Google Cloud ADK Technical Guide Released

From manual clicks to IaC: Why Customer contributions matter 🚀 I’m thrilled to share that my latest contribution to the Google Cloud Terraform provider was recently merged! As a practitioner working at de Bijenkorf, a Google Cloud customer, I’m a firm believer that users make the best contributors. We are the ones in the trenches, feeling the friction of manual "ClickOps" every day. Recently, we encountered a gap in the automation workflow: there was no way to manage Cloud Logging Saved Queries via Terraform. Refusing to settle for ClickOps, I decided to build the solution. I’ve added the new `google_logging_saved_query` resource to the provider. The Customer Perspective When customers contribute to Open Source, everyone wins. We bring real-world requirements to the table, ensuring the tools we all use actually solve the problems we face in production. Why this resource is a game-changer 🔹 Scalability: Stop manually recreating queries across your projects. 🔹 Auditability: Version-control your shared filters and security dashboards just like the rest of your (logging) infra. 🔹 Consistency: Standardize your debugging and audit queries across your entire organization. A huge thanks to the maintainers at Google for the smooth review process! Ramon Vermeulen, thanks for the help and the initial review. This shows why it matters to have an empowering partner like Xebia! This is a perfect example of how a great partner empowers a customer to give back to the community. Check out the PR here: https://lnkd.in/euvSTsbg Check out the resource here: https://lnkd.in/eviH3JB3 #GoogleCloud #Terraform #OpenSource #CloudLogging #PlatformEngineering #GCP #Xebia

Serverless doesn't magically solve state management. We often reach for AWS Lambda or Cloud Functions for their auto-scaling and pay-per-use benefits. They excel at stateless, event-driven tasks. But the model breaks down when you introduce long-lived connections or in-memory state. A classic example is building a real-time collaboration feature using WebSockets. An API Gateway WebSocket connection might trigger a Lambda function, but that function is ephemeral. It can't maintain the persistent connection required for a real-time session. It processes its event and terminates, dropping the state along with it. For this pattern, a long-running container on a service like AWS Fargate or Google Cloud Run is often a better fit. The container can hold the WebSocket connection open, manage session state in memory, and broadcast messages to connected clients efficiently. The right architecture often involves using both serverless functions and long-running containers, not choosing one over the other. Use Lambda for the stateless API endpoints and Fargate for the stateful, real-time components. #SystemDesign #CloudArchitecture #Serverless

We just published a quick write-up on how we closed ECS auto scaling gaps with a Lambda driven approach 🚀 Native ECS scaling works well, but in real environments, we needed tighter control, centralized logic, and smarter safety checks across services. By moving scaling decisions into Lambda and letting capacity providers respond to real demand, we gained more predictable behavior and cleaner control. If you’re running ECS at scale, this may help. Read it here: https://lnkd.in/gDbPXNdK #TrinityDevOps #CloudScaling #AmazonECS #AWSLambda #AWS

🌪️ Chaos Engineering - Building Resilience the Smart Way 🚫 Not breaking things randomly ✅ Testing resilience on purpose ⚙️ Modern cloud systems are complex (microservices, multi-region, async). Traditional tests miss real failures — chaos engineering finds them early. 🧪 The formula that matters: Steady state → Hypothesis → Controlled failure → Learn 🔍 It reveals hidden dependencies, weak failovers, and risky assumptions. 🛡️ Result? Less firefighting 🚒 More confidence 💪 Stronger systems 🚀 Check it out how Chaos Engineering on Google Cloud - https://lnkd.in/dUX942qM #ChaosEngineering #Google #Cloud #DevOps #SRE #CloudNative #Resilience #Observability

Google might have just changed how we use “localhost.” Gemini Code Studio on Vertex AI doesn’t just give code suggestions. It creates a full cloud development environment for you. You can say, “Build a React dashboard for this CSV,” and it: • Sets up the container • Installs dependencies • Writes the code • Gives you a live preview link No local setup. No environment issues. It feels like AI-powered cloud development built directly on Google’s infrastructure. This could seriously speed up how we build and test ideas. Would you try this instead of working on localhost?

AWS Insight Monday Cost Implications of AWS’s Newest Services: What You Need to Know Our analysis of $2M in AWS spending reveals that new services can inflate costs by up to 30% if not monitored closely. Here's what we discovered: • 25% of users overspend on underutilized EC2 instances. • New Lambda pricing can increase costs by 20% for sporadic workloads. • 40% of S3 users are not leveraging Intelligent-Tiering. Most common mistake: Ignoring service-specific pricing changes. 💡 Quick fix: Regularly review your billing dashboard and set alerts for new service launches. What's your experience with managing costs on new AWS services? #AWSCosts #CloudOptimization #DevOps #CloudWise

We have been working a lot with OpenAI's Agent module, to orchestrate multiple agents into a user-based workflow. In this short video I show an example of how the OpenAI agentkit builder maps to a user interface module on BITE Data that we are building. There is quite a lot of integration on the back-end that is still required and lots we are still learning. Traversing the line of sight from the OpenAI --> Amazon Web Services (AWS) (Cloud) --> API --> MCP --> Server --> Client is a decent amount of work but they way we've architected our microservices is helping with this. We are looking to release our first module using the agent builder in a month and will be testing it pretty thoroughly prior to release.

Google launched the Developer Knowledge API and MCP Server in public preview! 🚀 LLMs are only as effective as the context they are given. To solve the challenge of outdated training data and fragile web-scraping, Google is providing a canonical, machine-readable gateway to official documentation (Android, Firebase, Cloud, and more). Key highlights: ✅ Real-time Truth: Documentation is re-indexed within 24 hours of updates. ✅ Markdown Ready: Retrieve docs as clean Markdown for easy LLM ingestion. ✅ MCP Integration: An official Model Context Protocol server that plugs directly into your IDE, CLI or AI agents. Check out the documentation --> https://lnkd.in/dsBcZiNi #GoogleDevelopers #GenerativeAI #MCP #SoftwareEngineering #CloudComputing #Firebase

A startup was paying $11k/month on AWS. After a 2-week audit, we brought it down to $6.5k. Same performance. Same uptime. Here's what I found (and I see this on almost every project): → 14 EC2 instances running 24/7. Actual peak usage? 3 hours a day. We moved non-critical workloads to spot instances and set up autoscaling. Savings: ~$2k/mo. → A 3-node Elasticsearch cluster for "logging." Nobody had queried it in 4 months. The team had switched to CloudWatch months ago and forgot to decommission it. Savings: ~$1,200/mo for doing literally nothing — just turning it off. → $1,100/mo in cross-AZ data transfer. Microservices calling each other across availability zones on every request. Moved tightly coupled services into the same AZ + added a caching layer. Savings: ~$800/mo. → Three NAT Gateways processing traffic that could go through VPC endpoints. Savings: ~$500/mo. → Oversized instances everywhere. A t3.2xlarge running at 8% CPU average. Right-sized across the board. The thing is — none of this was complicated. No architectural redesign. No migration. Just looking at what's actually running vs what's actually needed. Most companies overspend on cloud by 25-40% without realizing it. Not because they're careless — because nobody has time to look.

If you’re building cloud apps and your pipeline still feels “fragile” (deploys break, security is bolted on later, rollbacks are manual), this article gives a practical reference CI/CD pipeline you can adapt quickly—structured around a clean lifecycle: Build → Scan → Deploy → Rollback, with real-world steps and decisions that help teams ship faster and safer. CI/CD works best when it’s a repeatable loop you can trust: ✅ Build → reproducible builds, versioned artifacts, consistent environments ✅ Scan → SAST/SCA + container/image scanning + policy gates before prod ✅ Deploy → automated releases with approvals, progressive delivery, observability checks ✅ Rollback → instant recovery using versions, blue/green or canary strategies, database-safe rollbacks Read the full reference pipeline here: https://lnkd.in/g4_brWYr #CICD #DevOps #DevSecOps #Cloud #PlatformEngineering #Kubernetes #SRE #Automation #SoftwareDelivery #AWS #Azure #GCP