PSA: Don't chase shiny.

Serverless stacks are fast, easy, and cheap... until they're not. 75% of the way through an implementation when the company started to realize that we would be done by now if we continued to use our own infrastructure.

"Serverless" (or FaaS, I still hate that name) is great for *some* use cases; they can really reduce costs and increase performance, especially for naturally spiky traffic. I don't really see an issue with development taking a long time either if the end result is cheaper, more maintainable and more scalable.

That said, there's *definitely* some non-obvious pitfalls that people fall into. Cold start times, managing DLQs, concurrency limits, the possibility of duplicate executions, the need for good logging, maximum timeouts, etc. are all things that should be taken into account well before development starts - and too often they're not.

I love this whole serverless bullshit.

Especially the name because it's so fucken wrong. A server serves something. So going up and asking a machine on the internet for data is essentially "asking a server". Nothing that makes this newass shit special makes it serverless.

I think this is just a marketing thing. Because this way you can sell thousands of units running on one machine, which wouldn't work performant enough for 24/7 use. Just my 2 cents.

It's like blockchain. You could use it for everything but you don't use the same cloth to wipe your ass and your desk, do you?

@nitwhiz "Serverless" has a terrible name, and I hate the marketing, but it's hard to argue with (for some use cases) from a pricing perspective. If I have a personal service that I'm playing around with and might hit a few times a month, paying just a few pence for that is really rather nice.

Likewise, for a commercial service that usually deals with sod all traffic, but occasionally has unpredictable peaks of a few million requests concurrently, it's an absolute godsend.

I am sorry but when someone says serverless I think a mailman on a cycle delivering mail, so no , serverless isn’t the most reliable solution not scalable.

I remember when one of software houses was trying to argue with us that we should take serverless option. I asked them for arguments and got whole lot of nothing. No, seriously, what the fuck, guys let my open my "why_own_infra.txt" and copare it to yours why not...

I asked 2 technical questions, pulled out calculator, figured out much worse costs (constant load), and counted it will be cheapper to use overkill dedicated server to load that if we ever achieve... Oh boi, i dont know if thats doable.

Yeah, i dont buy that bullshit like
"Cloud"
"Serverless"
"You pay only what you use"

Of course, your milage may vary, but remember, take your fucking calculators, and count that shit, before moving a finger a single millimetter.

Jeez, i feel today im more toxic than usual.
Peace.

@irene Not always. It's slashed our costs by more than 50%.

But applying it blindly is ridiculous, and yes, anyone that claims you can is scamming you.