Building and Operating a Site with the GKE Tutorial

Overview

I previously built the Toco-chan Bus app on a Sakura VPS.

Flask Python3 で 戸田市 tocoちゃんバスあと何分？ Webアプリ作成♪ - 長生村本郷Engineers'Blog

以下に移行しました。 kenzo0107.github.io

The Sakura VPS hosted several personal projects, but since some of them were
discontinued, I moved the Toco-chan Bus app over to GCP.

I’ve put together a fresh write-up of that experience, doubling as a GKE tutorial.

Why GCP?

The Toco-chan Bus app is a lightweight site that doesn’t even have a database.
Because of that, I figured I could use GCP’s free tier, which led to the migration.

Architecture

On GCP, I’m using a Container Cluster, with an architecture like the following.

Below, I’ve summarized the build steps following GCP’s tutorial.

gcloud Default Configuration

Please refer to the article from when I previously built this using the free tier.
It’s a lonely single-Pod setup, but bear with me.

無料枠で運用！ GKE + Kubernetes で Hubot 〜CLIから実行編〜 - 長生村本郷Engineers'Blog

以下に移行しました。 kenzo0107.github.io

Creating the Container Cluster

As with the previous article, in order to use the free tier
I first create the cluster with 3 nodes and then reduce it to 1 node once it’s done.

I specified cluster version 1.7.2. ((As of August 2, 2017, the latest cluster version was 1.7.2))

$ gcloud container get-server-config

Fetching server config for us-west1-b
defaultClusterVersion: 1.6.7
defaultImageType: COS
validImageTypes:

• CONTAINER_VM
• COS
• UBUNTU
  validMasterVersions:
• 1.7.2
• 1.6.7
  validNodeVersions:
• 1.7.2
• 1.7.1
• 1.7.0
• 1.6.7
• 1.6.6
• 1.6.4
• 1.5.7
• 1.4.9

• Create the container cluster

$ gcloud container clusters create tocochan-cluster-free \
      --cluster-version=1.7.2 \
      --machine-type f1-micro \
      --disk-size=30 \
      --num-nodes=3

• Set the node count to 1

$ gcloud container clusters resize tocochan-cluster-free --size=1

• Verify

$ gcloud container clusters describe tocochan-cluster-free | grep currentNodeCount

currentNodeCount: 1

We can confirm that the current node count is 1.
And with that, it’s on the free tier!

After creating the cluster, I obtain the container cluster’s credentials so that
I can connect to and operate the cluster with kubectl.

$ gcloud container clusters get-credentials tocochan-cluster-free

Registering with the Container Registry

I create an image from a container started locally and
register it with the Container Registry, GCP’s private Docker repository.

I use the following repository.

https://github.com/kenzo0107/toda-tocochan-bus

• Start the Docker container

$ git clone https://github.com/kenzo0107/toda-tocochan-bus
$ cd toda-tocochan-bus
$ docker-compose up --build -d

• Create an image from the running Docker container and push it to GCR ((The project ID is “mametsubuservice-175801”))

$ container_id=$(docker ps | grep [f]lask | awk '{print $1}')
$ docker commit $container_id gcr.io/mametsubuservice-175801/tocochan:latest
$ gcloud docker -- push gcr.io/mametsubuservice-175801/tocochan:latest

Deploying a Single Pod, as Tutorial ①

Deploying a single Pod on its own is rarely done in practice.
That’s simply because there’s no way to manage cases such as a container inside the Pod stopping abnormally.
This time, it’s a tutorial exercise for the sake of free-tier operation and understanding the concepts.

• pod.yaml

apiVersion: v1
kind: Pod
metadata:
  name: tocochan
spec:
  containers:
    - image: gcr.io/mametsubuservice-175801/tocochan:latest
      imagePullPolicy: Always
      name: tocochan

Running the Single-Pod Deployment

$ kubectl create -f pod.yaml

Checking the Pod Status

$ kubectl get pods

• Access configuration

flask starts on port 5000.

$ kubectl port-forward tocochan 5000

Forwarding from 127.0.0.1:5000 -> 5000
Forwarding from [::1]:5000 -> 5000

• Access http://localhost:5000 from a browser

You can confirm that the top page is retrieved.

Deleting by Specifying the Pod Name

Now that we’ve confirmed the single-Pod deployment, let’s delete it.

$ kubectl delete pods tocochan

ReplicaSet Deployment, as Tutorial ②

When you create a single Pod on its own, nothing in particular is recovered even if the Pod stops abnormally.
A ReplicaSet always manages the number of containers that should be running normally and,
if there’s an abnormal stop, it adds a new Pod.

This too is written as a tutorial. I’ll delete it once I’m done.

• replicaset.yaml

apiVersion: extensions/v1beta1
kind: ReplicaSet
metadata:
  name: tocochan
spec:
  replicas: 1   # number of containers to keep running at all times
  template:
    metadata:
      labels:
        name: tocochan
    spec:
      containers:
        - image: gcr.io/mametsubuservice-175801/tocochan:latest
          imagePullPolicy: Always
          name: tocochan

Running the ReplicaSet Deployment

$ kubectl create -f replicaset.yaml

Checking the ReplicaSet

$ kubectl get rs -l name=tocochan

NAME DESIRED CURRENT READY AGE
tocochan-4006188167 1 1 1 10m

What Happens If We Try to Delete the Pods?

$ kubectl delete pods -l name=tocochan

You can see that the number of running containers never drops to 0; instead, new ones are created.

NAME READY STATUS RESTARTS AGE
tocochan-14s3b 1/1 Running 0 4s
tocochan-tsvfn 1/1 Terminating 0 5m

Deleting the ReplicaSet

$ kubectl delete rs tocochan

Deployment Deploy

A ReplicaSet deployment leaves no history on k8s.
A Deployment deploy keeps a history, and if there’s a problem with the current version
you can easily roll the version back.

The Deployment deploy is what solves the problems we’ve seen so far.

• deployment.yaml

apiVersion: apps/v1beta1
kind: Deployment
metadata:
  name: tocochan
spec:
  replicas: 1
  template:
    metadata:
      labels:
        name: tocochan
  spec:
    containers:
      - image: gcr.io/mametsubuservice-175801/tocochan:latest
        imagePullPolicy: Always
        name: tocochan

Running the Deployment Deploy

Adding --record lets you keep a history of operations.
By keeping a history, you can use kubectl operations to roll back to a past version if there’s a problem.

$ kubectl create -f deployment.yaml --record

Checking the Deployment

$ kubectl get deployments -l name=tocochan

NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE
tocochan 1 1 1 1 10m

Checking the ReplicaSet

$ kubectl get rs -l name=tocochan

NAME DESIRED CURRENT READY AGE
tocochan-2006588533 1 1 1 10m

Checking the Pod

$ kubectl get pods -l name=tocochan

NAME READY STATUS RESTARTS AGE
tocochan-4006188167-3zrn9 1/1 Running 0 10m

Checking the Deploy Result

$ kubectl rollout status deployment/tocochan

deployment “tocochan” successfully rolled out

We can see that the rollout was published correctly.

Checking the History

$ kubectl rollout history deployment tocochan

deployments “tocochan”
REVISION CHANGE-CAUSE
1 kubectl create –filename=deployment.yaml –record=true

Editing

$ kubectl edit deployment tocochan

vim launches and you can edit the deployment.

- image: gcr.io/mametsubuservice-175801/tocochan:latest
+ image: gcr.io/mametsubuservice-175801/tocochan:v0.0.1

When you edit as above, save, and exit,

NAME READY STATUS RESTARTS AGE
tocochan-1297744065-2qb87 1/1 Terminating 0 15m
tocochan-4006188167-3zrn9 1/1 Running 0 10s

you can see that the existing container is being stopped and a new container has been started.

• Check the history

$ kubectl rollout history deployment tocochan

REVISION CHANGE-CAUSE
1 kubectl create –filename=all.yaml –record=true
2 kubectl edit deployment tocochan

Checking the rollout history shows that the edit has been added.

Rolling Back the Version

Roll back to REVISION 1.

$ kubectl rollout undo deployment tocochan --to-revision=1

$ kubectl get pods -l name=tocochan

NAME READY STATUS RESTARTS AGE
tocochan-1297744065-2qb87 1/1 Terminating 0 6m
tocochan-4006188167-zswcj 1/1 Running 0 7s

Just like before, you can see the existing container stopping and a new container starting up.

Can We Connect from Outside?

With the Pod in the state we’ve reached so far, run the following command.

$ kubectl get svc

NAME CLUSTER-IP EXTERNAL-IP PORT(S) AGE
kubernetes 10.3.240.1 443/TCP 3h

Other than kubernetes having a cluster-ip assigned, no IP is assigned in particular,
so it’s in a state where it can’t be accessed from outside.

We need to configure it so that it can be accessed from outside.

Creating a Service

We set up an external-facing IP and route so that the Pod can be accessed from outside.

• service.yaml

apiVersion: v1
kind: Service
metadata:
  name: tocochan
spec:
  type: LoadBalancer
  selector:
    name: tocochan
  ports:
    - port: 5000

Creating the Service

$ kubectl create -f service.yaml

Checking the Service

$ kubectl get svc

After a few minutes, it changes from to an IP.

NAME CLUSTER-IP EXTERNAL-IP PORT(S) AGE
kubernetes 10.3.240.1 443/TCP 10m
tocochan 10.3.240.70 xx.xxx.xxx.xxx 5000:32429/TCP 10m

You can confirm that the web page is accessible with the following command.

$ curl -v http://$EXTERNAL-IP:5000

Creating a Load Balancer

It’s possible to set up a load balancer.
It receives on port 80 and forwards to port 5000 on the backend.

• ingress.yaml

apiVersion: extensions/v1beta1
kind: Ingress
metadata:
  name: hello-world
spec:
  rules:
  - http:
      paths:
       - path: /*
         backend:
           serviceName: hello-world
           servicePort: 5000

Creating the Ingress

$ kubectl create -f ingress.yaml

Checking the Ingress

$ kubectl get ingress tocochan

NAME HOSTS ADDRESS PORTS AGE
tocochan * yy.yyy.yy.yy 80 10m

You can see that accessing it as below yields the same result as the curl -v http://$EXTERNAL-IP:5000 we ran earlier.

$ curl http://$INGRESS_IP/

Consolidating the Config Files

$ echo '---' > hyphen.txt; \
  cat \
  deployment.yaml \
  hyphen.txt \
  service.yaml \
  hyphen.txt \
  ingress.yaml \
> all.yaml; \
rm hyphen.txt

• all.yaml

apiVersion: apps/v1beta1
kind: Deployment
metadata:
  name: tocochan
spec:
  replicas: 1
  template:
    metadata:
      labels:
        name: tocochan
    spec:
      containers:
        - image: gcr.io/mametsubuservice-175801/tocochan:v0.0.1
          imagePullPolicy: Always
          name: tocochan
---
apiVersion: v1
kind: Service
metadata:
  name: tocochan
spec:
  type: LoadBalancer
  selector:
    name: tocochan
  ports:
    - port: 5000
---
apiVersion: extensions/v1beta1
kind: Ingress
metadata:
  name: tocochan
spec:
  rules:
  - http:
      paths:
       - path: /*
         backend:
           serviceName: tocochan
           servicePort: 5000

From here on, the following command is all you need!

$ kubectl create -f all.yaml --record

Obtaining a Domain

toda-tocochan-bus.tk is a free domain obtained at freenom,
with the Ingress IP set and published.

Overall Assessment

The flow of developing locally and then deploying has become truly easy.
I really felt that kubernetes contributes greatly to carrying out the philosophy of containers.

That’s all.
I hope this is helpful.