TLDR DevOps 2026-07-01

Vercel now supports deploying any HTTP service from a Dockerfile.vercel, building the image, storing it in the project registry, and running it on Fluid compute with preview deployments, autoscaling, observability, and active CPU pricing. The feature is aimed at backends and services that do not fit Vercel's framework detection model, letting teams deploy Go, Rails, Spring Boot, Express, Laravel, ASP.NET, FastAPI, nginx-backed services, and other containerized apps as long as they listen on $PORT.

AWS Certificate Manager now supports the ACME protocol for automated TLS certificate issuance, letting organizations use standard clients like Certbot while maintaining centralized control over certificate policies and domain validation. The new managed ACME endpoint offers PKI administrators fine-grained access controls through IAM roles, domain scoping capabilities, and unified visibility across all certificates, addressing the growing challenge of shorter certificate validity periods dropping to 47 days by 2029.

Dragonfly v2.5.0 was released with direct download support for AI model repositories from Hugging Face and ModelScope using Git LFS acceleration, plus a new Kubernetes webhook injection system that automatically adds P2P capabilities to Pods without rebuilding container images. The update also introduced dfctl command-line tool for managing local tasks, enhanced rate limiting across control planes, and a blocklist feature in the Manager console to block problematic downloads.

Load tests should use production telemetry from Grafana Cloud instead of guessed VUs, deriving arrival-rate k6 scenarios, baselines, peaks, and latency thresholds from real request and traffic patterns, which can be auto-generated via Grafana Assistant for realistic testing loops.

Feature flag migrations are often delayed due to risks around logic parity, configuration drift, and cutover safety across legacy and new systems. A safe migration uses parallel systems with shadow evaluation, incremental redirection of new and short-lived flags, and monitored parity checks before staged cutover.

Herdr, a new open-source terminal multiplexer designed specifically for managing multiple AI coding agents, launched as a lightweight Rust binary that shows real-time agent status (blocked, working, or done) across persistent sessions that survive disconnects. Unlike GUI-based alternatives like Conductor or cmux, herdr runs directly in any terminal with SSH access and provides native integrations for popular AI agents including Claude, Cursor, Devin, and Copilot, while offering a Unix socket API for custom agent development.

OmniRoute, an open-source AI gateway, launched as a local proxy that aggregates 236 AI providers (including 50+ with free tiers) into a single endpoint, offering developers up to 2.1 billion free tokens monthly across services like Claude, GPT, and Gemini while using RTK and Caveman compression to reduce token usage by 15-95%. The MIT-licensed tool runs locally on users' machines, features automatic fallback routing when providers fail or hit limits, and integrates with popular coding tools like Cursor, Copilot, and Claude Code without requiring code changes.

Discord's engineering team developed a custom CPU profiling system to track hosting costs across its 1,700+ API endpoints by extending their Python profiler to tag code execution time by feature, then joining that data with cloud billing information to accurately allocate infrastructure spending. The system solved the challenge of cost attribution in their unified codebase that runs across hundreds of Kubernetes deployments, giving product teams real-time visibility into how code changes affect operational costs without requiring infrastructure changes.

Argo CD 3.5 introduces a first-class ApplicationSet UI with list views, health monitoring, resource tree visualization, owner relationships, and browser-based previews that show generated applications and configuration diffs before changes are applied. The new read-only interface makes ApplicationSets easier to manage without relying on kubectl or App-of-Apps workarounds.

CNCF's Kepler project completed a major rewrite that eliminates its reliance on eBPF in favor of standard /proc and /sys access, improving accuracy in tracking power consumption for Kubernetes workloads while requiring significantly lower privileges for deployment. The changes reduced attribution gaps to nearly zero watts in testing and boosted code coverage to 90%, addressing previous issues with missing short-lived processes and inaccurate energy footprints that had created bottlenecks for power estimation models.

Amazon CloudWatch Logs now provides managed syslog ingestion, allowing network devices and Linux servers to send RFC 5424, RFC 3164, and Cisco syslog over TCP, TCP+TLS, or UDP to a VPC endpoint without agents.

Amazon CloudWatch now offers OTel Container Insights for Amazon EKS, providing 30-second granularity infrastructure metrics via OpenTelemetry receivers like cAdvisor, Kube State Metrics, and NVIDIA DCGM.