Order Event-stream Topology

Named events, ordered topics, processors, consumer groups, state, replay, and dead letters

Order Event-stream Topology Named events, ordered topics, processors, consumer groups, state, replay, and dead letters 01 / Producers 02 / Transit 03 / Processors 04 / State + recovery 05 / Consumers Checkout API · order producer · 01 / Producers · team commerce Checkout API order producer team commerce Billing API · payment producer · 01 / Producers · team money Billing API payment producer team money orders.v1 · 12 partitions · 02 / Transit · key: order_id orders.v1 12 partitions key: order_id payments.v2 · 8 partitions · 02 / Transit · key: order_id payments.v2 8 partitions key: order_id Order Validate · group fulfillment · 03 / Processors · ordered Order Validate group fulfillment ordered Payment Enrich · group analytics · 03 / Processors · at-least-once Payment Enrich group analytics at-least-once Order State · materialized view · 04 / State + recovery · idempotent Order State materialized view idempotent events.dlq · poison events · 04 / State + recovery · 7-day retention events.dlq poison events 7-day retention Fulfillment · shipping workflow · 05 / Consumers · consumer Fulfillment shipping workflow consumer Analytics · streaming facts · 05 / Consumers · consumer Analytics streaming facts consumer Replay Tool · approved batch · 05 / Consumers · operator gate Replay Tool approved batch operator gate On-call · DLQ owner · 05 / Consumers · SRE On-call DLQ owner SRE OrderPlaced schema v1 PaymentCaptured schema v2 ordered orders consumer group payment facts at-least-once valid order idempotent enriched payment idempotent ready orders read model order facts non-PII invalid event dead letter poison event dead letter failure sample restricted approved replay audited batch Legend primary data policy / PII async batch data store

Transit Contract

  • • Every event and topic is named
  • • Partition keys preserve per-order ordering
  • • Consumer groups expose processing ownership

State + Delivery

  • • Processors write an idempotent materialized view
  • • Fulfillment and analytics consume distinct assets
  • • At-least-once delivery never implies duplicate business effects

Failure Ownership

  • • Poison events land in a retained dead-letter topic
  • • On-call inspects samples before replay
  • • Replay is gated, batched, and auditable

Data-flow diagram • Built with Archify • Hover to trace • R route • Click to focus • +/− zoom • M radar • [/] views • P play story • T theme • E export