Discord slash commands in AWS Lambda

Rationale

Serverless is becoming more and more popular. This kind of infrastructure is not only very budget friendly, but also hugely scalable and once set up requiring almost 0 maintenance. Recently, discord has introduced a new kind of feature, interactions. The only kind of interactions at the moment are slash commands. Interactions can be configured to act as post requests to custom endpoints. This repository is a template for AWS Lambda functions serving as endpoints for discord to use.

Downsides

So far I've mostly only mentioned the upsides of a serverless approach. It's worth considering the downsides of this approach as well, amongst them are

• Latency. Cold starts are performance killers & some API limitations force us to employ less than optimal solutions.
• Versatility. Slash commands essentially only provide us with some input data and a context to send webhooks. Many features a conventional bot might implement are therefore unavailable.
• Working with AWS services in general can be a bit tedious

Setup

In this section we'll be creating 2 lambda functions. To understand why that is necessary, please read the Usage section.

Discordgo

This template uses discordgo for its types & some AWS sdk libraries. As of today, 2021-02-16, discordgo hasn't merged support for slash commands yet. As I cannot get Go's packages to work, I've done some black magic and merged the relevant branch myself locally.

cd $(go env GOPATH)/src/github.com/bwmarrin/discordgo
git remote rm origin
git remote add origin https://github.com/FedorLap2006/discordgo
git pull origin slashes

In the future just using discordgo normally should suffice.

Building

To build our lambda code, we just invoke

go build lambda-task/main.go
go build lambda-endpoint/main.go

To use them in lambda, we need to

• set some enviromental variables, e.g. GOOS=linux and CGO_ENABLED=0
• zip the binaries

The bash script build.sh does that for us and conveniently places the resulting zip files in an out folder.

AWS

These are the steps which can be used to get lambda running. If you're already experienced with lambda, you can change these as you see fit. I'm using the aws cli for clarity & efficiency's sake.

First we create a policy for our first lambda. Pay attention to the ARN we get back. We're gonna need it.

aws iam create-policy --policy-name LambdaInvokeAll --policy-document file://trust-policy.json

Now, we create an execution role for our first and second lambda function. We are again going to need the returned ARNs

aws iam create-role \
--role-name DiscordEndpointRole \
--assume-role-policy-document file://trust-policy.json

aws iam create-role \
--role-name DiscordTaskRole \
--assume-role-policy-document file://trust-policy.json

And attach the needed policies

aws iam attach-role-policy \
--role-name DiscordTaskRole \
--policy-arn arn:aws:iam::aws:policy/service-role/AWSLambdaBasicExecutionRole

aws iam attach-role-policy \
--role-name DiscordEndpointRole \
--policy-arn arn:aws:iam::aws:policy/service-role/AWSLambdaBasicExecutionRole

aws iam attach-role-policy \
--role-name DiscordEndpointRole \
--policy-arn {OUR INVOCATION ARN}

Lastly, we create the actual lambda functions.

aws lambda create-function --function-name DiscordEndpoint \
--handler main --runtime go1.x \
--role {Endpoint Role ARN} \
--zip-file fileb://out/endp.zip

aws lambda create-function --function-name DiscordTask \
--handler main --runtime go1.x \
--role {Task Role ARN} \
--zip-file fileb://out/task.zip \
--timeout 900

We also need to add a trigger to the endpoint lambda, there's probably a complicated way to do this from the cli, but I've just used the web interface. Make sure to choose API Gateway -> Create an API -> HTTP API and choose Security=Open

Now, under details you'll see the API endpoint. Copy that link into discord.

Configuration

We now need to set all the correct values in config/config.go. Appid and Publickey can both be found in the discord overview of our application. For the Bottoken we need to create a bot and use its token. For the Tasklambda value use the name of our DiscordTask function.

After recompiling, to update our code we can use these commands (Note: these do take a few seconds to update):

aws lambda update-function-code --function-name DiscordEndpoint \
--zip-file fileb://out/endp.zip 

aws lambda update-function-code --function-name DiscordTask \ 
--zip-file fileb://out/task.zip 

Which I've put into update.sh as well. (Note: update.sh also builds the lambda code and cli, more on that later)

Usage

After you've set up AWS (by either following my setup instructions or doing your own thing), you can start developing & testing your application.

Program structure

All discord specific types are supplied by discordgo. (Or rather PR 856 which hasn't been merged as of now - see the Setup)

The directory commands, or equivalently the package, contains a Commands map, which maps a string (the command name) to a Command struct as defined bellow

type Command struct {
	Command      discordgo.ApplicationCommand
	Handler      HandlerSig
	Continuation ContinuationSig
}

To add a command add an entry to that map. For example:

"timer": {
	Command: discordgo.ApplicationCommand{
		Name:        "timer",
		Description: "Sets a timer. Tick tock, tick tock...",
		Options: []*discordgo.ApplicationCommandOption{
			{
				Type:        discordgo.ApplicationCommandOptionInteger,
				Name:        "Seconds",
				Description: "How many seconds to set the timer for",
				Required:    true,
			},
		},
	},
	Handler: AckSourceContinue,
	Continuation: Timer,
},

On import, the module checks that each key matches the Name field in the accompanying ApplicationCommand struct. The Handler field is a function with the signature

func(*discordgo.Interaction) (discordgo.InteractionResponse, bool, error)

The handler not being nil is checked during import. The input is the interaction received from discord. It returns an interaction response, a boolean signifying whether to continue (more below) and an error. If the error is not nil the server will return 500 (Everything is logged in Lambda - including the error).

The Discord API requires us to respond within 3 seconds of receiving the interaction. Any tasks longer than that are delegated to a second lambda. If your initial handler returns true for its boolean return value, the endpoint lambda will invoke the second task lambda (using its name as defined in the config struct (config/config.go)). The lambda is simply passed the interaction struct. It's not possible to pass other custom data to it (you're welcome to modify the code and/or issue a PR). The task lambda then invokes the Continuation field. If it's nil the lambda will fail, logging any errors. (But since this separate from the HTTP endpoint lambda failing does not affect discord) The signature for the continuation handler is:

func(*discordgo.Interaction) error

If error isn't nil the lambda will fail, logging the error.

To add custom commands you need to

1. Write a handler (and optionally a continuation handler)
2. Add the command to the Commands map
3. Recompile and update the code of both lambdas
4. Update the commands on discords side - see the CLI

Note that since your lambdas are standalone and each command is equal to an invocation, you can automate the testing of your commands pretty easily.

Also note that slash commands by themselves only actually allow us to receive commands and send responses. AWS Lambda has a hard limit for runtime at 15 minutes, making many tasks such as playing music hugely infeasible. On the other hand, you obviously can interact with discord in Lambda using discordgo, REST API calls, or any other way you see fit. (Do note that for many discordgo operations you conditionally need to have called the Open function for them to work)

CLI

To actually use slash commands, you must update them on discords side. Obviously, this could be done using curl, python, etc. However since all of your commands are already defined in commands/commands.go, I have added a CLI to automate some tasks. (The code is pretty messy & could be improved upon, although it suffices at its job, feel free to open an issue or PR)

Note: to upload your commands you need to recompile the CLI

There are currently 4 actions the CLI can perform

• list
• delete
• show
• upload

The CLI also takes in these flags:

• -commandid: Specifies the command's ID where applicable
• -guildid: Specifies the guild's ID where applicable
• -all: Specifies for an operation to be performed on all commands

Example uses:

List all guild commands:

./slashes -guildid={ID} list   

Show a command:

./slashes -commandid={ID} show

Delete all global commands:

./slashes -all delete

Upload your commands globally: (Upload will create a command anew OR modify an already existing one if the names match):

./slashes -guildid=784518984947073025 upload

To upload globally just omit the -guildid flag.

Note: You need to have config/config.go configured correctly for the CLI to work

Note: order matters. ./slashes -guildid={ID} list will word whereas ./slashes list -guildid={ID} won't