Best of Faun — April 2026

  1. 1
    Article
    Avatar of faunFaun·7w

    Kubernetes Is Not DevOps : A Short Story

    A hiring manager shares an interview experience where a candidate knew Kubernetes commands but couldn't explain what happens internally when running kubectl apply. The story illustrates a broader industry trend: engineers learning tools without understanding the underlying systems. True DevOps expertise goes beyond tool familiarity — it requires understanding infrastructure provisioning, distributed systems, automation principles, and reliability design. Kubernetes is just one tool; the fundamentals of systems thinking and automation will outlast any specific technology.

  2. 2
    Article
    Avatar of faunFaun·4w

    Qwen3.6–35B-A3B: The Most Practical Open-Source AI Model Yet?

    Qwen3.6-35B-A3B is a Mixture-of-Experts open-source model with 35B total parameters but only ~3B active per request, making it highly efficient. It features a 262K context window (extendable to 1M with YaRN), multimodal support (text, image, video), and an Apache 2.0 license. The model is designed for agentic coding workflows, achieving top scores on SWE-bench Verified (73.4), Terminal-Bench 2.0 (51.5), and strong STEM reasoning benchmarks. Key architectural innovations include Gated DeltaNet linear attention and Grouped Query Attention (GQA). It supports a switchable thinking/non-thinking mode and a new thinking preservation feature that reuses reasoning across conversation turns. Deployment is supported via vLLM, SGLang, KTransformers, and Hugging Face.

  3. 3
    Article
    Avatar of faunFaun·7w

    Async Logging Is Not a Silver Bullet — What Actually Limits Performance

    Async logging is commonly assumed to make logging cheaper, but it only redistributes the cost rather than eliminating it. The real pipeline includes a capture step (copying or serializing data) before enqueuing, which is required for correctness due to data lifetime issues with deferred formatting. Formatting often dominates logging cost, and moving it to a background thread doesn't make it faster. A single backend thread creates a hard throughput ceiling, and queues only delay when overload becomes visible. An alternative approach — optimizing formatting on the caller thread and only offloading I/O asynchronously — avoids lifetime issues, reduces copying, and produces more predictable behavior.

See all Faun archives