TLDR DevOps 2025-12-17

According to a recent Google Cloud + Chronosphere report:

• 59% of orgs feel “very confident” in production readinesses, but almost 25% encounter downtime during releases.
• >92% monitor apps frequently or continuously, but many still report limited visibility.

Find out what's going wrong (and right) by grabbing a copy of Unveiling the Landscape of Day 1: Release Readiness in Cloud-Native CI/CD. This report looks at recent survey data to show how orgs are preparing for day 1 -- with practical advice on how to prepare your apps, infrastructure, and processes for Day 1 success.

⬇️ Download the eBook

The new CoreDNS release introduces initial DoH3 support, core performance and stability fixes, and plugin improvements across forwarding, GeoIP, cache, and orchestration integrations. GeoIP has deprecated returning 0,0 for missing coordinates, switching to empty values in the next release.

exe.dev is launching a public developer preview of a VM hosting service that lets users spin up many real, fast-starting Ubuntu VMs via an SSH-based API, sharing a fixed CPU/RAM budget across as many VMs as they want. The platform focuses on private-by-default, persistent, agent-friendly VMs with sub-second startup and no per-VM marginal cost, aiming to make running lots of small tools and agents frictionless.

Whole-system observability using metrics and sampling is cheap and effective for capacity planning and fault detection, but it cannot explain what happens inside individual operations. Per-process tracing provides exact timelines needed to identify and prioritize the highest-ROI optimizations, and can be implemented efficiently using techniques like leveraging early-exit syscall paths to correlate application events with system traces.

The Kube Resource Orchestrator (kro) has been integrated into Amazon's new EKS capabilities, signaling major cloud provider recognition of its potential to simplify Kubernetes-native resource grouping for platform engineers. The powerful mechanism abstracts complex Kubernetes objects through ResourceGraphDefinitions, enhancing composition while integrating with broader orchestration frameworks.

Machine Learning Engineering (MLE) has emerged as a critical and highly lucrative field, serving as the essential bridge between data science research and the deployment of production-grade artificial intelligence systems at scale. This in-demand role requires a diverse skill set encompassing programming, ML algorithms, and data engineering, all guided by a proven methodology to ensure successful and maintainable solutions.

Kubernetes 1.34 adds dynamic resource allocation, enabling precise GPU and accelerator requests beyond simple counts. An upcoming workload abstraction aims to coordinate multinode AI pods with all-or-nothing scheduling, reshaping how complex AI workloads are deployed.

AI ETL platforms are automating data integration, leveraging machine learning to dynamically map schemas, transform data, and detect quality issues. This significantly overcomes traditional ETL's struggles with complex unstructured data and schema evolution by automatically adapting to changing conditions and optimizing performance.

Microsoft has announced major US cloud infrastructure expansions to meet growing cloud and AI demand, including a new East US 3 Azure region in the Atlanta area opening in early 2027, expanded Availability Zones across multiple existing regions through 2026–2027, and increased capacity for commercial and government customers with a focus on resiliency, sustainability, and secure, compliant architectures.

TestNitro is a fully managed service that delivers 100% automatable coverage in weeks — with only human-verified bugs reported. Get 15x faster scripting without AI-washing. Start here

Docker Sandboxes and the MCP Toolkit can run AI agents like Kiro in isolated containers, preventing access to host files and credentials while enabling safe code changes, testing, and tool use.

Google is celebrating GKE's 10th anniversary by advancing Kubernetes for AI and agentic workloads with Agent Sandbox, massive scale up to 130,000 nodes, faster autoscaling, and optimized inference.