Loading blogs...
Записи с тегом Observability.
One degrading optic can raise thousands of alerts and stall a thousand GPUs. What NVIDIA Minerva does, what it is worth in GPU hours, goodput, MTBI, MTTR and tokens, which of its three consumption models fits, worked examples for a Vera Rubin NVL72 neocloud, a university Blackwell cluster and an enterprise with Grafana and ServiceNow, and how it relates to UFM.
One flapping InfiniBand link can stall a 256-GPU job. What NVIDIA Unified Fabric Manager does, when UFM Telemetry, Enterprise or Cyber-AI fits, worked examples for a 4-node DGX B300 cluster, a 32-node SuperPOD-style cluster and a university Slurm service, and what fabric visibility is worth in GPU hours.
AI agents fail quietly: a confident answer from the wrong document, a tool called with the wrong argument. Learn how to trace every model, retrieval and tool step with OpenTelemetry, attach quality scores, and keep the records auditors ask for.
Unit tests prove a function is correct, not that a system survives a slow database or a lost zone. A practical guide to cold-start integration tests, fault injection with Toxiproxy, circuit breakers, recovery drills and safely automated mitigation.
A practical playbook for running AI agents in production: classify every tool by side effect, gate mutations behind a human, set hard stop conditions, trace each step with OpenTelemetry and evaluate continuously.
Why fragmented monitoring creates blind spots, duplicate spend, and slow incidents — and how to make centralised observability a non-negotiable company standard.
Workstation по выявлению узких мест агентов LLM с возможностью наблюдения: OpenTelemetry, Prometheus/Grafana/Thanos, Langfuse — с четкими плюсами и минусами, рычагами затрат и показателями использования, которые защищают прибыль.
Удобный для поддержки, готовый к SRE обзор последовательных обновлений сервера Couchbase под управлением автономного оператора Couchbase (CAO) с использованием spec.paused для шлюзования и перебалансировки подкачки по одному узлу за раз, чтобы вы могли стабилизировать, проверять работоспособность и оставлять открытыми варианты отката.