Using Kubernetes with Ballerina

An API written in Ballerina and deployed to Kubernetes that returns website rankings

Contents

• Check your environment
• Install Minikube
• Create a Kubernetes cluster
• Setup a Docker Client
• Install, run and test Kubectl
• Run Ballerina and deploy to minikube

Check your environment

The setup described here used a standard laptop with 8GB of RAM running

Microsoft Windows [Version 10.0.18362.357]

Hyper-V should be enabled and we need to add a Virtual Switch using the Hyper-V Manager. My switch was named minikube. There was an issue with IPV6 addressing (race condition or something like that) I disabled ipv6 and started over. When using CLI tools I prefer bash and have Windows Subsystem for Linux (WSL) setup wit 18.04 LTS (Bionic)

Install minikube

Initially minikube was installed on WSL but then I realised that this won't work

# systemctl status docker

Because WSL does not have systemd. It tells you it can't operate. So following this tutorial I installed minikube on Windows. And then I ran straight into:

>minikube start --vm-driver=none
* minikube v1.5.0 on Microsoft Windows 10 Enterprise 10.0.18362 Build 18362
X The driver 'none' is not supported on windows

Ah well. Here is a summary of what I found out about where to run minikube with what driver.

	WSL Ubuntu	Windows	Ubuntu
VM	no	yes	yes
Native	no	no	yes

But then anyway it turned out that minikube on Windows was necessary aaaanyway. Because when Docker Desktop and minikube are running simultaneously Windows ran out of RAM. To side-step this I used the docker daemon that runs inside minikube and stopped Docker Desktop.

In this design, I also like the separation between WSL and Windows. It means that when I come to actually deploy to a cloud, I only need to reconfigure the clients on WSL. Windows is acting like a fake cloud :)

Create a Kubernetes cluster

Run the windows cmd as Administrator and run the following.

>minikube start --vm-driver hyperv --hyperv-virtual-switch minikube

Check what just happened by asking for the IP address.

>minikube ip
192.168.1.67

Setup a Docker Client

Although we're not going to run a docker daemon on WSL we need a client so Ballerina can prepare the image.

# sudo apt install docker.io

$ docker --version
Docker version 18.09.7, build 2d0083d

Now back in Windows I grab the values for the environment vars.

>minikube docker-env

The vars are used to create the following entries in .bashrc on WSL.

export DOCKER_HOST="tcp://192.168.1.67:2376"
export DOCKER_TLS_VERIFY=0
export DOCKER_CERT_PATH="/c/Users/gavin/.minikube/certs"

A couple of changes that were made. The DOCKER_TLS_VERIFY is turned off as we are working on a private network. Also note that the path has no /mnt/. This is because in /etc/wsl.conf I have set the root as /.

To check if the client is talking nicely to the daemon try this from WSL.

$ docker info

This will return a lot of output, check for the line Name: minikube. See set up docker for wsl for more.

Install, run and test Kubectl

Kubectl will communicate with the Kubernetes cluster that was created using minikube. Kubectl version v1.16.2 was set up in WSL.

curl -LO "https://storage.googleapis.com/kubernetes-release/release/$ver/bin/linux/amd64/kubectl"
mv kubectl /usr/local/bin

Now to confgure the client I did the same as docker and used the windows config.

>kubectl config view

These values were used in this bash script

#!/bin/bash -x
minikubeip='192.168.1.67:8443'
winuser='gavin'

kubectl config set-cluster minikube \
  --server="https://$minikubeip" \
  --certificate-authority="/c/Users/$winuser/.minikube/ca.crt" \

kubectl config set-credentials minikube \
  --client-certificate="/c/Users/$winuser/.minikube/client.crt" \
  --client-key="/c/Users/$winuser/.minikube/client.key" \

kubectl config set-context minikube --cluster=minikube --user=minikube

This is the output of running the shell script.

+ minikubeip=192.168.1.67:8443
+ winuser=gavin
+ kubectl config set-cluster minikube --server=https://192.168.1.67:8443 --certificate-authority=/c/Users/gavin/.minikube/ca.crt
Cluster "minikube" set.
+ kubectl config set-credentials minikube --client-certificate=/c/Users/gavin/.minikube/cert.crt --client-key=/c/Users/gavin/.minikube/client.key
User "minikube" set.
+ kubectl config set-context minikube --cluster=minikube --user=minikube
Context "minikube" created.

Now that we have switch our kubectl to use the minikube context let's test it.

$ kubectl cluster-info
Kubernetes master is running at https://192.168.1.67:8443

Almost there! I learnt most of this from minkube on win with wsl

Setup Kubernetes using Ballerina

Using ballerina I defined a Rankings API to deploy to my new cluster. This is a simple test service based on data sourced from Alexa.

The following shows how the artifacts built by ballerina were applied to minikube and tested. First I tested the container that Ballerina built.

$ docker run -d -p 9090:9090 snow6oy/fnarg:rankingsApiBal

Now the container is running, check the service

$ curl http://localhost:9090/rankings/status
ok

Looks good, so let's publish to the cluster.

$ kubectl apply -f /home/gavin/fnarg/target/kubernetes/rankingsApiBal
service/rankings-svc created
ingress.extensions/rankings-ingress created
secret/rankings-secure-socket created
deployment.apps/rankingsapibal-deployment created

And to see what just happened

$ kubectl get svc
NAME           TYPE        CLUSTER-IP      EXTERNAL-IP   PORT(S)          AGE
kubernetes     ClusterIP   10.96.0.1       <none>        443/TCP          81m
rankings-svc   NodePort    10.104.84.120   <none>        9090:30191/TCP   44s

Using the minikube IP and the NodePort I can run an HTTP query.

$ curl https://192.168.1.67:30191/rankings/?query=goo -k
[{
	"rank": "1",
	"domain": "google.com"
}, {
	"rank": "7",
	"domain": "google.co.in"
}]

Hooray!