Custom Compute Classes, CIS Benchmarks and Cloud Run AI

The News

GKE

• GKE Custom compute classes: I have been waiting for this feature for a while. You can define a list of compute configurations you want your workload to be schedule on. You can decide on the priority of configurations with a fallback options, scalability limits and hardware requirements. Check the docs this is pretty neat.
• GKE Backups support Committed use discounts: A few weeks ago we introduced CUD’s to GKE, Cloud Run and Compute Engine. Now you can use it with Backups for GKE also.
• CIS Benchmarks for GKE Standard and Autopilot: ?CIS Benchmarks are a set globally-recognized best practices for securing IT systems and data. We works with the Center for Internet Security?(CIS) to release specific benchmarks for GKE Standard and Autopilot.
• GKE Enterprise Observability Dashboard's Sample Library: GKE Enterprise added some sample dashboards for Observability to the GKE Dashboard. See the screenshot below and navigate to Google Kubernetes Engine > Dashboard > Sample Library to check them out.

Google Cloud

• Cloud Service Mesh and Gateway API: In Public Preview you can use the Gateway API to configure Cloud Service Mesh for either proxyless gRPC or Envoy proxy deployments.
• Run AI on Cloud Run event: We are running something cool with Cloud Run and AI. Dial into this event on August 21st to learn more.

The Editorial

GKE

• Serve Open Models with TPU’s and Optimum TPU: Optimum TPU?is an Open Source Serving framework from HugginFace. This article shows how you can use it with GKE and TPU’s to serve Gemma 2B or Llama3 8B models.
• Disable GKE Kubelet read-only port: A communication was sent a while ago to GKE users to advice them to disable the GKE Kubelet read-only unauthenticated port. If you missed it this article details why you should.
• Exposing gRPC on GKE through the Gateway API: The tutorial shows how to expose a gRPC app on GKE with Global Load balancers via the Gateway API.
• Custom Org Policy and Policy Controller to secure GKE: Kubernetes is a very complex system due to it’s multi-layer architecture and securing each layer requires special tooling and techniques. Custom Org Policies and Policy Controller can be used together to secure some of these layers. Learn how to use both in this article.
• Build a Search Engine with GKE and Vertex AI Agents: Not sure why you will want to build your own search engine while Google exist already. But In case you need to do so you can use GKE and VertexAI Agents to build one.

Google Cloud

• Check when a Service Account was used last: There are probably tons of Service Account keys roaming freely in the world in various servers, private githubs and drive stores. This article show you a simple trick to figure out if Service Accounts are still used and when was the last time they were used.

Tools of the week

• kubectl-cond: is a CLI plugin to monitor the conditions of specific objects in the cluster.