Dear friend, you have built a Kubernetes

Original Article ↗ Hacker News Discussion ↗

Kubernetes complexity and when it’s justified

  • Many argue Kubernetes is “complexity abstracting over complexity” and unnecessary for small/medium systems that can run on 1–2 servers with shell scripts, systemd, or docker-compose.
  • Others say once you need many services, environments-on-demand, autoscaling, HA/DR, cert management, and standardized deployments, K8s becomes the least-bad option.
  • Several note that K8s is designed for Google-scale problems; 99% of orgs will never reach that scale, yet copy the tooling anyway.

Migration and operational pain

  • Multiple accounts of painful, multi‑month or multi‑year migrations (dozens to thousands of services) with outages caused by misconfigured limits, networking, or storage.
  • Some migrations eventually paid off via better autoscaling and resource utilization; others saw little to no cost savings versus well-tuned VMs/ASGs.
  • A recurring theme: orgs underestimate project design, testing, and observability work needed; “culture that wants optimistic timelines” is blamed.

Managed vs self‑hosted Kubernetes

  • Managed offerings (EKS/GKE/AKS) are seen as reasonable if you’re already in public cloud; self‑hosting K8s as a small org is often called a waste of money and staffing.
  • Where public cloud is off-limits (regulation, internal “cloud”), self‑managed K8s clusters are reported as fragile and hard to debug without strong infra teams.

Alternatives and “middle ground”

  • Alternatives praised: Docker Compose, Docker Swarm, Nomad, k3s/microk8s, ECS/Fargate, Kamal, NixOS scripts, CaaS platforms, and PaaS offerings (Heroku, Fly.io, Render, DO App Platform).
  • Many argue an opinionated, simpler orchestrator that sits between “bash + ssh” and full K8s is missing or underused.
  • Some say a well‑designed setup with ASGs, load balancers, Ansible/Terraform, and boring Unix tools is enough up to thousands of instances.

Shell scripts, “boring tech”, and bad practices

  • Strong support for simple deploys: “scp + script + HAProxy” works at surprising scale if you’re disciplined.
  • Critics counter that large shell-based systems tend to accumulate undocumented tweaks, brittle error handling, and ad‑hoc reimplementations of half of K8s.
  • Others reply that many “wins” attributed to K8s actually come from cleaning up bad practices (12‑factor, CI/CD, clearer configs), which could have been done without K8s.

Careers, hiring, and standardization

  • Several note K8s has become a de facto checkbox in job ads; lack of K8s experience can be an automatic rejection, regardless of broader sysadmin skills.
  • Supporters emphasize K8s as a common API and mental model across orgs and clouds, lowering onboarding cost—critics call this “resume-driven development” and vendor‑motivated standardization.