Using Kubernetes

Tendermint plus Kubernetes

Tendermint plus Kubernetes

This should primarily be used for testing purposes or for tightly-defined chains operated by a single stakeholder (see the security precautions). If your desire is to launch an application with many stakeholders, consider using our set of Ansible scripts.

Quick Start

For either platform, see the requirements

MacOS

curl -LO https://storage.googleapis.com/kubernetes-release/release/$(curl -s https://storage.googleapis.com/kubernetes-release/release/stable.txt)/bin/darwin/amd64/kubectl && chmod +x kubectl && sudo mv kubectl /usr/local/bin/kubectl
curl -Lo minikube https://storage.googleapis.com/minikube/releases/v0.18.0/minikube-darwin-amd64 && chmod +x minikube && sudo mv minikube /usr/local/bin/
minikube start

git clone https://github.com/tendermint/tools.git && cd tools/mintnet-kubernetes/examples/basecoin && make create

Linux

curl -LO https://storage.googleapis.com/kubernetes-release/release/$(curl -s https://storage.googleapis.com/kubernetes-release/release/stable.txt)/bin/linux/amd64/kubectl && chmod +x kubectl && sudo mv kubectl /usr/local/bin/kubectl
curl -Lo minikube https://storage.googleapis.com/minikube/releases/v0.18.0/minikube-linux-amd64 && chmod +x minikube && sudo mv minikube /usr/local/bin/
minikube start

git clone https://github.com/tendermint/tools.git && cd tools/mintnet-kubernetes/examples/basecoin && make create

Verify it worked

Using a shell:

First wait until all the pods are Running:

kubectl get pods -w -o wide -L tm

then query the Tendermint app logs from the first pod:

kubectl logs -c tm -f tm-0

finally, use our Rest API to fetch the status of the second pod’s Tendermint app.

Note we are using kubectl exec because pods are not exposed (and should not be) to the outer network:

kubectl exec -c tm tm-0 -- curl -s http://tm-1.basecoin:46657/status | json_pp

Using the dashboard:

minikube dashboard

Clean up

make destroy

Usage

Setup a Kubernetes cluster

Please refer to the official documentation for overview and comparison of different options.

Kubernetes on Digital Ocean

Available options:

As you can see, there is no single tool for creating a cluster on DO. Therefore, choose the one you know and comfortable working with. If you know and used terraform before, then choose it. If you know Ansible, then pick kargo. If none of these seem familiar to you, go with kubeadm. Rancher is a beautiful UI for deploying and managing containers in production.

Kubernetes on Google Cloud Engine

Review the Official Documentation for Kubernetes on Google Compute Engine.

Create a cluster

The recommended way is to use Google Container Engine. You should be able to create a fully fledged cluster with just a few clicks.

Connect to it

Install gcloud as a part of Google Cloud SDK.

Make sure you have credentials for GCloud by running gcloud auth login.

In order to make API calls against GCE, you must also run gcloud auth application-default login.

Press Connect:

../_images/gce1.png

and execute the first command in your shell. Then start a proxy by executing kubectl` proxy.

../_images/gce2.png

Now you should be able to run kubectl command to create resources, get resource info, logs, etc.

Make sure you have Kubernetes >= 1.5, because you will be using StatefulSets, which is a beta feature in 1.5.

Create a configuration file

Download a template:

curl -Lo app.yaml https://github.com/tendermint/tools/raw/master/mintnet-kubernetes/app.template.yaml

Open app.yaml in your favorite editor and configure your app container (navigate to - name: app). Kubernetes DSL (Domain Specific Language) is very simple, so it should be easy. You will need to set Docker image, command and/or run arguments. Replace variables prefixed with YOUR_APP with corresponding values. Set genesis time to now and preferable chain ID in ConfigMap.

Please note if you are changing replicas number, do not forget to update validators set in ConfigMap. You will be able to scale the cluster up or down later, but new pods (nodes) won’t become validators automatically.

Deploy your application

kubectl create -f ./app.yaml

Observe your cluster

web UI

The easiest way to access Dashboard is to use kubectl. Run the following command in your desktop environment:

kubectl proxy

kubectl will handle authentication with apiserver and make Dashboard available at http://localhost:8001/ui

shell

List all the pods:

kubectl get pods -o wide -L tm

StatefulSet details:

kubectl describe statefulsets tm

First pod details:

kubectl describe pod tm-0

Tendermint app logs from the first pod:

kubectl logs tm-0 -c tm -f

App logs from the first pod:

kubectl logs tm-0 -c app -f

Status of the second pod’s Tendermint app:

kubectl exec -c tm tm-0 -- curl -s http://tm-1.<YOUR_APP_NAME>:46657/status | json_pp

Security

Due to the nature of Kubernetes, where you typically have a single master, the master could be a SPOF (Single Point Of Failure). Therefore, you need to make sure only authorized people can access it. And these people themselves had taken basic measures in order not to get hacked.

These are the best practices:

  • all access to the master is over TLS
  • access to the API Server is X.509 certificate or token based
  • etcd is not exposed directly to the cluster
  • ensure that images are free of vulnerabilities (1)
  • ensure that only authorized images are used in your environment
  • disable direct access to Kubernetes nodes (no SSH)
  • define resource quota

Resources:

Fault tolerance

Having a single master (API server) is a bad thing also because if something happens to it, you risk being left without an access to the application.

To avoid that you can run Kubernetes in multiple zones, each zone running an API server and load balance requests between them. Do not forget to make sure only one instance of scheduler and controller-manager are running at once.

Running in multiple zones is a lightweight version of a broader Cluster Federation feature. Federated deployments could span across multiple regions (not zones). We haven’t tried this feature yet, so any feedback is highly appreciated! Especially, related to additional latency and cost of exchanging data between the regions.

Resources:

Starting process

StatefulSet

StatefulSet

Init containers (tm-gen-validator) are run before all other containers, creating public-private key pair for each pod. Every tm container then asks other pods for their public keys, which are served with nginx (pub-key container). When tm container have all the keys, it forms a genesis file and starts the Tendermint process.