AWS Is Asleep at the Lambda Wheel

Custom runtimes with Docker images solves so many issues with Lambda. Artifact management for free and the ability to run whatever language/applicaiton you want on Lambda. It works well, it's simple, it's portable. I can run my containers I deploy to Lambda on Lambda, Kubernetes, my local machine, pretty much anywhere if I abstract the AWS services i use.

Building a Python Docker image isn't sustainable as a customer? I don't understand that as most people package apps as containers already as a standard unit of deployment.Most large companies maintain their own base images. It looks like the only in-convenience for Python is doing `pip install awslambdaric`

With custom runtime Docker images I think AWS was smart, removing AWS maintenance as the limiting factor to let users have pretty much full control of the runtime, fairly trivially with tools they already know.

I worked in the past under a slew of former AWS product managers, and I know from them how the AWS roadmap works:

1.) Real, crucial customer feedback is incorporated into roadmap, usually for the core services that are already used extensively like Lambda. 2.) Executives have new, novel, sexy flavor of the month projects like "we need to catch up with that ChatGPT thing my kid was showing the Au Pair the other day!" 3.) Product Management leaders are incentivized to prioritize said new/novel projects, and proceed to do a worker crushing sprint to build what any sane person would call a PoC. 4.) PoC is barely tested and pushed out as a half-baked garbage product. Garbage product languishes. (See AWS Data Science Studio for a great example of giant, useless POS)

Meanwhile, developers are siphoned from real projects like, gee, I don't know, getting Python 3.10 out on Lambda? Or how about when you launch a sagemaker notebook instance you still top out at Python 3.8???? And don't get me started on recent versions of CUDA being available on GPU instances.

My team is relegated to docker containers deployed to lambda due to the antiquated, ridiculous limits on layer sizes for any of our ML oriented lambdas.

I have a buddy who works at AWS and talks about how utterly inefficient they are due to horrific leadership. At this point, you have a choice between a few highly dysfunctional cloud vendors. Google just SUCKS, Azure is a blatant financial rip off, and AWS is always developing new features nobody wants while ignoring the ones we do.