the dependencies and build the project:
```
cd mypackage
npm install
stack build
sls invoke local -f mypackage-func
```
This should invoke serverless locally and display output once everything has built.

Manually

Create a Stack package for your code:
```
stack new mypackage
```
LTS 10-17 are supported, older versions are likely to work too but untested.
Initialise a Serverless project inside the Stack package directory and install the serverless-haskell plugin:
```
cd mypackage
npm init -y
npm install --save serverless serverless-haskell@x.y.z
```
The version of the NPM package to install must match the version of the Haskell package.

Create serverless.yml with the following contents:

service: myservice
provider:
  name: aws
  runtime: haskell
functions:
  myfunc:
    handler: mypackage.mypackage-exe
    # Here, mypackage is the Haskell package name and mypackage-exe is the
    # executable name as defined in the Cabal file. The handler field may be
    # prefixed with a path of the form `dir1/.../dirn`, relative to
    # `serverless.yml`, which points to the location where the Haskell
    # package `mypackage` is defined. This prefix is not needed when the
    # Stack project is defined at the same level as `serverless.yml`.
plugins:
  - serverless-haskell

Write your main function:

import qualified Data.Aeson as Aeson
import AWSLambda
main = lambdaMain handler
handler :: Aeson.Value -> IO [Int]
handler evt = do
  putStrLn "This should go to logs"
  print evt
  pure [1, 2, 3]

Add aeson and serverless-haskell to package.yaml:

dependencies:
- base >= 4.7 && < 5
- aeson
- serverless-haskell

Build and test locally using sls invoke local:

The serverless-haskell plugin will build the package using Stack. Note that the first build can take a long time. Consider adding export SLS_DEBUG=* so you can see what is happening.
```
export SLS_DEBUG=*
sls invoke local -f myfunc
```
Use sls deploy to deploy the executable to AWS Lambda.

The serverless-haskell plugin will build the package using Stack, then upload it to AWS together with a JavaScript wrapper to pass the input and output from/to AWS Lambda.
```
export SLS_DEBUG=*
sls deploy
```
You can test the function and see the invocation results with:
```
sls invoke -f myfunc`
```

API Gateway

This plugin supports handling API Gateway requests. Declare the HTTP events normally in serverless.yml and use AWSLambda.Events.APIGateway in the handler to process them.

Serverless Offline can be used for local testing of API Gateway requests. You must use --useDocker flag so that the native Haskell runtime works correctly.

When using Serverless Offline, make sure that the project directory is world-readable, otherwise the started Docker container will be unable to access the handlers and all invocations will return HTTP status 502.

Notes

Only AWS Lambda is supported at the moment. Other cloud providers would require different JavaScript wrappers to be implemented.

See AWSLambda for documentation, including additional options to control the deployment.

Development

master branch is the stable version. It is normally released to Hackage once new changes are merged via Git tags.

The package is also maintained in Stackage LTS, provided the dependencies are not blocking it.

Testing

Haskell code is tested with Stack: stack test.
TypeScript code is linted with eslint.

Integration tests

Integration test verifies that the project can build and deploy a complete function to AWS, and it runs with expected functionality.

Integration test is only automatically run up to deployment due to the need for an AWS account. To run manually:

Ensure you have the required dependencies:
- curl
- jq
- NPM
- pkg-config
- pwgen
- Stack
Get an AWS account and add the access credentials into your shell environment.
Run ./integration-test/run.sh. The exit code indicates success.
To verify just the packaging, without deployment, run ./integration-test/run.sh --dry-run.
By default, the integration test is run with the LTS specified in stack.yaml. To specify a different series, use RESOLVER_SERIES=lts-9.
To avoid creating a temporary directory for every run, specify --no-clean-dir. This can speed up repeated test runs, but does not guarantee the same results as a clean test.

Releasing

Ensure you are on the master branch.
Ensure that all the changes are reflected in the changelog.
Run the integration tests.
Run ./bumpversion major|minor|patch. This will increment the version number, update the changelog, create and push the Git tag and the branch.
If you have released an LTS version, merge the version branch into master, taking care of the conflicts around version numbers and changelog, and release the latest version as well.