Since last week and to continue the beginning of my learning journey with GCP, this week I would like to deep dive with GKE, around few advanced setup and features.

First I needed to complete the setup of this blog on GKE by setting up my SSL certificate and map my DNS with the new Public IP address exposed. To accomplish this I’m using the Google-managed SSL certificates (which is still in Beta as we speak). Pretty straight forward by following the tutorial. I just needed to align the version of my cluster and the version of the Managed Certificates feature:

  • < 1.16.5-gke.1 = v1beta1
  • >= 1.16.5-gke.1 = v1beta2

Create a Public static IP address:

gcloud compute addresses create $staticIpName \
staticIpAddress=$(gcloud compute addresses describe $staticIpName \
    --global \
    --format "value(address)")

Grab this $staticIpAddress variable value and put it on my DNS where I host my domain name, it will guarantee that I own that domain for the following steps. Then, I need to expose myblog with a NodePort Service:

kubectl run myblog \
    --image=$imageName \
kubectl expose pod myblog \
    --type NodePort \
    --port=8080 \

Create the ManagedCertificate resource:

kubectl apply -f - <<EOF
kind: ManagedCertificate
  name: myblog
    - $domainName

And then, create the associated Ingress resource:

kubectl apply -f - <<EOF
kind: Ingress
  name: myblog
  annotations: $staticIpName myblog
    serviceName: myblog
    servicePort: 8080

After waiting for few minutes for the load balancer and the certificate to be provisioned, here we are, that’s it! 🤘 I have now myblog up-and-running with an SSL certificate configured: 🚀

That’s cool, but now let’s get more insights around the features of my current GKE cluster:

gcloud container clusters describe $clusterName \
    --zone $zone

addonsConfig.kubernetesDashboard.disabled: true, having the Kubernetes dashboard disabled by default is good practice on a security standpoint. If you need more UI interactions with your GKE clusters you could securely leverage the GKE Workloads menu in Cloud Console.

addonsConfig.networkPolicyConfig.disabled: true, Network Policies is an important feature for your Security Posture with Kubernetes. By default it’s not enabled, but good news, you could enable/disable Network Policy (Calico) on your GKE cluster without recreating your cluster: gcloud container clusters update --update-addons=NetworkPolicy=ENABLED or gcloud container clusters update --enable-network-policy. We could then apply few Network Policies like illustrated here.

currentMasterVersion and currentNodeVersion, you could learn more about Release channel: none, rapid, regular or stable. regular is the default one, and stable is the one eligible for the GKE’s SLA. This Release Channel is how you could manage your GKE’s version (master versus worker nodes, auto-upgrade, etc.). Complementary to this, we could see that management.autoUpgrade: true is by default, which is really great and important. Maintenance windows and exclustions features will bring flexibility and fine-grained control with your upgrades. Note: There is also the concept of Alpha clusters.

databaseEncryption.state: DECRYPTED, you have the option to encrypt your Kubernetes’ Secrets (etcd) with your own key via Cloud KMS.

location and zone: there is 3 choices regarding the availability of your cluster: Single-zone, Multi-zonal or Regional. If you choose a Region with 3 Zones, your will have the control plane nodes as well as the number of your nodes replicated on the 3 zones of that region. The choice to make your cluster Regional or Zonal will influence the GKE’s SLA too.

diskType: pd-standard, for better performance for your workloads, but with a cost, you may want to opt-in for a custom boot disk with SSD: pd-ssd.

machine-type: n1-standard-1 is by default with 1vCPU and 3.75GB memory (FYI: soon it will be e2-medium). I found very interesting this session Choosing the Right Compute Engine Instance Type for Your Workload to see the differences between the VM’s families and types. By default, you get benefit of the Sustained used discounts, and you could also get more discounts with more commitments with Commited use discounts. Finally, you may want to consume reservations with GKE.

nodeConfig.diskSizeGb: 100, choosing the proper size of your OS disk could be important, especially if you are facing application performance throttling. The section will guide you through the factors that affect performance between the OS disk and the Node type that you should be aware of.

Note: you could learn more about performance optimization based on the choices you could make regarding to the node’s disk setup, by reading this great article from PayPal.

nodeConfig.imageType: COS. Here are the different Node image types you could use. On my end, I would like to go with containerd for a security standpoint, you could read more about the differences between the Docker/Moby versus containerd approach to see what fits best for your own context. I could actually upgrade my existing cluster with the new COS_CONTAINERD image type: gcloud container clusters upgrade --image-type COS_CONTAINERD.

management.autoRepair: true, Nodes auto-repair is enabled by default with version 1.17+, which is important to guarantee your nodes are healthy.

shieldedNodes: {}, Shielded GKE nodes bring a more security node credentials boostrapping implementation, starting with version 1.18 clusters will have shielded GKE nodes by default. We could even update a current GKE cluster without this feature enabled at creation time: gcloud container clusters update --enable-shielded-nodes.

NodeLocal DNSCache, if you think you have Kubernetes’ DNS issues, great option for stability and performance within your cluster, especially with large clusters. Good news, you could enable/disable without recreating your cluster: gcloud container clusters update --update-addons NodeLocalDNS=ENABLED|DISABLED.

That’s a lot of concepts and there is more for sure (VPC, Monitoring, etc.), but anyway, that’s what you need to know and learn about to have a proper GKE cluster for Production. I don’t know for you but I love the flexibility with the gcloud container clusters update|upgrade commands to be able to enable/disable features on existing GKE cluster without having to recreate a cluster.

Last week I went with this very simple command line to create a straight forward GKE cluster: gcloud container clusters create --zone. With few concepts and features about resiliency, performance and security discussed throughout this blog article, I will better go now on with:

gcloud container clusters create \
    --release-channel rapid \
    --region \
    --disk-type pd-ssd \
    --machine-type n2d-standard-2 \
    --disk-size 256 \
    --image-type cos_containerd \
    --enable-network-policy \
    --addons NodeLocalDNS \
    --enable-shielded-nodes \
    --shielded-secure-boot \
    --enable-autorepair \

And there is more concepts and features not discussed in this blog article, so I will adjust this snippet accordingly as I’m diving more deeply with GCP/GKE.

Complementary resources:

Awesome learnings, isn’t it!? More to come during the coming weeks for sure, cheers!