TLDR DevOps 2026-01-02

Cloudflare Radar 2025 reported 19% internet traffic growth, Googlebot dominance, aggressive AI crawling with extreme crawl-to-refer ratios, post-quantum encryption securing about half of human web traffic, and Go-based API clients exceeding 20% adoption.

2025 was defined by reasoning-driven models and practical agents—especially coding agents and CLI workflows—unlocking longer autonomous tasks and widespread prompt-based image editing, while raising new safety risks around YOLO usage, AI browsers, and prompt injection. Meanwhile, Chinese open-weight models surged, OpenAI's lead narrowed as Gemini advanced, cloud models pulled ahead of local ones, AI “slop” went mainstream, and data centers drew increasing backlash.

MongoBleed (CVE-2025-14847), a critical vulnerability in MongoDB's zlib1 message compression path, has allowed unauthenticated attackers to read arbitrary heap memory, including sensitive data, across most versions since 2017. Though a fix has been issued for supported versions, over 213,000 internet-exposed MongoDB databases remain vulnerable to this "dead-easy" exploit.

Aurea Imaging, a Dutch agricultural tech startup, addressed the challenge of managing and remotely updating a global fleet of NVIDIA Jetson-powered remote sensing devices by adopting a cloud-native approach, including K3s and the CNCF Kairos project. This enabled atomic, image-based OS upgrades, eliminating inconsistent "snowflake" devices and significantly improving operational efficiency.

This post explains how to observe and scale MLOps infrastructure on Amazon EKS using Prometheus, Grafana, and Kubernetes autoscaling, with detailed guidance on monitoring GPUs, AWS accelerators, ML-specific metrics, and integrating open source and third-party observability tools.

This post explains Terraform parallelism, how concurrent resource operations affect provisioning speed, and how to configure and manage parallelism within Terraform and external systems, along with best practices to optimize infrastructure deployment time.

Zendesk engineers reduced Datadog observability costs by auditing metrics, traces, and logs, adopting single-span tracing, targeted sampling, and log deduplication, flattening spend while preserving visibility, performance insights, and engineering workflows.

This post details caching and storage strategies for AI and ML workloads on Amazon EKS. It covers container image caching, data loading, checkpointing, and storage services like Amazon S3, S3 Express One Zone, and FSx for Lustre to optimize performance and cost.

Kubernetes v1.35 introduced significant workload-aware scheduling improvements, including the new Workload API (`scheduling.k8s.io/v1alpha1`) for defining multi-Pod application requirements and an initial implementation of gang scheduling for all-or-nothing placement.

C++ and Rust have been the fastest-growing major languages because computing is constrained by power and chips, making performance-per-watt and performance-per-transistor efficiency increasingly critical as AI drives demand.