Prologue: If you want to start playing with Kubernetes, read it through. If you are an expert looking for a low-cost cluster like me, read it through.
Demand of Cheap Kubernetes
Since its emerge, Kubernetes(k8s) has been the game changer in the IT industry. Today it can be viewed as the de facto standard container orchestration platform. The only problem is the high cost and the deep learning curve.
For employers, hiring a k8s expert is not an easy and cheap task. For k8s enthusiasts, practicing on a k8s cluster is not cheap either. Hence there is a gap between the number of available experts and the demand. The solution to this dilemma is a cheap cluster with which amateurs can easily practice and improve their skills.
GKE
Google Cloud provides managed k8s clusters at a cost. The sweet thing is that the cost can be reduced to near zero. Let me break down the costs and explain how to avoid them. All the information comes from the official doc.
Cluster
Every cluster is charged is charged at maximum $74.4 per month as the management fee, except the first cluster in the account. Hence $0 for management.
Nodes
A cluster needs at least one node to function. In some data centers(Oregon, Iowa
and S. Carolina) every account can use one e2-micro(2 cores, 1G RAM) instance
for free. Besides the resources consumed by the basic k8s pods, there are around
400-500 Mb of RAM available for workloads. Hence $0 for a single node. If more
nodes are needed, spot instance is another cheap option if availability is not
the first priority, which is true for most personal use cases.
Traffic
When workloads are deployed, you may wish to expose them to the internet. The cost of egress traffic can be reduced by 2 measures:
- Use Premium network under Free Tier limit if:
- clients are not in China or Australia
- monthly usage below 1 GB
- Use Standard network at a very low cost($0.085 per GB)
Storage
For stateful applications like databases it is vital to keep data persistent across workloads' lifecycle. 30 GB of standard persistent disk is free for every account.
How To Use
All the needed pieces are provided by GKE, the only problem is how to combine them for practical use.
The first step is to create the cluster. You can choose free single-node cluster or cheap multi-node cluster depending on your need. The firewall rules should be set so that HTTP and HTTPS requests are permitted.
The second step is to install an ingress controller in order to provide service from multiple workloads in the cluster. The details can be found in this repo.
Now the cluster is ready for use. Although being too weak to support heavy workloads, it is sufficient for personal practice. My last advice is to setup a budget so that you won't be surprised if Google decides to withdraw its free tier offers.