Do you REALLY need Kubernetes?

[@CobyTamayo]

Never thought I'd say this about a _cloud provider_ but I feel like I have a parasocial relationship with Fly's devrel team. Their branding is just...too good.

[@flydotio]

Im so glad you like the videos!!

[@chrislanejones]

Great video, that "FKS" comment was priceless.

[@lulzmachineify]

Kubernetes saves us metric bucketloads of money compared to renting PaaS/SaaS for things like kafka, cassandra, postgres etc. It's amazing for running infrastructure on

[@AshesWake-sf7uw]

Great video! Kubernetes allows to do so much more than just orchestrating servers in a microservice

[@AmirHosseinHonardust]

Hmmm. K8s is ok, until you have to manage it. Vertical scaling is under appreciated. Although the requirement for vertical scaling is a language and ecosystem that can scale well with the hardware. I love elixir for this. Scales incredibly well both ways. Without any other tools needed. Remote function call? No need for grpc or rest! BEAM does it out of the box! And to be honest, the common language only works when the language itself is not more complicated than the sum of the actors trying to utilize it! Which is exactly what you get with self-hosted k8s. Unless its a requirement for work (as it is right now) I stay clear of the beast. There is a 100 more ways to fix most of my pain points without introducing 10 times more with k8s.

[@halftome]

Kubernetes does a lot of stuff that I used to have to write scripts to do, and it does it in a declarative and standardised way. It has its place though, and I certainly wouldn’t use it for everything.

[@cykatroopa]

It absolutely makes sense to run K8S even at small scale due to its nice ecosystem. You really get a lot of functionality for free. If you're not sure, give it a good try. It's not hard to use or install. You get free HA if you can abstract storage volumes. I love K8S.

[@deado7282]

No. But it's super helpful. It provides a Uniform Standard for almost everything. You should definitely give it a try. It's about much more than scaling.

[@gearboxworks]

So, the alternative to K8s is to couple one's application to the Fly service? Please correct me if my takeaway was wrong. I'm not arguing everyone needs K8s, but the only alternative you mentioned was to use Fly's built-in functionality instead, which seems a lot like jumping out of the frying pan (of complexity) and into the fire (of proprietary dependencies.)

[@hslee169]

99% of projects do not need K8S from the start. Some might after achieving some amount of scale.

[@ask_carbon]

One can always start with bare minimum node config resulting in a minimum capx spending on infra. Its modular and its scalable so instead of re-base your infra from monolith or rando docker containers so imo it's much better to start with bootstrap k8s and scale it up/down to your needs.

[@deado7282]

It's about much more than scaling. Consider a simple application deployment (usually at least 2 seperate machines): Frontend, Api-Server, Reverse-Proxy with ACME, SQL-Database, Failover, Backup-Catalog, some Blob for files, Grafana for basic Monitoring, simple Pipeline for test & build convenience. If you want to operate that on your own, you gain a lot already by having a unified API & a kubernetes operator for the database. Of course you can buy a bunch of managed services but: - you get problems in B2B because you can't explain to them, where their data is. - low flexibility. These days devs tend to build around managed services instead of just running the thing thats fits best on their machines. - vendor lock-in is real - no single mandate apps (again b2b) - cost Even if you only have very few services & you want basic telemetry you already gain a lot.

[@medel4174]

100% don’t need it even at scale. do you think facebook or google use that shit for scale

[@AshesWake-sf7uw]

@@hslee169 it's atleast cheaper than what AWS is 💀

[@seansingh4421]

@@AshesWake-sf7uwBro if tomorrow we get invaded by the Aliens, even they would keep their battle plans on a AWS S8 bucket

[@vicaya]

If you're in a position to ask the question these days, the answer is probably yes, because the alternatives are probably worse. It's fairly easy to use k8s, esp. k3s for pretty much anything even on a single node. The complexity of k8s itself is vastly exaggerated, while the complexity of devops in general is often underestimated.

[@vyrwu]

These days, setting up K8s, trying it out, rolling it out at a small scale and learning about it became so easy that I’d say more people than ever should startup on K8s. Don’t even need to manage the control plane with managed service offerings. But if not, then I’d disagree that buying into a managed service provider is the right answer. IMO a much higher ROI is to learn basic Linux administration and setting up a few cloud boxes yourself, until you’ve scaled up so you can hire experts to help (or better yet, you become one).

[@OldKing11100]

I feel like horizontal scaling only becomes an issue early on with single threaded dynamic languages like Python/Node apps. With something simple like Go, html/template, and SQLite(WAL) it's amazing how far a monolith can be pushed on a VPS. Down the road under heavier loads though K8s can be nice if I'm not managing it so something like FKS could be a good option.

[@autohmae]

The way I see it, k8s when you start adding all the pieces you want/need is not small and has many moving parts. And I think as security is improving for k8s (and the underlying Linux kernel systems) we'll see more providers offering it like a multi-tenant platform, my guess is in the form of something like vcluster and a smart operator. Because we all want the gitops and standardization, but not deal with everything else if we don't have to and pay less.

[@fredio54]

6-10 or more independent services/apps and I'd say yes. Less and you can get away without all the sophistication without undue overhead. The reason to use it is a consistent approach to everything you do. There's almost nothing it can't handle. But if you are likely to grow above half a dozen in the near term, starting on kube when you don't need it and maturing your approach on those first few apps will save you getting bogged down later when you wind up running too much in ECS or EC2 or whatever your offerings do, etc. Plus no costly migration from X to kube. Also worth noting that kube on bare metal without readily available additional compute to scale into is less beneficial since your only options are full time head room or shuffling workload replica size between this and that to optimise efficiency.

[@ha5anmukhlis]

k8s would be cool if it weren't so expensive once you want to host it somewhere.

[@pixelfairy]

Im glad you give us FKS. But you might want to run that one by marketing. How about FlyK? You can pronounce it flick.

[@fernandobalieiro]

K8S is really great, a single api that you can use everywhere instead of leaning of learning n tools, I'd even argue you don't even need to use it at scale of how good it is, principally with managed services

[@mayikx]

I tried this platform, and this platform is unnecessary complex. However, their videos are awesome

[@liamnguyen9527]

Thanks for the good video, non-bias information.

[@adrianosleipnir854]

Fly is great. But yes sometimes we need Kubernetes

[@JAMaitim]

FKS demo please

[@pansartax]

Great video!

[@MrAlanCristhian]

Startups just use what is popular and use as many buzzwords to attract unsuspicious investors.

[@eswarnichtsmehrfrei]

yes

[@digitalsparky]

Oh no, we fks'd up...

[@ramonpereira4460]

probably not

[@deiminator2]

No, unless your company has more money than common sense

[@malvoliosf]

Please don’t say “begs the question” when you mean “raises the question”.

[@flydotio]

I'd actually never thought of that before, thanks!

[@lulzmachineify]

Why? Begs the question is valid english

[@malvoliosf]

@@lulzmachineify It’s perfectly valid English, but it does not mean “raises the question”. It means assuming the conclusion to are trying to prove. For example, if I wrote “You should not use the phrase ‘begging the question’ because it’s not wise to say it in conversation”, I am trying to use the conclusion I want to prove. That would be begging the question.

[@simonmartinelli]

Great video. But better beware of Betteridge's law 😉 en.wikipedia.org/wiki/Betteridge%27s_law_of_headlines