Kubernetes on Indian infrastructure: when you need it, when you don't

Kubernetes is the most popular container orchestrator in the world. It is also the most over-applied. Most workloads do not need it. Here is a practical decision tree for engineering teams evaluating K8s in 2026.

When you need K8s

• You run 5+ stateless services that need to scale independently based on traffic (e.g. API tier + worker tier + cron tier).
• You have a multi-team org and need namespace isolation
  • per-team resource quotas.
• You do blue-green + canary deploys and need the declarative deployment object (Deployment, Service, Ingress) to manage them.
• You need to autoscale on a metric (CPU, memory, queue depth) and the cloud-provider autoscaler doesn’t fit.

When you don’t

• You run 1-2 long-running services. Just run them on container hosting and skip the orchestration layer.
• You have a single team. A docker-compose file + a reverse proxy is enough.
• You do rolling deploys. A docker run on the next instance is enough.
• Your traffic is steady (not bursty). Static capacity beats elastic capacity when the curve is flat.

What it costs on Indian infrastructure

K8s is free; the cluster it runs on is not. Pricing depends entirely on the SKUs you size from /pricing. Indicative ranges for a typical K8s cluster on the Linux Dedicated L-tier (all bandwidth unmetered):

• Small dev cluster (1 control plane + 2 workers, 4 vCPU + 8 GB each): roughly Rs 7,000-15,000 / month depending on whether you consolidate onto one L2 instance (cheaper) or spread across dedicated workers (more isolation). Use the /pricing la-carte slider to get a live quote.
• Production cluster (3 control planes + 4 workers, 8 vCPU + 32 GB each): in the L3-L4 range on the /pricing matrix; exact price scales with the SKU mix you choose.
• High-availability cluster (3 control planes + 6 workers, 16 vCPU + 64 GB each): at the top of the Linux Dedicated pool, plus an external load balancer.

We don’t ship “Kubernetes” as a fixed-cost SKU because every workload is different. Get a real quote on WhatsApp and we’ll size the right tier.

When you should use a managed K8s service instead

If you don’t have a platform team that can run a K8s control plane 24x7, don’t run your own K8s. Use a managed service (EKS, GKE, AKS) or our k3s-managed Apps tier (already live; see /hosting/container). The cost of the control plane engineer is greater than the cost of the cluster.

If your workload fits in 1-3 containers, run it on the Apps tier and skip the orchestration layer. The K8s tax (learning curve, control plane, observability) is not worth it for a single-digit-container workload.