Streams, metrics & aggregation

5 题

  • 广告点击聚合系统

    设计广告点击聚合系统。请覆盖High-throughput ingestion of K-M clicks/sec via a log/stream (Kafka) before aggregation, Exactly-once vs at-least-once delivery and idempotent aggregation sinks, Windowed aggregation...

    免费指南
  • Metric Monitoring and Alerting

    设计监控平台背后的指标管线——从部署在数千台主机上的 SDK 发出数值测量值开始,经过流式预聚合,写入持久的时序存储,让仪表盘能在超大规模下快速查询的整条路径。

    预览
  • Monitoring and Alerting

    设计一套监控平台的告警与可观测性层——这套系统会用规则比对进来的信号、决定要呼叫谁,并在事故期间避免值班的人被噪声淹没。

    预览
  • Top K Songs (Spotify)

    设计Top K Songs (Spotify)。请覆盖Framing top-K as a heavy-hitters/frequency-estimation problem over a play-count stream, Count-Min Sketch + heap (bounded error, memory trade-off) vs exact per-partition...

    免费指南
  • Fraud Detection

    设计一套实时的 ML 欺诈/风险评分系统,在 100 ms 内为支付交易评分。涵盖特征类型(velocity、graph、device)、流式特征计算、标签延迟与反馈回路、分数阈值与成本的权衡、模型加规则的混合架构,以及对抗性漂移。

    预览