Everything comes with a trade off. What are some of the downsides?

* Function "sprawl". Managing lots of small functions

* AWS Support Dependency. IMO way better to have amazon managing more but it still increases reliance on amazon and support tickets

* Limited set of input events available. For example it's difficult to use rabbitmq but lambda orchestrates batching and orchestration for SQS

I didn't really mention these because I think they are better than the alternatives :)

I would personally rather manage a bunch of small functions leveraging amazon's documentation, community, clear interfaces, and SDK, than a random OSS async worker framework or homegrown processing system.

Companies that have committed to AWS are already dependent on AWS support, I haven't really found an issue with being more dependent on AWS support :p

The limited set of events has also been a net positive. There are few clearly defined options. I have found the limited choices helps delivery and simplifies architecture; less discussion. Also once a company learns SQS and kinesis, that knowledge is applicable everywhere, since those are the constrained choices.

-----

I would love to hear any experiences where lambda was not a good fit, or did not work out

Can you stop to call it serverless? It's like for the cloud, there are servers, just you don't have any control over the servers running your snippets.

Azure functions is seems better to me, it supports multiple hooks integration