Building Fast Distributed Synchronous Replication at SingleStore – Webinar Recap 2 of 2

SingleStore's recent release of 7.0 features fast replication as one of its major features, offering high-throughput and synchronous replication that only slows down performance by about 10% compared to asynchronous replication. Rodrigo Gomes, a senior engineer at SingleStore, discusses how the company achieved these results and presents alternatives for replication, including considering primary and secondary nodes, serializing transactions, and using sequence numbers or IDs to order transactions. The presentation also touches on the challenges of implementing fast replication, such as avoiding conflicts between concurrent transactions and ensuring data consistency. SingleStore's approach uses pessimistic concurrency control, reserving disk space for each transaction, and managing metadata separately. The company is working to improve isolation levels in future releases, aiming for higher performance and reliability.