Event Streams and Durable Workflows

Adding a broker between services does not make an asynchronous system reliable. Production still needs answers about order, acknowledgment, duplicate delivery, replay, side effects, schema changes, poison data, and work that may remain unfinished for months.

Event Streams and Durable Workflows is a production-oriented guide to systems built around queues, durable logs, event streams, change data capture, stateful processing, and workflow histories. It helps intermediate and advanced engineers choose an asynchronous state model because its guarantees match the problem - not because event-driven architecture sounds modern.

The book follows messages and intent across partitions, consumers, databases, processors, and long-running workflows. It explains what delivery guarantees really cover, how idempotency boundaries should be drawn, why replaying old events through new code is dangerous, and how durable execution differs from ordinary background jobs.

Inside, you will learn how to:

- Distinguish queues, pub/sub, durable logs, streams, and workflow engines
- Design partitions, consumer groups, offsets, ordering, and replay
- Handle at-least-once delivery, deduplication, idempotency, and transactions
- Connect transactional outbox patterns with CDC and schema evolution
- Reason about windows, watermarks, late data, joins, and materialized state
- Model workflow commands, signals, timers, compensation, and human work
- Isolate side effects during replays, backfills, and recovery campaigns
- Operate backpressure, poison-data handling, observability, and migrations

Failure timelines, production reports, open-source system analysis, runnable labs, and incident-oriented review material expose the boundaries hidden by simplified diagrams. Special attention is given to recovery: how operators can pause, inspect, replay, promote, compensate, and prove that progress resumed safely.

Event Streams and Durable Workflows gives engineers a concrete framework for preserving intent and controlling progress when computation spans services, failures, deployments, and long periods of time.