Skip to content

Latest commit

 

History

History
77 lines (60 loc) · 3.95 KB

README.md

File metadata and controls

77 lines (60 loc) · 3.95 KB
Raw

Note

This V2 package is to resolve the 503 error issue, which is usually caused by the 30s timeout limitation of HTTP API gateway

Changes

  • Architecture change
  1. Add a WebSocket API Gateway, which is used to setup a long connection between client and backend.
  2. Decouple the chat function by using AWS SNS. Now OpenAI's API usually takes more than 30s to generate the response text, so that we cannot using HTTP API Gateway to trigger that function, because the HTTP API gateway has timeout limition of 30s.
    architecture-v2

Setup

Create Lambda functions

We will create 4 lambda functions: (Code files are in four sub folers under v2/server/) The detail steps are the same as previous v1. please refer to v1 README.md .

  1. lambda_login
    Tips:
  • This lambda is integrated with the /login route of HTTP API, and perform the user validation.
  • The execution role of this lambda requires read permission of DynamoDB tables. You need to attach the DynamodDB access policy accordingly
  • Need to configure an environment variable named TOKEN_KEY.
  1. lambda_connect_handle
    Tips:
  • This lambda is integrated with the $connect route of WebSocket API, and perform the token authorization.
  • Need to configure the same TOKEN_KEY environment variable as lambda_login.
  1. lambda_handle_chat
    Tips:
  • This lambda is integrated with the sendprompt route of WebSocket API, and send the connectionId and the prompt from user input to a SNS.
  • The execution role of this lambda requires SNS publish permission. You need to attach the SNS policy accordingly.
  • Need to configure an environment variable named SNS_TOPIC_ARN, and fill the arn of your SNS topic ( Creation steps will be guided in later)
  1. lambda_chat
    Tips:
  • This lambda is triggered by SNS, so it will subscribe SNS topic later.
  • It calls OpenAI API to get the response text, and send back the text via the WebSocket API Gateway, so it needs to be attached with the permission policy of AmazonAPIGatewayInvokeFullAccess.
  • Change the lambda timeout value to a bigger one, for example 5 mins.
  • Configure your own OpenAI API key in environment variables as key named "OPENAI_API_KEY".

Create a HTTP API gateway

The detail steps are the same as previous v1. The only difference is that it has /login route. please refer to v1 README.md configure the route for /login part.

Create a WebSocket API gateway

  1. Create a WebSocket API gateway from console
    wsapigw-1
  2. Add route key $connect and sendprompt
    wsapigw-2
  3. Add integration to the lambdas accordingly wsapigw-3
  4. Get your WSS endpoint wsapigw-4

Create SNS topic

  1. Create a Standard SNS topic
    sns-1
  2. Create a subscription, choose lambda in policy field, and past your arn link of lambda_chat to the endpoint field.
    sns-2
  3. Copy the sns topic arn and paste to the environment variable's value of SNS_TOPIC_ARN of lambda_handle_chat sns-3

Build the client

  1. Change the API_http and API_socket in code apigw.js, to your actual endpoints of HTTP API GW and WebSocket GW accordingly. export const API_socket = 'wss://{apiid}.execute-api.{region}.amazonaws.com/dev'; export const API_http = 'https://{apiid}.execute-api.{region}.amazonaws.com';

  2. Run npm install and npm run build. The detail steps are the same as previous v1. please refer to v1 README.md .

Upload the static files to S3 bucket.

  1. Setup the s3 bucket with static website hosting enable
  2. Upload the files in build folder to the bucket. The detail steps are the same as previous v1. please refer to v1 README.md .

After all these done, Congrats!