Search code examples
amazon-web-servicesapilambdatimeoutgateway

Long call to aws api gateway

The problem I have is the following.

I currently have a system running in .NET. This system makes a call to a service which takes approximately 1 minute.

We are currently migrating the solution to AWS. And the problem I find is that the Lambda works runs in 1 minute (since it makes the call to the other system that takes 1 minute) and everything works fine. But when I make the call from the Gateway API, y have a timout. Investigating I found that it has a maximum timeout of 29 seconds.

Then I need to know what solution I can have to this problem, considering that i need wait 1 minute for the lambda function.

One that occurred to me is to trigger the call from the API, and that the lambda function runs, and from the client create a pool to see the status of the transaction. But I don't know how to keep the initial call "in memory" and when I call again the api to see the status, I know I'm talking about the same request, to get the result data

Solution

  • Here's an outline of one way to solve this:

    • the client makes a /start API request that triggers Lambda #1
    • Lambda #1 is short-lived and does the following:
      • generates a UUID as a correlator for the task about to be undertaken
      • creates a new item in DynamoDB, with the UUID as the key
      • triggers the long-lasting (1 minute) Lambda #2 to start, passing it the UUID
      • returns the UUID to the client
    • Lambda #2 is long-lived and does the following:
      • whatever work it needs to
      • periodically update its status and results to the UUID item in DynamoDB
    • the client can poll a /status?id=UUID API, on whatever schedule it likes, which triggers Lambda #3
    • Lambda #3 is short-lived and does the following:
      • query the UUID item from DynamoDB
      • return the current status and any results to the client

    When the /status?id=UUID API call indicates that the long-lived task is complete (or failed), the client can make a final API request to indicate that it has the result associated with the UUID and the DynamoDB item can be deleted, or you could just implement a TTL on the DynamoDB item.

    This process looks complicated, but it's really not.

    Rather than the client polling the back-end for status and results, it could alternately poll an SQS queue for the same, or subscribe to an SNS topic.