Tag
Netflix 介绍了其在超大规模基础设施上构建端到端知识图谱（本体驱动的可观测性）的实践，通过统一 MELT（指标、事件、日志、追踪）数据层并显式记录系统实体间关系，将典型故障根因排查时间从约 4 小时显著缩短。
This paper formalizes the problem of ordering filters in sequential filtering pipelines under independent cost and selectivity models, proving that ordering by increasing ratio of cost to rejection probability is optimal. Monte Carlo simulations demonstrate that this ordering dominates common heuristics both in expectation and across the full distribution of outcomes.