Skip to content

Latest commit

 

History

History
250 lines (192 loc) · 9.67 KB

README.md

File metadata and controls

250 lines (192 loc) · 9.67 KB
Raw

AWS API Gateway Websocket Asynchronous Pusher

Illustration

Fast AWS API Gateway websockets notifications' pusher using Python AsyncIO for managing asynchronous and concurrent non-blocking IO calls to DynamoDB connections store and API Gateway management API. making it suitable for broadcasting messages to multiple users with a fast and cost-effective approach.

You can test the Pusher with this demo application or use the live demo at https://sumu.kodhive.com

Features

  • Multiple sources: capable of receiving notification requests from SQS and SNS, which make it suitable for processing single notification request from SNS and high number of notification requests in batches when polling from SQS

  • Multicast Notifications: the pusher can send messages to only a subset of users (eg: users in a chat room).

  • Broadcast Notifications: sometimes you just want to send the same message for all connected users (eg: broad announcements)

  • Exclusion Notifications: the pusher can broadcast messages to all users except a list of excluded users (eg: online/offline presence events can be sent to all except the originator)

  • Stale Connections Pruning: capable of detecting and deleting stale connections from DynamoDB connections store in case API Gateway missed cleaning them.

  • Asynchronous Processing: the pusher is using AsyncIO to notify multiple users/connections concurrently to not wait for inflight requests to DynamoDB/API Gateway so you don't pay AWS the waiting time 😎

  • Batch Processing: when using SQS as event source the Pushed will be able to process batches of notification requests, also concurrently.

  • Duplicate Users Detection: able to detect duplicate users in a notification requests and make them unique set of users. to avoid double notifications.

Requirements

Illustration

  • DynamoDB Connections Store: a DynamoDB table where users connections are stored, It should have a hash key user_id and a range key connection_id, where every user can have multiple connections and the Pusher can retrieve specific user connections by querying with the hash key.

  • AWS API Gateway Websockets API: an already configured Websocket API Gateway. the Pusher will use the connection management endpoint passed as environment variable to post messages to users.

  • SNS Topic: the SNS Topic that the pusher will subscribe to for notifications requests.

  • SQS Queue: the SQS Queue that the pusher will poll notifications requests from.

Notification Requests

Backend applications can send notifications to AWS API Gateway Websocket connected users by sending a notification request to the service integrated with the Pusher (SNS|SQS), notifications requests should meet the following format:

For Multicast notifications, the message should be a JSON String that contains the list of users and the actual data:

import json

message = {
    "users": ["783304b1-2320-44db-8f58-09c3035a686b", "a280aa41-d99b-4e1c-b126-6f39720633cc"],
    "data": {"type": "notification", "message": "A message sent to multiple user"}
}
message_to_send = json.dumps(message)

For broadcast notifications, the same but do not provide users list or provide an empty users list:

import json

message = {
    "data": {"type": "announcement", "message": "A broadcast to all users"}
}
message_to_send = json.dumps(message)

For exclusion notifications, instead of providing users list, provide a list of excluded users:

import json

message = {
    "exclude_users": ["783304b1-2320-44db-8f58-09c3035a686b"],
    "data": {
      "type": "announcement", 
      "message": {
        "user_id": "783304b1-2320-44db-8f58-09c3035a686b", 
        "status": "OFFLINE"
      }
    }
}
message_to_send = json.dumps(message)

Integration

The Pusher can accept notifications requests from multiple sources like SQS and SNS. every source has its advantages and disadvantages when it comes to speed and cost.

Notification requests through SNS

The Pusher can subscribe to notifications SNS Topic, and whenever a backend applications Publish notification requests to SNS, the later will quickly notify the Pusher by sending the notification request to the subscribed Pusher Lambda.

This will result in a fast delivery because this approach does not introduce a polling mechanism and SNS will notify the Pusher whenever a notification request is available. however, at scale SNS will trigger a Pusher Lambda Function for every notification request and given that the Lambda Function Concurrent Invocations Limit is 1,000 per account (Can be increased to 100,000 by support-ticket) notification requests will be throttled for large applications.

Publish to SNS when you have small application with few users

import os
import json
import time
import boto3

message = {
    "users": ["783304b1-2320-44db-8f58-09c3035a686b", "a280aa41-d99b-4e1c-b126-6f39720633cc"],
    "data": {
        "type": "notification",
        "message": {
            "text": "Your order has been fulfilled!",
            "timestamp": int(time.time())
        }
    }
}
boto3.client("sns").sns.publish(
    TargetArn=os.environ["NOTIFICATIONS_TOPIC_ARN"],
    Message=json.dumps(message),
)

Sending notification requests through SQS

Unlike SNS, when sending notifications to SQS queue, the Pusher Lambda Function event source can be configured to poll notification requests from the SQS Queue, and it will periodically poll notification requests from the Queue using Polling Technique.

This will result in notifications requests to be processed in batches, which comes with many benefits:

  • Fewer Lambda Invocations - to not reach the Lambda Concurrency Limit.
  • Concurrent Notifications - as the pusher uses AsyncIO, it will be able to process batches of SQS Records concurrently.
  • Low cost - thanks to SQS Batches and fewer Lambda Invocations.

Pusher can meet the same speed and performance of SNS if the SQS queue receive_wait_time_seconds is set to 20. this will make the Lambda Service do Long Polling instead of Short Polling. In addition to that, Lambda service will have a background worker that has five instances polling every 20 seconds, this will ensure that the lambda will receive the notifications requests as soon as they arrive in the queue.

AWS: the automatic scaling behavior of Lambda is designed to keep polling costs low when a queue is empty while simultaneously letting us scale up to high throughput when the queue is being used heavily. When an SQS event source mapping is initially created and enabled, or when messages first appear after a period with no traffic, then the Lambda service will begin polling the SQS queue using five parallel long-polling connections. The Lambda service monitors the number of inflight messages, and when it detects that this number is trending up, it will increase the polling frequency by 20 ReceiveMessage requests per minute and the function concurrency by 60 calls per minute. As long as the queue remains busy it will continue to scale until it hits the function concurrency limits.

5 parallel connections, each will send 3 ReceiveMessage requests per minute which is 15 messages every minute. so 900 every hour, 21600 every day and 648,000 every month.

AWS gives you one million messages for free every month. After that it’s only $0.40 per million messages, so the cost is very low for consuming messages from SQS.

Send to SQS when you have a large application with millions of users

import os
import json
import time
import boto3

message = {
    "users": ["783304b1-2320-44db-8f58-09c3035a686b", "a280aa41-d99b-4e1c-b126-6f39720633cc"],
    "data": {
        "type": "notification",
        "message": {
            "text": "Your order has been fulfilled!",
            "timestamp": int(time.time())
        }
    }
}
boto3.client("sqs").send_message(
  QueueUrl=os.environ.get("NOTIFICATIONS_QUEUE_URL"),
  MessageBody=json.dumps(message),
)

Deploy

Before deploying the pusher, version control it on your github account and then call the Terraform module shipped with the Pusher to provision the Pusher Lambda Function, and the Terraform module to create CI/CD pipeline:

module "pusher" {
  source      = "./infra"
  prefix      = local.prefix
  common_tags = local.common_tags

  agma_arn       = "${aws_apigatewayv2_api._.execution_arn}/${aws_apigatewayv2_stage._.name}/POST/@connections"
  apigw_endpoint = "https://live.kodhive.com/push"
  # Custom API Gateway Domain
  connections_table = {
    name = aws_dynamodb_table.connections.name
    arn  = aws_dynamodb_table.connections.arn
  }
  notifications_topic_arn = aws_sns_topic.notifications.arn
  notifications_queue_arn = aws_sqs_queue.notifications.arn
}

module "pusher_ci" {
  source      = "git::https://github.com/obytes/terraform-aws-lambda-ci.git//modules/ci"
  prefix      = "${local.prefix}-pusher-ci"
  common_tags = var.common_tags

  # Lambda
  lambda                   = module.pusher.lambda
  app_src_path             = "src"
  packages_descriptor_path = "src/requirements/lambda.txt"

  # Github
  pre_release  = true
  s3_artifacts = {
    arn    = aws_s3_bucket.artifacts.arn
    bucket = aws_s3_bucket.artifacts.bucket
  }
  github = {
    owner          = "obytes"
    webhook_secret = "not-secret"
    connection_arn = "arn:aws:codestar-connections:us-east-1:{ACCOUNT_ID}:connection/{CONNECTION_ID}"
  }
  github_repository = {
    name   = "apigw-websocket-pusher"
    branch = "main"
  }
  # Notifications
  ci_notifications_slack_channels = {
    info  = "ci-info"
    alert = "ci-alert"
  }
}