Welcome to issue #410 August 5th, 2024

News

Bigtable transforms the developer experience with SQL support - Bigtable, a fast and flexible NoSQL database used by Google services like Search and YouTube, now supports GoogleSQL, an ANSI-compliant SQL dialect. With over 100 SQL functions at launch, Bigtable SQL support makes it easier to analyze and process large amounts of data directly within Bigtable, unlocking its potential for a wider range of use cases. Key use cases include NoSQL migrations, interactive and ever-changing data, low-latency lookup of non-uniform data, BigQuery serving layer, SQL-based time-series solutions, and write-time aggregates.

Introducing Spanner Graph: Graph databases reimagined - Spanner Graph reimagines graph data management by seamlessly integrating graph, relational, search, and AI capabilities with virtually unlimited scalability. It supports the ISO Graph Query Language (GQL) for intuitive graph pattern matching and traversal, and offers full interoperability between GQL and SQL, breaking down data silos and empowering developers to choose the optimal tool for each query. Spanner Graph also features built-in vector and full-text search capabilities, enabling efficient retrieval of graph data using semantic meaning and keywords.

Unlock the potential of your data: Build reliable and intelligent applications with Spanner editions - Spanner, Google's globally consistent database, introduces new editions with enhanced capabilities like Spanner Graph, full-text search, and vector search. Spanner editions offer a tiered pricing model with Standard, Enterprise, and Enterprise Plus options, providing cost transparency and optimization. The Standard edition includes all current generally available features, while Enterprise edition adds multi-model capabilities and Enterprise Plus offers the highest availability and data sovereignty features. Spanner editions will be available starting September 24, 2024.

Unlocking the power of your Cloud SQL data with Dataplex Catalog - Cloud SQL customers can now easily catalog their Cloud SQL data within Dataplex Catalog, Google Cloud’s next-generation data asset inventory platform. This integration provides a centralized repository for all Cloud SQL assets, allowing users to easily search, browse, and explore their Cloud SQL data.

Cloud SQL Enterprise Plus delivers enhanced performance, availability and disaster recovery for SQL Server - Cloud SQL Enterprise Plus for SQL Server is now generally available. It delivers enhanced performance with two new machine families, a data cache for improved read performance, advanced disaster recovery capabilities, and a 99.99% availability SLA. You can create a new instance of Cloud SQL Enterprise Plus using the Google Cloud console, CLI, or API. You can also upgrade your existing Cloud SQL for SQL Server instance in-place to the new Cloud SQL Enterprise Plus edition with minimal downtime.

Announcing IAM group authentication in Cloud SQL - Cloud SQL IAM group authentication allows customers to use Google Cloud Identity groups to manage access to Cloud SQL instances and databases.

Can’t stop, won’t stop: More innovations from the Google Cloud database portfolio - Google Cloud is introducing several new capabilities to make it easier for customers to build intelligent AI apps with Spanner, including Spanner Graph, vector search, and advanced full-text search. Bigtable SQL support and Bigtable distributed counters are also announced to transform the developer experience. Cloud SQL Enterprise Plus edition for SQL Server is launched, delivering improved read performance, higher availability, and advanced disaster recovery. A new strategic partnership with Oracle is formed to host Oracle database services within Google Cloud data centers, enabling easy migration of Oracle-based applications to Google Cloud.

Google Cloud announces new data analytics innovations to help fuel customers’ AI use cases - New data analytics innovations include Gemini in BigQuery for AI-powered experiences, Gemini in Looker for conversational analytics, a unified data platform with support for open formats and streaming, and simplified data processing and streaming with Apache Spark and Apache Kafka.

Understanding new Gemini model inference, tuning, grounding and safety settings in BigQuery - BigQuery now supports the latest Gemini models, including Gemini 1.5 Pro and Gemini 1.5 Flash, enabling users to analyze text, images, audio, and PDF files from a single model. Grounding with Google search and customizable safety settings ensure responsible AI responses, while support for Gemini 1.0 model tuning and evaluation allows further customization of AI capabilities.

Announcing PyTorch/XLA 2.4: A better Pallas and developer experience, plus “eager mode” - PyTorch/XLA 2.4 introduces several improvements to enhance developer experience and model performance. It includes updates to Pallas, a custom kernel language supporting TPUs and GPUs, with added support for Flash Attention, Paged Attention, and Megablocks' block sparse kernels.

Enhancing LLM quality and interpretability with the Vertex AI Gen AI Evaluation Service - The Vertex AI Gen AI Evaluation Service helps developers improve the quality and interpretability of large language models (LLMs) by generating diverse responses, automating the selection of the best response, and providing quality metrics and explanations. This workflow can be applied to any modality or use case, including text, images, and audio, and can be parallelized to minimize latency.

Now GA: Compute Engine C3 bare-metal and X4 instances - Google Cloud Compute Engine C3 bare-metal and X4 machine types are now generally available. X4 instances are ideal for extra-large, memory-intensive workloads such as relational in-memory databases and graph in-memory databases. C3 bare-metal instances are ideal for customers with workloads that require direct access to the machine’s CPU and memory resources, e.g., hypervisors or container platforms.

Build large-scale AI/ML and HPC clusters with Cluster Toolkit (formerly HPC Toolkit) - The Cluster Toolkit (formerly HPC Toolkit) simplifies the creation and management of high-performance computing environments on Google Cloud. It offers easy deployment and management of clusters, quickstart options for HPC and AI/ML workloads, integration of Google Cloud best practices, regular updates and new features, and open-source accessibility.

AI capabilities and controls to power up your workday with Chrome Enterprise - Chrome Enterprise introduces AI-powered shortcuts to streamline workflows and enhance productivity. Features like Google Lens, Gemini, Tab Compare, and Help me write leverage AI to simplify tasks, improve search, and assist with writing. IT teams gain granular controls to manage AI usage, ensuring alignment with company policies and security standards. Chrome Enterprise Premium customers can block specific activities on unsanctioned AI tools.

Google Cloud Private Marketplace, now GA, helps control costs and maintain governance - Google Cloud Private Marketplace helps organizations control costs and maintain governance by curating a collection of vetted products specific to their needs. It enhances compliance and security with product-level access control, empowering end users to discover new products while maintaining control.

Collect OTLP data from GKE apps with new curated OpenTelemetry ingestion pipeline - Google Cloud has released a new curated OpenTelemetry Protocol (OTLP) ingestion pipeline for Google Kubernetes Engine (GKE). This pipeline simplifies the process of collecting OTLP data from instrumented applications and exporting it to Cloud Observability.

Launching a landmark partnership to support AI startups with Y Combinator - Google Cloud and Y Combinator have partnered to support AI startups. YC startups will receive access to Google Cloud's AI platform, including GPUs, TPUs, cloud credits, and support and mentorship.

mWISE Conference 2024: Your front-row seat to the future of cybersecurity - mWISE Conference 2024, a two-day vendor-agnostic event on September 18-19 in Denver, offers cybersecurity professionals the opportunity to immerse themselves in the latest threat intelligence, cutting-edge tools, and engage with industry leaders. Experts from Mandiant, Google Cloud, and the wider cybersecurity community will share insights, experiences, and battle-tested solutions to defend against impactful threats.

Articles, Tutorials

Infrastructure, Networking, Security, Kubernetes

UNC4393 Goes Gently into the SILENTNIGHT

Cloud CISO Perspectives: Why water security can’t wait - In this edition of Cloud CISO Perspectives, Sandra Joyce, VP of Google Threat Intelligence, discusses the urgent need to prioritize water security in the face of increasing cybersecurity risks. She highlights recent incidents involving Russian hackers disrupting a water facility in Texas and the targeting of U.S. water and wastewater treatment systems by hackers from China and Iran.

What’s next with Istio Service Mesh? - Istio Ambient Mesh, a new data plane for Istio service mesh, eliminates the need for sidecar proxies by introducing a shared proxy layer at the node level, reducing resource overhead and simplifying deployment and management. It offers flexibility by allowing mixing of both sidecar and sidecarless deployment modes in the same environment and the option to deploy only L4 proxies or inject L7 ones later.

Events, and Lists, and Rules, oh my! - In this blog post, the author shares their experience setting up a lab environment with Google Security Operations to learn and experiment with the platform.

App Development, Serverless, Databases, DevOps

Accelerate PostgreSQL migrations to Google Cloud: move terabytes of data with Database Migration Service - Database Migration Service (DMS) offers a comprehensive way to migrate data to Google Cloud databases, including PostgreSQL to Cloud SQL and AlloyDB. To improve migration speeds, you can adjust settings for parallel initial load and change data capture (CDC), configure PostgreSQL database parameters, and optimize machine and network configurations. For example, setting the `max_wal_size` parameter on the destination instance to a range of 20GB-50GB can reduce checkpoint frequency and improve resource allocation for migration processes. Additionally, increasing the `shared_buffers` parameter on the source instance to 60% of RAM can enhance initial load performance.

Simplifying VM deployments on Google Cloud Marketplace with a Terraform-based UI - Google Cloud Marketplace now offers a simplified way to deploy VMs using Terraform directly from its user interface. With just a few clicks, you can deploy enterprise-grade cloud solutions from partners like F5, Inc. You can also switch to command-line deployment or download the Terraform module for integration with your CI/CD pipeline.

Best practices for streamlining log centralization with Cloud Logging - Centralize log management with Cloud Logging for unified visibility, efficient management, enhanced security, and streamlined operations. Utilize aggregated sinks for efficient routing, establish a central observability project as a management hub, customize log storage for optimal retention and cost efficiency, and manage log storage access control for data security. Monitor log volume to proactively manage storage costs and investigate anomalies.

Maxxton improves app performance up to 60% with Bare Metal for Oracle on Google Cloud - Maxxton, a provider of property management systems for the hospitality industry, improved app performance by up to 60% by migrating to Bare Metal Solution for Oracle on Google Cloud.

Automate or Stagnate: Master Salesforce Task Automation with Python and Google Cloud Platform - Salesforce task automation with Python and Google Cloud Platform streamlines operations, reduces manual effort, and ensures timely attention to at-risk partners.

[SOLVED] Google App Engine deployment fails because of failing readiness check - A custom Google App Engine environment fails to start due to failing readiness checks. The issue was resolved by adding a CMD statement to the Dockerfile and reverting to default health check settings. The default nginx instance shipped by the Google base container handles health checks.

Bulk closing alerts with Python and the Google Security Operations API - This article presents a method for bulk closing alerts in Google Security Operations using Python and the Google Security Operations API.

AlloyDB Adaptive AutoVacuum and how AlloyDB Cluster Storage Space is Released. - Real World Example of Adaptive AutoVacuum in AlloyDB and AlloyDB Cluster Storage Shrink/Usage in Action.

Big Data, Analytics, ML&AI

Edit Your BigQuery Columns Without Ever Dropping A Table - Prevent BigQuery type errors and table deletions with a 5-minute method to edit columns without putting your table at risk.

Automatically Update BigQuery View Schema Changes

Deploying dbt Projects at Scale on Google Cloud - Containerising and running dbt projects with Artifact Registry, Cloud Composer, GitHub Actions and dbt-airflow.

Logging new Airflow DAG entires in Cloud Composer - DAG Upload Audit.

Load Balancing Blitz — data pipeline - This blog post explores a near real-time data pipeline to gather metrics for a demo game called Load Balancing Blitz. Pub/Sub, BigQuery, and Looker were used to ingest, process, and visualize data in real-time.

New strides in making AI accessible for every enterprise - Google Cloud's Gemini 1.5 update brings significant cost reductions, with input costs slashed by up to 85% and output costs by up to 80%. Additionally, Gemini 1.5 Flash and Pro now support over 100 languages, enabling global users to interact in their native tongues. Provisioned Throughput in Vertex AI, coupled with a 99.5% uptime SLA, ensures predictable performance for enterprise-scale AI applications.

Unlock personalized web experiences: Enhancing user context with Vertex AI embeddings - PLAID uses Vertex AI embeddings to improve customer support. Embeddings help KARTE, PLAID's real-time analytics platform, better understand user intent, providing relevant recommendations and boosting customer satisfaction. By tuning the embeddings model, PLAID achieved significant improvements in recommendation accuracy, demonstrating the effectiveness of embedding-based models for web content recommendations.

Virtual Try On for Saree - Meesho, an Indian e-commerce platform, partnered with Google Cloud Consulting to develop a Virtual Try On solution for sarees. The solution utilizes Google Cloud Platform to process and enhance saree images, enabling end suppliers to easily upload and showcase their products.

Designing Generative AI Solutions: Key Lessons Learned - Generative AI is transforming how we interact with technology. Google Cloud's Applied AI Engineering team shares valuable lessons learned from designing and developing generative AI solutions. They emphasize three overarching design principles: offering frictionless experiences, building trust through transparency, and prioritizing goals. These principles help create trustworthy, empowering, and beneficial AI solutions for users and businesses.

Sentiment analysis with few-shot prompting - This article explores sentiment analysis using few-shot prompting with Gemini and LangChain4j. It demonstrates three approaches: using a big string of inputs/outputs, a low-level list of messages, and an AiServices abstraction.

Retrieval Augmented Generation (RAG) with Cloud SQL for MySQL - How to build a Generative AI application with your Cloud SQL for MySQL database.

Putting Generative Agents behind authentication and controlling their Access. - This article discusses how to build an authenticated generative AI agent on Google Cloud. The article provides a step-by-step guide on setting up authentication using Google as the identity provider, implementing authorization through OAuth2 scopes, and enabling auditability through conversation history tracking.

Gemini Code Assist - Gemini Code Assist, a new AI-powered extension for popular IDEs like Visual Studio Code and IntelliJ IDEA, aims to revolutionize software development by assisting engineers with various tasks. It leverages advanced features like context-aware prompting, license attribution, code customizations, and full codebase awareness to provide high-quality code suggestions and make coding faster and more efficient.

Beyond temperature: Tuning LLM output with top-k and top-p - What are top-k and top-p? What do they mean, how do they work, and how can they be tuned?

Various

Toronto Meetup: Building Serverless AI/ML Applications on Google Cloud Run with Wietse Venema - Learn how to build Generative AI applications using Google Gemini and Vertex AI, and deploy them as Serverless Containers (that scale to zero and require no operations) on Cloud Run.

Slides, Videos, Audio

Security Podcast - #183 Cloud Security Journeys: Improve, Evolve, Transform with Cloud Customers.

Releases

GKE new features - The NVIDIA GPU Operator can now be used as an alternative to fully managed GKE for both Container-Optimized OS and Ubuntu node images. You can now enable NCCL Fast Socket on your multi-GPU Autopilot workloads. You can now keep a GKE Standard cluster on a minor version for longer with the Extended release channel.

Load Balancing - Cloud Load Balancing now supports failover for global, classic, and regional external Application Load Balancers.

Marketplace - Google Private Marketplace and Product Requests are now Generally Available.

StratoZone - StratoZone is scheduled for end of sale on July 30, 2024, and will reach end of life (complete shutdown) on October 1, 2024.

Migration Center - StratoZone is scheduled for end of sale on July 30, 2024, and will reach end of life (complete shutdown) on October 1, 2024. The Migration Center discovery client v5 is scheduled for end of sale on July 30, 2024, and will reach end of life (complete shutdown) on June 30, 2025. The Rapid Migration Assessment API is scheduled for end of sale on July 30, 2024, and will reach end of life (complete shutdown) on June 30, 2025.

Cloud Monitoring - You can now troubleshoot Compute Engine issues involving host events, MIG autoscaling and health-check failures, resource-availability errors, and VM performance by using the new "interactive playbook" dashboards in Cloud Monitoring.

Security Command Center - Detector for Container Threat Detection released to General Availability Container Threat Detection, a built-in service available in Security Command Center Premium and Enterprise, has launched a new detector, Unexpected Child Shell, in General Availability. Assign high-value resources based on Sensitive Data Protection insights for Cloud Storage The attack path simulations feature can now automatically set the resource value of a Cloud Storage resource based on the sensitivity of the data that the bucket contains. Preview of curated detections for Microsoft Azure data New curated detections in the Cloud Threats category that identify suspicious patterns in Microsoft Azure data are currently available in Preview.

Sensitive Data Protection - The DATE_OF_BIRTH infoType detection model that was previously only accessible by setting InfoType.version to latest has been promoted to be the default detection model for this infoType.

Service Mesh - 1.20.x & 1.21.x & 1.22.x. Cloud Service Mesh is incompatible with Envoy version v1.31.0.

Cloud Spanner - Spanner now offers Spanner Graph in Preview, which unites purpose-built graph database capabilities with Spanner. Spanner full-text search (Preview) lets you search a table to find words, phrases, or integers, instead of just searching for exact matches in structured fields.

Cloud SQL MySQL - Gemini in Databases assistance in Cloud SQL for MySQL is now available in Preview for query insights, system insights, index advisor, and active queries. You can now use the gcloud CLI or the Cloud SQL Admin API to switch the storage location of transaction logs used for point-in-time recovery (PITR) on your instance from disk to Cloud Storage. Migrating your external MySQL 5.7 and 8.0 databases into Cloud SQL for MySQL by using Percona XtraBackup physical files is now generally available (GA).

Cloud SQL Postgres - Gemini in Databases assistance in Cloud SQL for PostgreSQL is now available in Preview for query insights, system insights, index advisor, and active queries.

Cloud SQL SQL Server - Cloud SQL for SQL Server now offers two editions of Cloud SQL to support your various business and application needs: Cloud SQL Enterprise Plus edition and Cloud SQL Enterprise edition.

Cloud Storage - Beginning November 1, 2024, BigQuery users will start seeing charges for Cloud Storage usage as per pricing documentation, which was not metered before due to a billing bug. You can now use list caching with Cloud Storage FUSE, which is a cache for directory and file list, or ls, responses that improves list operation speeds. You can now use the Google Cloud console to set a default soft delete retention duration. A weekly digest of client library updates from across the Cloud SDK.

Vertex AI - The Vertex AI Model Registry now offers Preview support for model copy across different projects.

VMware Engine - You can now purchase Fully licensed commitments, Fully licensed convertible commitments, and Portable License commitments.

VPC Service Controls - VPC Service Controls feature: VPC Service Controls supports using identity groups and third-party identities (only single identities) in ingress and egress rules to allow access to resources protected by service perimeters.

AlloyDB - Gemini in Databases assistance in AlloyDB for PostgreSQL is now available in Preview for standard and enhanced query insights, system insights, index advisor, and active queries.

Google Distributed Cloud Bare Metal - 1.28. Release 1.28.800-gke.111 Google Distributed Cloud for bare metal 1.28.800-gke.111 is now available for download. Functionality changes: Updated Kubernetes audit logging to include request and response payloads from the Kubernetes API server for bare metal custom resources, including the following: Cluster, NodePool, BareMetalMachine and BareMetalCluster. Fixes: The following container image security vulnerabilities have been fixed in 1.28.800-gke.111: Medium-severity container vulnerabilities: CVE-2024-33599 CVE-2024-33600 CVE-2024-33601 CVE-2024-33602 CVE-2024-35255. Known issues: For information about the latest known issues, see Google Distributed Cloud for bare metal known issues in the Troubleshooting section.

GKE on AWS - You can now launch clusters with the following Kubernetes versions.

Anthos clusters on Azure - You can now launch clusters with the following Kubernetes versions 1.29.6-gke.600, 1.28.11-gke.600, 1.27.14-gke.1600.

Apigee Integration - Redirecting from Apigee Integration to Application Integration We are converging Apigee Integration and Application Integration.

Apigee Advanced API Security - The preview release of generative AI summaries and recommendations for Advanced API Security Abuse Detection incidents is now re-enabled after resolution of the known issue noted on July 19.

Apigee X - On August 1, 2024, we released an updated version of Apigee (1-13-0-apigee-1). New flow variables are now available: request.headers.names.string request.queryparams.names.string request.formparams.names.string message.headers.names.string message.queryparams.names.string message.formparams.names.string response.headers.names.string These context variables can be used to return header, query parameter, and form parameter names in string format that can be used in API proxy logic. Bug ID Description 308583363, 332464869 Security fix for apigee-mart. Bug ID Description 293150694now supports theelement and can trust TLS certs from non-public CAs. On July 30, 2024, we released an updated version of Apigee. With this release, Apigee expanded its support for data residency to an additional region in Europe: europe-west6 (Zurich).

Apigee Hybrid - On July 31, 2024 we released an updated version of the Apigee hybrid software, 1.11.2-hotfix.2. Bug ID Description 351868444 Tolerations are now working for Redis's Envoy pod. 1.12.1-hotfix.1. hybrid 1.12.1-hotfix.1 On July 30, 2024 we released an updated version of the Apigee hybrid software, 1.12.1-hotfix.1. Bug ID Description 351868444 Tolerations are now working for Redis's Envoy pod.

AppEngine Flexible Ruby - Ruby 3.3 is now generally available. In the App Engine page in the Google Cloud console, you can now filter your existing App Engine versions by runtime lifecycle stages.

AppEngine Standard Ruby - Ruby 3.3 is now generally available. In the App Engine page in the Google Cloud console, you can now filter your existing App Engine versions by runtime lifecycle stages.

Cloud Asset Inventory - The following resource types are now publicly available through the ExportAssets, ListAssets, BatchGetAssetsHistory, QueryAssets, Feed, and Search (SearchAllResources, SearchAllIamPolicies) APIs.

BigQuery - When you translate SQL queries from your source database, you can use configuration YAML files to optimize and improve the performance of your translated SQL. Workload management now provides the following benefits: The autoscaler now scales up immediately. You can now use the output_dimensionality argument of the ML.GENERATE_EMBEDDING function when you use the function with a remote model based on a Vertex AI multimodalembedding model. The RANGE data type is now a supported JSON encoding. You can now use the administrative jobs explorer to help you quickly monitor jobs activity across your organization. Vector indexes support the TreeAH index type, which uses Google's ScaNN algorithm.

BigTable - Bigtable supports querying in SQL. The Bigtable Studio query editor is available in Preview. Bigtable aggregates let you build distributed counters and aggregate your Bigtable data at write time using min, max, sum, or HLL.

Capacity Planner - Preview: Capacity planner supports the following for data aggregated by folder ID: View and export the actual and forecasted usage data of the VMs, Persistent Disk volumes, and GPUs in your folder.

Chronicle - Curated Detections has been enhanced with new detection content for Cloud Threats to include rule packs covering Microsoft Entra ID, Entra ID Audit and Azure Compute and are in public preview for customers with a Google Security Operations or Enterprise Plus license.

Chronicle Security Operations - Customers can now configure direct ingestion of Google Cloud data without using a 1-time Google Security Operations access code.

Chronicle SOAR - Release 6.3.13 is now in General Availability. Release 6.3.14 is currently in Preview. Unable to rerun a failed playbook step when the parameter is very large. The platform does not show the correct error when trying to save a playbook which is open in another tab. Can't remove the remote agent after host/container has stopped (ID #49024310). List and multi-select parameters not appearing correctly in the IDE (ID # 51995565). Playbook simulator sometimes not executing actions in the correct order (ID #48264534). Missing audit log entries when deleting permission groups (ID #51496411). Release 6.3.12 is now in General Availability.

Cloud Composer - In July and August 2024, new Cloud Composer 2 environments keep switching to using GKE 1.29 and PSC as a connectivity channel to the GKE control plane. (Cloud Composer 3 only) The July 24th issue with Airflow upgrade operations for Cloud Composer 3 environments is now fixed.

Compute Engine - Generally available: You can use instant snapshots to take in-place disk backups that can be restored to new disks in under a minute.

Database Migration Service - Database Migration Service support for homogeneous MySQL to Cloud SQL for MySQL migrations with physical backup files created by using the Percona XtraBackup utility is now Generally Available (GA).

Dataform - The maximum size limit for repositories encrypted with customer-managed encryption keys (CMEK) is 512 MB.

Dataplex - Metadata import for Dataplex Catalog entries and their aspects is available in preview.

Dataproc Serverless - New Dataproc Serverless for Spark runtime versions: 1.1.74 1.2.18 2.0.82 2.2.18. Dataproc Serverless for Spark: Upgraded Spark BigQuery connector to version 0.36.4 in the latest 1.2 and 2.2 Dataproc Serverless for Spark runtime versions.

Datastore - You can now apply range and inequality filters to multiple fields in a query.

Cloud Data Loss Prevention - The DATE_OF_BIRTH infoType detection model that was previously only accessible by setting InfoType.version to latest has been promoted to be the default detection model for this infoType.

Cloud Networking Products - Health checks for external endpoints in Cloud DNS routing policies are now available in Preview.

Cloud Firestore - You can now apply range and inequality filters to multiple fields in a query.

Cloud Functions - Cloud Functions now supports the Ruby 3.3 runtime at the General Availability release level.

IAM - You can use IAM attributes in custom organization policies to control how your allow policies can be modified.

Backup for GKE - Backup for GKE now offers committed use discounts (CUDs) that provide 20% discount (one year term) and 45% discount (3 year term) on backup management fee in exchange for your commitment.

Google Kubernetes Engine - The NVIDIA GPU Operator can now be used as an alternative to fully managed GKE for both Container-Optimized OS and Ubuntu node images. You can now enable NCCL Fast Socket on your multi-GPU Autopilot workloads. Autopilot nodes use Google Virtual NIC (gVNIC) for 1.30.2-gke.1023000 and later. (2024-R27) Version updates GKE cluster versions have been updated. You can now keep a GKE Standard cluster on a minor version for longer with the Extended release channel. GKE Autopilot versions 1.28.6-gke.1317000 and later fixes a known issue for ephemeral storage requests.