GKE Extended Support. Static Pods IP’s and Ray Add-on

The News

GKE

• GKE Extended Support: It sounds like we are following the trend in the market. GKE will support minor version for up to 24 months (vs 12 months upstream). So you can keep your cluster within the same minor version and only get patches for an additional 12 months beyond it’s EOL upstream.
• Persistent IP addresses for Pods: This is something we heard from customers a lot. I’m glad we have a feature for it now. You can assign one or more persistent IP addresses to specific Pods. The feature is useful if you are making outbound calls to an external endpoint which allowlist based on IP’s. Now you can make sure your outbound traffic will use the same set of IP’s. Think of it as NAT for Pods [Preview].
• GKE is now publishing SLSA VSA: GKE is now publishing SLSA VSA (Security Levels of Software Artifcats Verification Summary Attestation). VSA allows users to verify that an artifact is meeting a specific SLSA level and how that verification took place. GKE is the first GCP product to publish those. We will be publishing an article to explain this better.
• Ray Operator on GKE is GA: Ray is a popular Python framework for writing and executing distributed apps and used for AI/ML workloads. We added an Add-on that allows users to deploy Ray on GKE with one click (or one flag in gcloud cli). This is available for GKE 1.30+ so you will have to select the Rapid channel.
• Preloading container images on disk is GA: You can create a disk with a pre-loaded image (or data) and attach it to GKE as a secondary disk. This way you speed up loading time for data. This is useful for large ML models. Starting in GKE version 1.28.3-gke.1067000 in Standard clusters and in GKE version 1.30.1-gke.1329000 in Autopilot.
• GKE Autopilot support Flexible CUD’s: Flexible Committed-Use Discounts(CUD’s) are a way for customers to commit to a certain spend and receive discounts. This now available for GCE, GKE and Cloud Run across any project, region belonging to a single Billing Account.
• Monitor Nvidia GPU’s with DCGM: Send?NVIDIA Data Center GPU Manager (DCGM) metrics?to Cloud Monitoring to monitor how your GPU’s are performing.
• Globally distributed applications on GKE with Gateway and Cloud Service Mesh: It took months for this solution guide to be published but I'm glad it's out. We updated our Edge to Mesh solution to use Cloud Service Mesh and Gateway API with multi-cluster setup. Now you have a full solution for running globally available Apps on GKE.

Google Cloud

• Free Google Gemini API Course on Udacity: Learn how to use the Gemini API to build GenAI Apps with this free course.
• Limit the run time for a VM: I don’t know who might need this feature but you can now set a time limit on how long a VM would be up for. After the time runout we will stop or delete the VM for you.
• Disable Cloud Run URL: Cloud Run generate an URL for your service but you can also choose to make your service private and hide it behind a LoadBalancer. In the past even if your service is private the generated URL will remain in place (even if it’s not used). Now you can disable it all-together.

The Editorial

GKE

• Leverage the Secret Manager add-on for GKE: Learn how to use the Secret Manager GKE Add-on to manage secrets on GKE.
• GKE Security Posture: Learn how GKE Security Posture can help monitor and secure you GKE clusters.
• Finetune an LLM with LLaMa-Factory on GKE: LLaMA-Factory is an open source tool used for fine-tuning large language models. In this tutorial you will see how you can use it with GKE.
• Performing FleetOps with GKE Enterprise: This article goes into the details of how GKE Enterprise fleet can allow you to manage Clusters at scale in a consistent way.
• Simplifying VPC Service Control: Do you know what VPC SC (Service Controls) are? Me Neither. This tutorial got your back. It covers the basics you need to know.

Kubernetes

• Argo Rollouts Now Supports Version 1.0 of the Kubernetes Gateway API: Argo added native support for the Gateway API v1.0. You can now perform Canary or Blue/Green with a simple config.
• An alternative to Helm? There is a new kid in the block. Nelm which is part of the Werf project a CNCF Sandbox project and claims to be a good replacement for helm (or at least that’s what the cool kids are saying).

Google Cloud

• Accessing Google API’s without the Internet: You don’t need to put your VM’s or GKE cluster on the Internet to access Google API’s (BigQuery, GCS…etc) you can use Private Google Access (PGA) or Private Service Connect (PSC) to achieve that. Read how in this tutorial.

Tools of the week

• https://github.com/kluctl/kluctl: This tool puts together your (and any third-party) deployments into one large declarative Kubernetes deployment.

Video, audio, podcasts

• Kubernetes Podcast Dapr Episode: I have been hearing about Dapr for a while now. It’s a pretty popular tool and is part of the CNCF. I’m glad I managed to sit down with Mauricio to talk Dapr, Java, Knative and more.