Hot off the press: aws.amazon.com/about-aws/…/aws-lambda-90-minute-function

I wonder if the pricing over time on using this will drive architects back to servers…. 🤔

90-minute Lambda function executions, on set-aside EC2 instances? So much for "serverless"! 😂 I was an early adopter of AWS Lambda and I still believe the total cost (development, operation, maintenance) is competitive, but what are users doing if they are running one Lambda function for 90 continuous minutes? I fear that raising the time limit from 15 to 90 minutes is an accommodation for large corporate customers that have run amok with Lambda and implemented monolithic functions (even this language is an oxymoron). I had a different but related concern about Lambda durable functions. On one hand, that feature raises the total execution time limit beyond 15 minutes (except for synchronous invocation, like an a source mapping from an SQS queue to a Lambda function) and makes interruptions/wait time free, and on the other hand, it requires developers to write code that is deeply dependent on an AWS-proprietary module/library. All of these changes make Lambda useful for more applications but at the expense of Lambda's advantages: simplicity and small units of work.

I agree. One of the key use cases they called out was AI inference; I guess folks were bumping up against the 15 min limit. But even then, I would be more inclined to have the function be a proxy to an AWS Batch job or AWS Step Function which puts the results on a queue for another slim function to pick up and report. yes, “monolithic functions” is a funny statement in deed! 😅

Worked with a few companies that were using Lambda functions to fetch so much data from APIs that they would timeout, but the answer was to move to a better solution like MWAA or an actual data engineering toolset. Agreed that it feels like an antipattern to accommodate this