← CatZep/Engineering Guides/System Design
Counters, top-K, time series, and logging pipelines.
Learn the reason behind each component
5 independent guides. Each includes local concept explanations, a correctness contract, a worked failure trace, and a runnable Python learning model.
Read in this order: requirements → concepts → data flow → concurrency and failure behavior → sizing → worked example. Explain what can fail before choosing a technology.
5 guides
Count impressions and clicks from a replayable event stream, while distinguishing provisional dashboard numbers from auditable billing.
Deduplication · Bloom filters · Offsets and checkpoints · Event time and watermarks · Rollups and weighted aggregation · Transactional outboxEstimate distinct active users cheaply and understand why merging sketches is different from adding daily counts.
HyperLogLog · Cardinality · Event time and watermarks · Offsets and checkpoints · DeduplicationMove structured events from applications to independent searchable, analytical, and archival sinks with a bounded loss/replay policy.
Pub/sub and durable queues · Offsets and checkpoints · Write-ahead log and fsync · Backpressure and load shedding · Deduplication · Transactional outboxStore timestamped metric series while preserving the meaning of counters, gauges, and aggregated distributions.
Cardinality · Rollups and weighted aggregation · Write-ahead log and fsync · Inverted indexes and columnar storage · Event time and watermarks · Offsets and checkpointsFind frequent items while separating approximate frequency error from missing candidate identities.
Count-Min Sketch · Top-k candidate discovery · Time windows · Offsets and checkpoints · CardinalityNo guide matches. Try a broader concept such as replication or retries.
Download the Python concept models · Python 3.10+, no packages required.
Student clarity and correctness review · 19 October 2026