@MichalBryxi I mean, k8s is useful. The main problem is how it continually trickles down, and gets blindly suggested as a “solution” for much smaller problem sets than it was designed for.
When the scale doesn’t justify the management overhead, that’s when it’s counterproductive.
@MichalBryxi yeah, I can see where that would be overkill for that use case.
K8s does a really strong job of solving problems related "distributed production workloads at scale." So, the compute, networking, storage, scheduling, RBAC, configuration, security, permissions, health monitoring, lifecycle of containers, etc. That's what k8s was made for, and it does an admirable job at it.
The trouble happens when you then have a relatively small problem, like "I need to deploy a static website for a limited customer base", and then just throw k8s at it because you can.
That's like going through pilot training, getting all of your ratings, and buying an Airbus A320 with your own money....just so you can visit your grandma who lives a 2 hour drive away from you on the weekend.
Is A320 obscene overkill for visiting your grandma? Definitely.
Does that mean it's a bad airplane? Not at all. It's a great airplane when it's doing the things that it's designed to do.
@MichalBryxi @Shephallmassive
Eh, I dunno. If you're tired of learning new things, this isn't the field to be in.
New != "always better" && New != "always bad"
Also, not every tool is meant for me or the team I'm working with. I just may not be in that problem space.
An example for myself is Kerberos. (just pulling something out of the air.) I've never had the need to go learn Kerberos, nor install a Kerberos server anywhere. I feel ambivalent about the tool itself, because I don't know much about it. But I take ownership of the fact that I don't care about it *because* I don't know much about it, and my daily work isn't in the inner workings of authentication flows.
So, it's not that Kerberos isn't a useful tool. It just means I'm not familiar with the problems it's meant to solve.
@srfaudio Yup. The words **for me** should have been in bold there. I can see the use & usefulness of k8s in cases.
The case I outlined above is about me not having an option, because "everything is k8s" is status quo in many places now. Even for my 5MB bundle of JS+CSS+HTML.
I would say absolutely nothing if k8s _itself_ would provide one-liner config "this folder, expose on this URL" (for this case). But it can't because k8s a tool, not a solution.
@srfaudio And as I said many times before: I don't care about tools, I care about solutions.
Yes, IT is not the space that makes me happy. But sadly the space where I have the biggest expertise so far.
I do have some follow-up opinions about "need to keep learning" & IT. But let's keep the Sunday nice :)
@MichalBryxi right. Totally get that.
My only comment back, which I'm sure you already know, is that even "this set of files, at this URL" takes a lot of actual, other things to make it actually happen on a live website.
There's no world where you can just do that, k8s or not.
Any tool that you've used that allows you to say "this set of files, at that URL" and you're done is doing a lot of behind the scenes work for you. Which means, someone is 1) creating/managing it, or 2) paying for it.
@Shephallmassive @MichalBryxi oh yeah...that's a thing.
The other problem is that many times, the line between "we don't need k8s" and "oh shit...k8s would make this a *lot* easier" is only visible after you've already crossed it.
Like, early on, your Debian server running NGINX, and the certs you go out and create, and the DNS records you get, and redeploying your static pages is so much easier...*genuinely*...than having to worry about all of the moving pieces of k8s.
Then, many times the following loop happens over a few months/years:
A new requirement came down for a new site. Sure, no problem...ok, let's see. Well, I guess let's just add a second server, and we'll get a second DNS entry. And just add that to NGINX. Oh wait...the passwords and secrets. Hmm. No problem, let's just make a little tracker tool for the credentials. Ok cool. Done! See? No k8s needed.
<iterate this loop a few times until there's now 10+ of those>
And then, at that point now managing your "simple" solution is actually a huge nightmare of duct tape and esoteric knowledge by team members that come and go.
So, yeah...sometimes people want to slap k8s onto things when it's completely not required.
Other times though, they've got the background experience to know that it should be built out correctly, early, with a plan for future growth so that you never hit that "Whoops...this would be way easier with k8s" point.
As with everything in software engineering: it depends.