Hi everyone! I'm currently working at a large company that still uses traditional "Failover Clusters in Windows," but we're looking to transition to Kubernetes. Our team is excited about this change but we don't have much experience with it yet. I have a few questions specifically for BareMetal or On-Premise VMs:
1. How have you implemented GitOps for infrastructure management? For example, how do you handle things like defining the metrics server?
2. Is TalosOS the right choice for on-premise setups?
3. When it comes to local storage for SQL Server, should we go with SMB, NFS, or are there better alternatives?
4. Lastly, what strategies do you use for backups and ensuring quick recovery in case of a disaster? How do you feel secure about that?
Thanks in advance for your help! 😉
2 Answers
When it comes to GitOps and Kubernetes, you've got two main options: ArgoCD and FluxCD. They each have their advantages and disadvantages, and they operate using a "reconciliation loop" that checks your Git repos against what's currently running in your cluster.
For your infrastructure questions, defining the metrics server falls under deployment tooling, which you should categorize into several sections: cluster bootstrapping, platform tooling, app bootstrapping, and application deployment itself. Both ArgoCD and FluxCD can help you manage this. They work well on a Talos cluster or any other bare-metal setup. The idea is to make your GitOps setup transferable to avoid vendor lock-in.
Regarding storage, if you're using bare-metal, it's best to utilize fast, local storage, or you might want to invest in a robust network fabric to run Ceph for shared storage. However, the complexity involved can be significant.
Finally, backups are crucial for stateful applications. If you mainly have stateless apps, that concern diminishes. Automation can help, but it's not an easy task, and it's so dependent on your specific needs. I'd suggest getting some support to make your transition smoother. Good luck! Let us know how it goes!
You should definitely check out Rancher! They offer a free tier, and when deployed with Harvester, it can manage your OS and quickly spin up clusters. Harvester also supports other VM platforms. With Rancher, you get built-in cluster storage through Longhorn and GitOps tools via Fleet, which could cover all your needs.
I've heard about Longhorn as a distributed filesystem. Is it tough to set up? Also, what is Fleet? I saw it mentioned on Azure but I'm a bit confused about its functionality.
I agree that bare metal won't be an issue for vCPUs. We need to avoid internet dependence too, so options like Azure Disk won't work for SQL Server. Do you lean more towards ArgoCD or FluxCD for GitOps?