subscriptionless

GraphQL subscriptions using AWS Lambda and API Gateway Websockets

Usage no npm install needed!

<script type="module">
  import subscriptionless from 'https://cdn.skypack.dev/subscriptionless';
</script>

README

About

GraphQL subscriptions for AWS Lambda and API Gateway WebSockets.

Have all the functionality of GraphQL subscriptions on a stateful server without the cost.

Note: This project uses the graphql-ws protocol under the hood.

⚠️ Limitations

Seriously, read this first before you even think about using this.

This is in beta

This is beta and should be treated as such.

AWS API Gateway Limitations

There are a few noteworthy limitations to the AWS API Gateway WebSocket implementation.

Note: If you work on AWS and want to run through this, hit me up!

Socket timeouts

Default socket idleness detection in API Gateway is unpredictable.

It is strongly recommended to use socket idleness detection listed here. Alternatively, client->server pinging can be used to keep a connection alive.

Socket errors

API Gateway's current socket closing functionality doesn't support any kind of message/payload. Along with this, graphql-ws won't support error messages.

Because of this limitation, there is no clear way to communicate subprotocol errors to the client. In the case of a subprotocol error the socket will be closed by the server (with no meaningful disconnect payload).

Setup

Create a subscriptionless instance.

import { createInstance } from 'subscriptionless';

const instance = createInstance({
  schema,
});

Export the handler.

export const gatewayHandler = instance.gatewayHandler;

Configure API Gateway

Set up API Gateway to route WebSocket events to the exported handler.

💾 serverless framework example
functions:
  websocket:
    name: my-subscription-lambda
    handler: ./handler.gatewayHandler
    events:
      - websocket:
          route: $connect
      - websocket:
          route: $disconnect
      - websocket:
          route: $default
💾 terraform example
resource "aws_apigatewayv2_api" "ws" {
  name                       = "websocket-api"
  protocol_type              = "WEBSOCKET"
  route_selection_expression = "$request.body.action"
}

resource "aws_apigatewayv2_route" "default_route" {
  api_id    = aws_apigatewayv2_api.ws.id
  route_key = "$default"
  target    = "integrations/${aws_apigatewayv2_integration.default_integration.id}"
}

resource "aws_apigatewayv2_route" "connect_route" {
  api_id    = aws_apigatewayv2_api.ws.id
  route_key = "$connect"
  target    = "integrations/${aws_apigatewayv2_integration.default_integration.id}"
}

resource "aws_apigatewayv2_route" "disconnect_route" {
  api_id    = aws_apigatewayv2_api.ws.id
  route_key = "$disconnect"
  target    = "integrations/${aws_apigatewayv2_integration.default_integration.id}"
}

resource "aws_apigatewayv2_integration" "default_integration" {
  api_id           = aws_apigatewayv2_api.ws.id
  integration_type = "AWS_PROXY"
  integration_uri  = aws_lambda_function.gateway_handler.invoke_arn
}

resource "aws_lambda_permission" "apigateway_invoke_lambda" {
  action        = "lambda:InvokeFunction"
  function_name = aws_lambda_function.gateway_handler.function_name
  principal     = "apigateway.amazonaws.com"
}

resource "aws_apigatewayv2_deployment" "ws" {
  api_id = aws_apigatewayv2_api.ws.id

  triggers = {
    redeployment = sha1(join(",", tolist([
      jsonencode(aws_apigatewayv2_integration.default_integration),
      jsonencode(aws_apigatewayv2_route.default_route),
      jsonencode(aws_apigatewayv2_route.connect_route),
      jsonencode(aws_apigatewayv2_route.disconnect_route),
    ])))
  }

  depends_on = [
    aws_apigatewayv2_route.default_route,
    aws_apigatewayv2_route.connect_route,
    aws_apigatewayv2_route.disconnect_route
  ]
}

resource "aws_apigatewayv2_stage" "ws" {
  api_id        = aws_apigatewayv2_api.ws.id
  name          = "example"
  deployment_id = aws_apigatewayv2_deployment.ws.id
}

Create DynanmoDB tables for state

In-flight connections and subscriptions need to be persisted.

📖 Changing DynamoDB table names

Use the tableNames argument to override the default table names.

const instance = createInstance({
  /* ... */
  tableNames: {
    connections: 'my_connections',
    subscriptions: 'my_subscriptions',
  },
});
💾 serverless framework example
resources:
  Resources:
    # Table for tracking connections
    connectionsTable:
      Type: AWS::DynamoDB::Table
      Properties:
        TableName: ${self:provider.environment.CONNECTIONS_TABLE}
        AttributeDefinitions:
          - AttributeName: id
            AttributeType: S
        KeySchema:
          - AttributeName: id
            KeyType: HASH
        TimeToLiveSpecification:
          AttributeName: ttl
          Enabled: true
        ProvisionedThroughput:
          ReadCapacityUnits: 1
          WriteCapacityUnits: 1
    # Table for tracking subscriptions
    subscriptionsTable:
      Type: AWS::DynamoDB::Table
      Properties:
        TableName: ${self:provider.environment.SUBSCRIPTIONS_TABLE}
        AttributeDefinitions:
          - AttributeName: id
            AttributeType: S
          - AttributeName: topic
            AttributeType: S
          - AttributeName: connectionId
            AttributeType: S
        KeySchema:
          - AttributeName: id
            KeyType: HASH
          - AttributeName: topic
            KeyType: RANGE
        GlobalSecondaryIndexes:
          - IndexName: ConnectionIndex
            KeySchema:
              - AttributeName: connectionId
                KeyType: HASH
            Projection:
              ProjectionType: ALL
            ProvisionedThroughput:
              ReadCapacityUnits: 1
              WriteCapacityUnits: 1
          - IndexName: TopicIndex
            KeySchema:
              - AttributeName: topic
                KeyType: HASH
            Projection:
              ProjectionType: ALL
            ProvisionedThroughput:
              ReadCapacityUnits: 1
              WriteCapacityUnits: 1
        TimeToLiveSpecification:
          AttributeName: ttl
          Enabled: true
        ProvisionedThroughput:
          ReadCapacityUnits: 1
          WriteCapacityUnits: 1
💾 terraform example
resource "aws_dynamodb_table" "connections-table" {
  name           = "subscriptionless_connections"
  billing_mode   = "PROVISIONED"
  read_capacity  = 1
  write_capacity = 1
  hash_key = "id"

  attribute {
    name = "id"
    type = "S"
  }

  ttl {
    attribute_name = "ttl"
    enabled        = true
  }
}

resource "aws_dynamodb_table" "subscriptions-table" {
  name           = "subscriptionless_subscriptions"
  billing_mode   = "PROVISIONED"
  read_capacity  = 1
  write_capacity = 1
  hash_key = "id"
  range_key = "topic"

  attribute {
    name = "id"
    type = "S"
  }

  attribute {
    name = "topic"
    type = "S"
  }

  attribute {
    name = "connectionId"
    type = "S"
  }

  global_secondary_index {
    name               = "ConnectionIndex"
    hash_key           = "connectionId"
    write_capacity     = 1
    read_capacity      = 1
    projection_type    = "ALL"
  }

  global_secondary_index {
    name               = "TopicIndex"
    hash_key           = "topic"
    write_capacity     = 1
    read_capacity      = 1
    projection_type    = "ALL"
  }

  ttl {
    attribute_name = "ttl"
    enabled        = true
  }
}

Configure idleness detection (ping/pong)

Set up server->client pinging for socket idleness detection.

Note: While not a hard requirement, this is strongly recommended.

📖 Configuring instance

Pass a ping argument to configure delays and what state machine to invoke.

const instance = createInstance({
  /* ... */
  ping: {
    interval: 60, // Rate in seconds to send ping message
    timeout: 30, // Threshold for pong response before closing socket
    machineArn: process.env.MACHINE_ARN, // State machine to invoke
  },
});

Export the resulting handler for use by the state machine.

export const stateMachineHandler = instance.stateMachineHandler;
💾 serverless framework example

Create a function which exports the aforementioned machine handler.

functions:
  machine:
    handler: src/handler.stateMachineHandler

Use the serverless-step-functions plugin to create a state machine which invokes the machine handler.

stepFunctions:
  stateMachines:
    ping:
      role: !GetAtt IamRoleLambdaExecution.Arn
      definition:
        StartAt: Wait
        States:
          Eval:
            Type: Task
            Resource: !GetAtt machine.Arn
            Next: Choose
          Wait:
            Type: Wait
            SecondsPath: '$.seconds'
            Next: Eval
          Choose:
            Type: Choice
            Choices:
              - Not:
                  Variable: '$.state'
                  StringEquals: 'ABORT'
                Next: Wait
            Default: End
          End:
            Type: Pass
            End: true

The state machine arn can be passed to your websocket handler function via outputs.

Note: naming of resources will be dependent the function/machine naming in the serverless config.

functions:
  subscription:
    handler: src/handler.gatewayHandler
    environment:
      PING_STATE_MACHINE_ARN: ${self:resources.Outputs.PingStateMachine.Value}
    # ...

resources:
  Outputs:
    PingStateMachine:
      Value:
        Ref: PingStepFunctionsStateMachine

On connection_init, the state machine will be invoked. Ensure that the websocket handler has the following permissions.

- Effect: Allow
  Resource: !GetAtt PingStepFunctionsStateMachine.Arn
  Action:
    - states:StartExecution

The state machine itself will need the following permissions

- Effect: Allow
  Resource: !GetAtt connectionsTable.Arn
  Action:
    - dynamodb:GetItem
    - dynamodb:UpdateItem
- Effect: Allow
  Resource: '*'
  Action:
    - execute-api:*

Note: For a full reproduction, see the example project.

💾 terraform example

Create a function which can be invoked by the state machine.

resource "aws_lambda_function" "machine" {
  function_name    = "machine"
  runtime          = "nodejs14.x"
  filename         = data.archive_file.handler.output_path
  source_code_hash = data.archive_file.handler.output_base64sha256
  handler          = "example.stateMachineHandler"
  role             = aws_iam_role.state_machine_function.arn

  environment {
    variables = {
      CONNECTIONS_TABLE   = aws_dynamodb_table.connections.id
      SUBSCRIPTIONS_TABLE = aws_dynamodb_table.subscriptions.id
    }
  }
}

Create the following state machine which will be invoked by the gateway handler.

resource "aws_sfn_state_machine" "ping_state_machine" {
  name     = "ping-state-machine"
  role_arn = aws_iam_role.state_machine.arn
  definition = jsonencode({
    StartAt = "Wait"
    States = {
      Wait = {
        Type        = "Wait"
        SecondsPath = "$.seconds"
        Next        = "Eval"
      }
      Eval = {
        Type     = "Task"
        Resource = aws_lambda_function.machine.arn
        Next     = "Choose"
      }
      Choose = {
        Type = "Choice"
        Choices = [{
          Not = {
            Variable     = "$.state"
            StringEquals = "ABORT"
          }
          Next = "Wait"
        }]
        Default = "End"
      }
      End = {
        Type = "Pass"
        End  = true
      }
    }
  })
}

The state machine arn can be passed to your websocket handler via an environment variable.

resource "aws_lambda_function" "gateway_handler" {
  # ...

  environment {
    variables = {
      # ...
      PING_STATE_MACHINE_ARN = aws_sfn_state_machine.ping_state_machine.arn
    }
  }
}

Note: For a full reproduction, see the example project.

Usage

PubSub

subscriptionless uses it's own PubSub implementation which loosely implements the Apollo PubSub Interface.

Note: Unlike the Apollo PubSub library, this implementation is (mostly) stateless

📖 Subscribing to topics

Use the subscribe function to associate incoming subscriptions with a topic.

import { subscribe } from 'subscriptionless/subscribe';

export const resolver = {
  Subscribe: {
    mySubscription: {
      resolve: (event, args, context) => {/* ... */}
      subscribe: subscribe('MY_TOPIC'),
    }
  }
}
📖 Filtering events

Wrap any subscribe function call in a withFilter to provide filter conditions.

Note: If a function is provided, it will be called on subscription start and must return a serializable object.

import { withFilter, subscribe } from 'subscriptionless/subscribe';

// Subscription agnostic filter
withFilter(subscribe('MY_TOPIC'), {
  attr1: '`attr1` must have this value',
  attr2: {
    attr3: 'Nested attributes work fine',
  },
});

// Subscription specific filter
withFilter(subscribe('MY_TOPIC'), (root, args, context, info) => ({
  userId: args.userId,
}));
📖 Concatenating topic subscriptions

Join multiple topic subscriptions together using concat.

import { concat, subscribe } from 'subscriptionless/subscribe';

concat(subscribe('TOPIC_1'), subscribe('TOPIC_2'));
📖 Publishing events

Use the publish on your subscriptionless instance to publish events to active subscriptions.

instance.publish({
  type: 'MY_TOPIC',
  payload: 'HELLO',
});

Events can come from many sources

// SNS Event
export const snsHandler = (event) =>
  Promise.all(
    event.Records.map((r) =>
      instance.publish({
        topic: r.Sns.TopicArn.substring(r.Sns.TopicArn.lastIndexOf(':') + 1), // Get topic name (e.g. "MY_TOPIC")
        payload: JSON.parse(r.Sns.Message),
      })
    )
  );

// Manual Invocation
export const invocationHandler = (payload) =>
  instance.publish({ topic: 'MY_TOPIC', payload });

Context

Context values are accessible in all resolver level functions (resolve, subscribe, onSubscribe and onComplete).

📖 Default value

Assuming no context argument is provided, the default value is an object containing a connectionParams attribute.

This attribute contains the (optionally parsed) payload from connection_init.

export const resolver = {
  Subscribe: {
    mySubscription: {
      resolve: (event, args, context) => {
        console.log(context.connectionParams); // payload from connection_init
      },
    },
  },
};
📖 Setting static context value

An object can be provided via the context attribute when calling createInstance.

const instance = createInstance({
  /* ... */
  context: {
    myAttr: 'hello',
  },
});

The default values (above) will be appended to this object prior to execution.

📖 Setting dynamic context value

A function (optionally async) can be provided via the context attribute when calling createInstance.

The default context value is passed as an argument.

const instance = createInstance({
  /* ... */
  context: ({ connectionParams }) => ({
    myAttr: 'hello',
    user: connectionParams.user,
  }),
});

Side effects

Side effect handlers can be declared on subscription fields to handle onSubscribe (start) and onComplete (stop) events.

📖 Enabling side effects

For onSubscribe and onComplete side effects to work, resolvers must first be passed to prepareResolvers prior to schema construction.

import { prepareResolvers } from 'subscriptionless/subscribe';

const schema = makeExecutableSchema({
  typedefs,
  resolvers: prepareResolvers(resolvers),
});
📖 Adding side-effect handlers
export const resolver = {
  Subscribe: {
    mySubscription: {
      resolve: (event, args, context) => {
        /* ... */
      },
      subscribe: subscribe('MY_TOPIC'),
      onSubscribe: (root, args) => {
        /* Do something on subscription start */
      },
      onComplete: (root, args) => {
        /* Do something on subscription stop */
      },
    },
  },
};

Events

Global events can be provided when calling createInstance to track the execution cycle of the lambda.

📖 Connect (onConnect)

Called on an incoming API Gateway $connect event.

const instance = createInstance({
  /* ... */
  onConnect: ({ event }) => {
    /* */
  },
});
📖 Disconnect (onDisconnect)

Called on an incoming API Gateway $disconnect event.

const instance = createInstance({
  /* ... */
  onDisconnect: ({ event }) => {
    /* */
  },
});
📖 Authorization (connection_init)

Called on incoming graphql-ws connection_init message.

onConnectionInit can be used to verify the connection_init payload prior to persistence.

Note: Any sensitive data in the incoming message should be removed at this stage.

const instance = createInstance({
  /* ... */
  onConnectionInit: ({ message }) => {
    const token = message.payload.token;

    if (!myValidation(token)) {
      throw Error('Token validation failed');
    }

    // Prevent sensitive data from being written to DB
    return {
      ...message.payload,
      token: undefined,
    };
  },
});

By default, the (optionally parsed) payload will be accessible via context.

📖 Subscribe (onSubscribe)

Subscribe (onSubscribe)

Called on incoming graphql-ws subscribe message.

const instance = createInstance({
  /* ... */
  onSubscribe: ({ event, message }) => {
    /* */
  },
});
📖 Complete (onComplete)

Called on graphql-ws complete message.

const instance = createInstance({
  /* ... */
  onComplete: ({ event, message }) => {
    /* */
  },
});
📖 Ping (onPing)

Called on incoming graphql-ws ping message.

const instance = createInstance({
  /* ... */
  onPing: ({ event, message }) => {
    /* */
  },
});
📖 Pong (onPong)

Called on incoming graphql-ws pong message.

const instance = createInstance({
  /* ... */
  onPong: ({ event, message }) => {
    /* */
  },
});
📖 Error (onError)

Called on unexpected errors during resolution of API Gateway or graphql-ws events.

const instance = createInstance({
  /* ... */
  onError: (error, context) => {
    /* */
  },
});