Handling Correlation ID Changes in Event Sourcing When an Entity Switches Context

Question

I'm working on an event-sourced application that crawls sports betting games from different bookmakers. I have two primary aggregates in my system:

• Game: Represents a sports betting event for a specific bookmaker.
• Combined Game Bucket: A bucket that groups games with similar names across different bookmakers, based on a 30-minute interval (start time). Each bucket handles multiple combined games.

Problem

I want to ensure that all events within a stream share the same correlation ID. Ideally:

• Games within the same stream should share a correlation ID.
• When a game is combined into a Combined Game Bucket, it should inherit the correlation ID of the combined game, ensuring that all games within the same combined game share the same correlation ID.

This works well until a game's start time changes significantly (more than 30 minutes). In this case, the game is removed from its original combined game and placed into a new combined game bucket. The issue here is that the correlation ID from the original game stream becomes outdated, as it still points to the old combined game. Since correlation IDs shouldn’t be changed for previous events in a stream, there’s no straightforward way to update the correlation ID for future events related to this game.

Proposed Solution

One idea I had was to create a new stream when the game’s start time changes significantly. The old stream would emit a GameMovedToNewStream event, which points to the new stream. However, this results in the game’s event history being split across multiple streams.

Questions

• Is fragmenting a game's history across multiple streams considered a bad practice in event sourcing?
• Should I rethink the assumption that games should share the same correlation ID with their combined game?
• Are there better approaches for maintaining correlation IDs in this scenario while still preserving a consistent event history for games?

I'm looking for advice on whether my current approach aligns with best practices in event sourcing or if there's a more elegant way to handle this situation.

Answer 1

After reevaluating how I'm handling CombinedGame and BookmakerGame events within my existing projections, I've decided to change the approach.

Instead of having the bookmaker games share a correlation ID with the combined game, I instead duplicate the "game paired/combined" event in the BookmakerGame stream itself, within the same transaction as the original CombinedGame event. This removes the dependency on the combined game stream entirely.

So when a game is paired, two events are saved in a single transaction:

• CombinedGame::GameAdded
• BookmakerGame::GameAddedToCombinedGame