Event Sourcing in production — Borjan Tchakaloff

Event sourcing captures state changes as a sequence of events rather than storing current state, providing a complete history of changes

Aggregates are the smallest addressable units that enforce business rules and invariants - all changes must go through them

Separate read and write models (CQRS pattern) - write models enforce business rules while read models/projections optimize for querying

When handling projections:

• Create new projectors rather than modifying existing ones
• Replay full event history to rebuild projections
• Consider eventual consistency implications

For domain evolution challenges:

• Fail loudly when encountering unknown events
• Use upcasting for handling legacy events
• Consider backwards compatibility when adding new attributes
• Carefully handle unique constraints and data migrations

Production considerations:

• Use hybrid approach with single-threaded workers for concurrent requests
• Implement atomic saves for multiple aggregates
• Plan for projection rebuilds during schema changes
• Handle duplicate detection based on business requirements

Read models can be implemented multiple ways:

• Django models for simple cases
• Dedicated structures for complex queries
• Statistical aggregations
• Temporal tracking approaches

Technical implementation:

• Events should be immutable truth source
• Applications coordinate command handling
• Process applications handle cross-domain coordination
• Event store manages persistence

Keep projections and read models separate from core domain logic

• Projections can be rebuilt from events
• Different views can serve different query needs
• Consider eventual consistency tradeoffs