Category
Technology with judgment: what actually moves the needle, no passing fads.
cAdvisor is still embedded in kubelet and covers surface metrics, but falls short for production Kubernetes. The modern minimum stack pairs it with kube-state-metrics, node-exporter, Prometheus, and Grafana as a base, eBPF for deep network and syscall visibility, and OpenTelemetry for application context.
Llama 3 is the open-model family Meta released on April 18, 2024, in 8-billion and 70-billion-parameter sizes, trained on 15 trillion tokens. The 70B beat Claude Sonnet, Mistral Medium, and GPT-3.5 in Meta's own human evaluation, and its licence allows free commercial use up to 700 million monthly active users.
Chainguard Images are minimal Docker containers from the company behind Sigstore, with zero known CVEs, Cosign-signed SBOMs and daily rebuilds on top of Wolfi, its own glibc-based distribution. They pay off over official images when strict compliance, supply chain audits or sensitive production workloads are at stake.
In 2024 sustainable data centers move beyond PUE: liquid cooling becomes standard in AI GPU racks, carbon-aware workload scheduling is already practical with tools like the Carbon Aware SDK, and waste-heat reuse has real cases in Stockholm and Helsinki. The EU-wide energy efficiency directive already requires honest metrics instead of greenwashing.
Sigstore has become the standard signing layer for OCI artefacts. GHCR is the best-integrated registry; Harbor 2.5+ and Quay offer native support; AWS ECR pushes its own KMS scheme. Verification earns its keep at three points: the cluster admission controller, the GitOps layer, and the CI/CD pipeline. The public Rekor has rate limits that force self-hosting past a certain build volume.
A model trained in PyTorch or TensorFlow, running the same way on a server, a phone, a browser tab, or an ARM gateway on the factory floor: that is what ONNX Runtime solves. It turns the ONNX format into a genuinely portable artifact, exported once, at the cost of some peak performance versus a platform-native runtime.
Deno Deploy is the serverless-edge platform from the team that built Deno: native TypeScript without a transpile step, standard Web APIs, and global deployment to ~35 regions. Cold start runs ~50ms vs Cloudflare Workers (~5ms). Best for lightweight APIs and Fresh/Astro SSR; less ideal when your project needs the full Cloudflare infrastructure stack.
Loki indexes only labels, not log content, which cuts storage costs dramatically compared to Elasticsearch. The main production risk is cardinality explosion each unique label-value combination generates a stream that inflates the index and slows queries. Separating read and write paths ensures a heavy query cannot saturate ingestion.
Fastly Compute runs WebAssembly code on its ~70 global PoPs with a 35-microsecond cold start. Supports Rust, Go, and JavaScript compiled via Javy. Competes directly with Cloudflare Workers, with advantages for compute-intensive workloads and teams already on Fastly CDN, but at a higher entry cost.
