[qaq]

For the same amount of compute Lambda is priced far higher than Fargate which itself is priced higher than ec2. People run large workloads on k8s with base workload compute fully covered by dedicated ec2 instances not because it's fun, but because it saves you a lot of $.

[dijit]

Totally agree.

But it's the python argument.

Python is super slow, inefficient and let's say that the build environment is not so nice.

However, it's so easy to write functioning software in python that the alternative sometimes is not more efficient code, it's no code.

Lambda, in theory, follows a similar paradigm, if you can click a button and have a service that scales to zero (with logging and monitoring) then you're more likely to make toy webhooks and tiny services. If I have to make a build pipeline and a docker container and wrangle some yaml, configure service accounts and a service definition with the right labels and annotations.

Well, that's a decent chunk of work that means I'm probably going to think a bit longer about even deploying my little toy service that might see one request a day.

Going to reiterate though: I do not advocate for serverless in production. If you seriously think you're building something that will scale, it's fiscally illiterate to use a managed serverless provider.

[qaq]

If you under the impression that CI/CD and observability is easy with lambda I have a bridge to sell you. I worked on a large scale pure serverless project we wrote more CDK code than application code.

[dijit]

That’s sad, I saw a one click deploy button in my IDE and made an assumption.

What’s the point then if its not easier?

[qaq]

It scales to 0. Scaling to 0 and not having to write few lines Docker file is the only tangible benefits.