Serverless: Meme or not?

Is serverless just another fad? Is it good for anything? Do any of you actually use it? For what? Should I dive into it? Is it an employable skill?

I like the concept, I can get to use all the meme-tech from AWS, GCE and Azure, without having to write specific code for any of them.

Do you guys think that the serverless approach will get rid of DevOps?

it's another fad for now. Both Azure and AWS have terrible response times in this area, and if the route isn't already 'warmed up', you're looking at response times within the tens of seconds.

i like it in theory, because it's a very functional way of thinking, but its a fucking meme. only serves to fragment your codebase even more than microservices already do.

I quite like the idea because dealing with server-by-server configuration bullshit uses a lot of my time right now. I wouldn't currently start a new project that uses it, but I would happily work on one if employed to do so.

If you have a usecase why not. If you force yourself to use it then yes.

It will reduce them, you still need someone to hook the production code to whatever you are going to use.

>check out website
>its one of these infinite-scroll bullshit sites
fml, why is a site for sysadmins made like this? I can understand it for a normie site but I don't want a 20 MB website to see the description of a product.

Hello fellow programmers,

I'd like to remind you that I have a book out covering this very same topic. You can buy it on AmazonĀ® or your favorite book retailer. Don't forget to leave a good review!

t. Peter Sbarski

PS: Please don't download illegal copies of my book from places like Genesis Library, because remember: Piracy is Theft!

We definitely are heading in the general direction of making the cloud cluster care about where what runs, and where it stores its files.

BUT I'm not at all sure serverless is a good implementation.

> dealing with server-by-server configuration bullshit uses a lot of my time right now
But we do have ansible, salt, and a lot of cloud computing frameworks to make this not very server-by-server at all.

Are you even using these?

I'm using aws lambda at work. But unlike I'm using Zappa to write it as one monolithic flask app that defers some calls via sns to another instance of itself just using a decorated function. So very easy to get up and running of you have a specific microservice in mind and don't have to worry about handling horizontal scaling and such yourself. The longest delay I've seen is 3 seconds when a new container gets instanced (aws doesn't actually spin up a new lambda for every event).
But of you are writing an app of any decent size or complexity it makes a lot less sense because it doesn't really gain anything and costs more and you have to work around time limits.

Would plow.

devops is shit no one wants to do, so it will probably increase

>even the name is disingenuous
>is it a fad guys?

how many good goi points for renting overpriced hw?

depending on your user profile, you might save tons of money. Consider for example a sports live score portal, where people tune in mostly for the big games, and your traffic drops down to zero when there's nothing on.

> you might save tons of money
This doesn't seem to be one of these things that's designed to be very efficient on your budget.

> Consider for example a sports live score portal, where people tune in mostly for the big games, and your traffic drops down to zero when there's nothing on.
So you either have a CDN in front or a deal with your hosting center to be billed extra for traffic or whatever.

Doesn't seem like this thing will make anything better.

>So you either have a CDN
that's just the thing, isn't it? To avoid all this crap, which takes hundreds of man-hours to configure and maintain.

I don't think serverless is meant for every company, but for small ones where it is cheaper to outsource some things, like all the server maintenance, yeah, it might be a good idea

> takes hundreds of man-hours to configure and maintain
Dozens maybe. The premise is that the service is simple, right? Well, then nearly OOTB solutions (doesn't matter if preconfigured vm, container, some classic account on a preconfigured actual server) from all over the place should work?

Nobody really has demonstrated serverless to be actually easier / quicker as far as I can tell.
It seems to just require more skilled IT (closer to programmers) and more finances etc. to monitor and handle one more bill item.

PLEASE GUYS DONT FALL FOR THIS OBVIOUS FAKE SHIT

ADVERTISING DETECTED

PS: Actually, I get the feeling that you just get more of the most expensive costs, namely programmer costs.

'cause of course, it's not actually serverless, and they indeeed DO make you worry about size / memory / processing time.

It seems like they're "giving you a discount because it's not always running", but then again, what huge issue was there actually in not constantly stopping / starting TINY containers? They would already barely cost cents a year to keep running, right?

>Nobody really has demonstrated serverless to be actually easier / quicker as far as I can tell.
I think you are just biased. It's super quick and easy. Yeah, if you try to shoehorn something that uses tons if time or memory processing into some dinky lambda you are obviously doing something wrong. Using them to consume an event stream and do some triage and trigger further processing depending on the event though is more appropriate.

> Yeah, if you try to shoehorn something that uses tons if time or memory processing into some dinky lambda you are obviously doing something wrong.
I'd be just offloading the one thing that would be attractive to offload - computer work.

Again, what's the point in having some fancy service provider that starts and shuts down containers or whatever they do in their black box all the time to offload processing power and work that would cost only CENTS to keep up all the time without the latency and with possibly less annoying billing structure?

Oh, and with the further benefit that these then aren't written against some pretty specific API that you're not running yourself, ever, but a thing that your own server can just run now or next year or in five year when it's dirt cheap(er).

PS: And no, it probably doesn't save money either.

You'll have to keep up to date with the API / framework in question.
You'll have to keep software up with the COST structure you're programming against.
You'll have a harder time debugging this shit, or doing the dev stuff on your machine.

I do not at all get this thing. It just seems like the classic nickel & dime scheme to make an already super cheap thing -so cheap that often the transaction to pay the bill is more cost- (TINY amounts of processing power and RAM for brief periods of time) more expensive again, and tie you more closely to a small number of providers... rather than just about everyone who can run computers, as is the case with Linux hosted VMs and containers and the like.

It isn't supposed to save money you moron, it's supposed to save effort. And it isn't hard to debug because the only use case is dead simple services that can be covered with unit tests. And the api is also tiny.
Yeah, I can tell you don't get it because you can't see past your previous experience.

>serverless computing

Still requires a server.

> dead simple services
There is another crux. What is a dead simple service and how do you guarantee it will stay that way?

More to the point, aren't we usually building software with increasingly more dead simple services so, you know, software can do more work? But then this doesn't do the one valuable thing: Guarantee low latencies so this piling up of dead simple services remains dead simple and comfy.

And is this ALL you do? If not, why can't you just run dead simple things in VMs or containers that will not really have to keep pace with the framework's API, price structures from the few non-standandardized services that offer serverless, that may also do something stateful or scale a bit more, and so on?

>Yeah, I can tell you don't get it because you can't see past your previous experience.
I feel I shouldn't ignore everything I know, no?

Again, successful software generally grows, and outsourcing the cheapest tiniest events in event-based software using a special API (even if it's simple) to program against a few service provider's black box with PARTICULARLY no good latencies and no availability of "bursts" of huge processing power seems really dumb.

it might imply it's not for sysadmins

There's also this, www.openfaas.com, which is similar to Amazon Lambda or Google Clloud Functions, except it's open source, free as in free beer, and there's no vendor lock-in.

Basically, the serverless move is not about paying someone else to run the servers for you, it's just the next step after microservices.