Kit desain sistem observabilitas dan data

Workload profile, store categories (relational, KV, document, wide-column, search, time-series, vector), weighted matrix, derived stores

Replication topology, sync vs async, N/W/R quorums, consistency model per operation, conflicts, failover and RPO, PACELC

Metrics and alerting design: OTLP and remote write ingestion, Gorilla-style compression, cardinality limits, rollups, alert evaluation, meta-monitoring

45-minute path, W3C trace context and baggage, async propagation, head, tail and hybrid sampling, span metrics, log correlation, retention tiers, failure modes

45-minute path, click dedupe, event-time windows, exactly-once counting, exact and Count-Min top-K, batch reconciliation for billing, failure modes, 10x row

Stream processing engine: event time, watermarks, windows and triggers, late data, keyed state, barrier checkpoints, exactly-once sinks, backpressure, rescaling

CDC pipeline: log-based capture, watermark chunked snapshot, per-key order, schema evolution, outbox, search, cache and warehouse sinks, replay

S3-like blob store design: consistent metadata, packed volumes, replicas then RS(9,6) erasure coding across AZs, durability maths, repair, deletes

GFS/HDFS-style file system: metadata and chunk servers, chunk sizing, leases and write pipeline, record append consistency, re-replication, erasure coding

Vector search service: HNSW vs IVF-PQ sizing, re-ranking, recall vs latency tuning, filtered search by selectivity, inserts and deletes, sharding, failure modes

ML feature store: registry, batch and streaming pipelines, offline and online stores, point-in-time joins with TTL, training-serving skew, monitoring

Telemetry goals, OTel semantic conventions 1.44.0 baseline with stability per area, logs, metrics, traces, optional Profiles, dashboards, SLOs, alerts, privacy, retention, cost