SaaS Builders! What do you find challenging about cloud operations?

Question

Running your apps on the cloud hold many risks, especially on the costs side. An unmanaged cloud environment can be very expensive, but in order to manage it you need, for example Kubernetes specialists who will likely be one of your highest paid teammates.

What is it about cloud that you wish were simpler? What's a deal breaker when you make the decision whether to use cloud or not?

Ivan Ralic · Accepted Answer

I actually don't understand why people use AWS or Kubernetes. They both suck. First one is intentionally bad with the UX to make you lose money. Second one is too complex and missing key functionalities by default (networking, secure-by-desing principle) I would really suggest for anyone thinking about k8s to look at this video https://youtu.be/YGujnkAV3pc It really sums up my experience with k8s. I have 6-years old Enterprise SaaS (15+ services) running on Docker Swarm without any DevOps engineers for the past 2-3 years. Everything automated and without downtime for the past 2 years.

Ivan Ralic · Answer

@geri_mate when it comes to orchestration tool as I've already mentioned above and as video suggests Docker Swarm is a lot better 😄 And when it comes to cloud providers I often use Azure, but we've switched out whole infrastructure from AWS to DigitalOcean to Azure with Docker Swarm with no downtime and no big problems whatsoever. AWS just sucks because they have intentionally bad UX to make you lose money, just like "patterned" Amazon checkout 😄

Jesse Ezell · Answer

I don't understand why anyone thinks Kubernetes is hard. We've been using it for years without issues and never had any dedicated "specialists" for it. In any case it's only getting easier and easier with things like GKE autopilot, and if that's still too much, you can just use Cloud Run and get Kubernetes under the covers with all the things that scare people abstracted away.

Ivan Ralic · Answer

@john_radosta the video I've sent you is from the Citizens Bank, one of the oldest and biggest Banks in the US. Enterprise Financial Institution with 18k employees and billions in profit 😄

Have a good day.

Ivan Ralic · Answer

@john_radosta with all due respect most of your claims are incorrect 😄
Docker Swarm does have great health check and it does  automatically restart fault containers.
There is auto-scaling and it supports load balancers from other providers (we've used AWS before)
Monitoring is not harder from k8s when you need them, Graphana works it's magic, and if you want out-of-the-box solution you can look at swarmpit.
When it comes to networking Docker Swarm is simply superior to k8s because it's secure-by-design. You can have multiple environments running on the same logical infrastructure not worrying if BETA infrastructure might affect Production.
It's really not for bootstrapping I have an Enterprise-grade SaaS used by the Telco Infrastructure Providers to deploy FTTX and 5G networks in US and EU. It's B2B and B2G solution and it's running (as already mentioned) without any DevOps engineers. CI/CD, auto scaling and auto deployment work perfect without any interactions.
I really suggest you watch the video I've posted. I'm not related to them at all, it just explains my experience with Docker Swarm, k8s and back 😊

Geri Máté · Answer

@ralic Readme on GitHub is in desperate need of a touch up so instead of the value proposition included in it we usually explain that dyrector.io is an open-source IDP providing microservices and bundled configuration management for small to middle-sized teams without specialists to accelerate product development. The repo is available at link the below: https://github.com/dyrector-io/d... If you got time to read the docs it's available here: https://docs.dyrector.io Any feedback is helpful :)

John Radosta, FINRA 65 · Answer

@jesse_ezell agreed. Nodes have Services that point to Deployments comprised of Pods. Ingress provides access to the Services running on the Nodes. Kubernetes in two sentences.

Ivan Ralic · Answer

@geri_mate sure, I'm always glad to help a maker 😄

Geri Máté · Answer

@ralic What do you suggest to use instead of AWS and Kubernetes?

Geri Máté · Answer

@ralic My bad about k8s and Docker Swarm :) Thanks for your comment!

My teammates are making a developer platform allowing non-specialists to manage microservices architectures and their configs in any environment. Wanna hear more about it?

Thanassis Parathyras · Answer

@geri_mate running software apps on the cloud calls for automation, so that you can be fast and reuse past setup configurations. At the same time, building the automation requires technical know-how and maintenance.
That said, I find that the main cost is to find the right people and pay for their skills and time to help you manage software apps on the cloud with confidence. Essentially, I agree with your point of view. I am myself a maker of a SaaS platform that helps development teams to facilitate the whole DevOps automation cycle without having to spend time to learn and maintain the required management tools.
Glad that I found out today there are other makers with similar perspectives.
I'd be also glad to hear any feedback you might have about cycleops.io, which is the cloud management tool we wish to put in the market.

Geri Máté · Answer

@jesse_ezell Developers, especially freelancers or people working at outsourcing companies I talked to recently find k8s hard to use, or at least something they'd rather not deal with instead of working on their project.

John Radosta, FINRA 65 · Answer

@ralic well for one, Docker Swarm doesn't automatically restart faulty containers, so you have to constantly be watching your cluster or set up notifications for container errors.

It doesn't autoscale automatically as Kubernetes does with the Horizontal Pod Scaler.

If you want a managed service offering to continuously upgrade your cluster and apply patches, they are non-existent in the major cloud providers, every major cloud provider offers Kubernetes as a managed service, and for good reason in my opinion.

There's no built in cluster monitoring or trace for Docker Swarm. Third party only.

Docker swarm does not interface with cloud load balancers, so getting your ingress to play nicely with your VPC you need to set up your own reverse proxy which is a bottleneck, you have a lot less control on your ingress in that regard.

Also docker swarm offers node port only for ingress in mesh mode, which means private IP for cluster node ports is next to impossible as the nodes needs to be publicly discoverable. This is a security risk even for Kubernetes, type NodePort for ingress should never be used for production.

In my opinion, Docker Swarm (and Docker Compose) is for bootstrapping. Its something that you can use to build and test ideas on a small scale, but as your applications and microservices need to scale and grow, you're going to have to rebuild it on Kubernetes to get all the extra features and scale out your DevOps team, so to me it defeats the purpose of using Docker Swarm at all.

John Radosta, FINRA 65 · Answer

@ralic right, because when I think about digital and financial leaders, I think of Citizen's Bank. So what's next, you going to adopt AliBaba Cloud because AliBaba uses it? I'm loving these contrarian recommendations of yours. Makes my work a lot easier.

John Radosta, FINRA 65 · Answer

@ralic mmm, I'm worth a lot more than you think. And your junior developer opinions don't offer much insight. Junior, that's your new nickname.

SaaS Builders! What do you find challenging about cloud operations?

Replies