Two people, one seat: Designing for real-world concurrency (Azure Cosmos DB Conf)
Patty Chow uses a “two people, one seat” scenario to highlight why concurrency bugs show up in real systems, then points to Azure Cosmos DB patterns—like multi-region writes, conflict resolution, change feed, event sourcing, and replay/replayable architectures—to design for issues you can’t easily simulate.
Overview
The short frames a common distributed-systems failure mode: two actors perform valid actions at the same time (both bought a ticket), but the system can’t coordinate fast enough to prevent an invalid outcome (only one seat exists).
It highlights Azure Cosmos DB-focused approaches teams use to build resilient, globally distributed applications:
- Multi-region writes to support low-latency writes across geographies.
- Conflict resolution strategies to handle concurrent updates.
- Change feed for reacting to data changes and building downstream processing.
- Event sourcing to model state changes as an append-only stream of events.
- Replay / “time travel” replay techniques to reproduce and reason about concurrency issues that are hard to simulate in test environments.
Related playlist: https://aka.ms/CosmosConf26Playlist