TLDR DevOps 2026-11-16
kagent 1.0 π, Cloudflare Spend Management π°, TPU Sync β‘οΈ
Together With
π Users spike. Your stress level doesn't have to. (Sponsor)
Traffic surges are supposed to be good news. But when your infrastructure can't keep up, that "we're going viral π" moment can easily turn into βeverything is down and needs fixing right now.β
π Ditch the stress-coding with Microsoft Azure.
Azure autoscales and optimizes performance behind the scenes, so you can roll out features without worrying about downtime or slowdowns.
π With more global regions than any other cloud provider, Azure gives you the headroom to grow β without the growing pains.
Take the stress out of scaling with Azure π
Introducing Spend Management: Spending limits and alerts for Cloudflare (14 minute read)
Cloudflare's new Spend Management feature lets pay-as-you-go customers set spending thresholds and pause activity when limits are reached to avoid unexpected bills. The beta supports Workers and Durable Objects and provides near real-time cost monitoring in the dashboard, CLI, and API.
Announcing kagent 1.0 GA (22 minute read)
kagent 1.0 is an enterprise-ready distribution of Agent Substrate for running AI agents on Kubernetes. The system addresses pod-per-agent limitations by scheduling agent workloads as "actors" on warm worker pools, automatically suspending idle actors to free up compute and restoring them when new requests arrive. Agent Substrate is now a CNCF project.
Throwing Out The Phonebook (10 minute read)
Fly.io redesigned its 6PN private network, replacing a synchronized "phonebook" of hundreds of thousands of routing entries on every host with a lazy lookup system. The new design stores only 100 to 200 records per host, learning routes on demand when machines initiate connections. This approach uses VXLAN and standard kernel components instead of complex eBPF logic, which simplifies debugging and improves reliability.
Gang scheduling and topology-aware placement for AI training in Kubernetes (18 minute read)
Distributed AI training workloads require pods to land on hardware with the right inter-GPU bandwidth and to be scheduled simultaneously. Kubernetes' built-in scheduler cannot meet these requirements as it schedules pods individually. Topology-aware scheduling and gang scheduling, implemented by Kueue and the Coscheduling plugin, solve these placement and coordination problems.
π¨βπ»
Resources & Tools
Put every SaaS tenant behind its own VM boundary (Sponsor)
ISVs often give each customer a dedicated environment just to keep tenants apart, which wastes capacity.
AWS Nitro Sandboxes lets Amazon EKS and Amazon ECS bin-pack multiple tenants onto shared nodes, each in a hardware-isolated sandbox with its own kernel. For sensitive tenants, confidential mode adds attestation that proves data stays private from operators and AWS. See how it works
here.
Airgorah (GitHub Repo)
Airgorah is a WiFi security auditing tool that captures traffic, performs deauthentication attacks, and cracks access point passwords. It is written in Rust with a GTK4 interface and runs on Linux. The tool requires a wireless network card that supports monitor mode and packet injection.
Uptime Kuma (GitHub Repo)
Uptime Kuma is a self-hosted monitoring tool that tracks uptime for HTTP(s), TCP, DNS records, and Docker containers. It features a reactive UI, supports over 90 notification services including Telegram and Slack, and lets users create multiple status pages.
Get our free daily newsletter with curated tools π», trends π, and insights π‘, for DevOps Engineers π¨βπ»
Join 370,000 readers for
one daily email