Rootconf Mini 2024

Abstract

Operating Cassandra at scale for mission-critical workloads presents unique challenges. This talk explores the strategies we use at Quizizz to maintain a resilient Cassandra cluster for real-time gameplay, ensuring high availability and low latency. We discuss architectural considerations for building a scalable, fault-tolerant infrastructure, with a focus on data modeling, performance monitoring, and optimization.

At Quizizz, we ingest a billion question responses every month with each response then being used for real-time aggregation and getting a birds eye view for everyone in a school.

We cover data modeling techniques for balanced partitioning, compaction strategies for performance improvement, monitoring practices to identify bottlenecks, and performance tuning for real-time demands. Additionally, we share lessons learned on integrating Cassandra into a distributed architecture with resilience and reliability as the focus.

Key Takeways

• Strategies for Cassandra data modeling, partitioning, and compaction to ensure consistent performance
• Techniques for mitigating split-brain scenarios and conducting repairs to ensure data consistency with Cassandra Reaper.
• Approaches to prevent tombstone buildup and optimize bulk read performance.
• Methods for leveraging Change Data Capture (CDC) for analytics in a large-scale environment.

Audience

• Engineers and Architects developing high-throughput, low-latency applications with high availability requirements.
• Site Reliability Engineers (SREs) and Platform Engineers managing Cassandra or similar NoSQL databases in production.