How Uber Uses Cassandra to Power Core Applications at Scale?
summary
Uber runs Cassandra across thousands of nodes handling millions of QPS and petabytes of data, using an internal fork with production changes for service discovery, logging, backups and cross-region failover. Key engineered features include automatic in-node repair (compaction-like, global sequencing of repairs), a real-time anti-pattern detector that logs keyspace/table/partition-level violations, automated fleet operations for thousands of node replacements and bad-host evacuation, bulk upload via the open-source Marmar pipeline, and ongoing work on rate limiting, context propagation and upgrades to Cassandra 4.x.
AI-written summary · about
chapters (15)
- 0:00Uber’s Cassandra use
- 1:01Cassandra at Uber scale
- 3:31High-level architecture
- 5:15Automated fleet management
- 7:41Bulk data upload for ML
- 8:23Cassandra features at scale
- 9:07Cassandra repair mechanism
- 11:36Anti-pattern detection
- 15:07Bug fixes at scale
- 17:58Immutable Cassandra workloads
- 20:05Cassandra operations challenges
- 22:37Rate limiter development
- 23:39Context propagation in queries
- 24:24Future plans and upgrades
- 24:52Q&A
related talks
29mCassandra - an introduction
46mCassandra Serving Netflix @ Scale
18mKeynote: Data = AI: Leading the Future of Gen AI with Apache Cassandra
35mMission Critical Data, and Evaluating the Community in OSS Tech
35mPlatformizing OpenSearch at Uber
46mApache Cassandra: 8 years in production
41mLearnings, patterns and Uber’s metrics platform M3...
44mScaling OpenSearch at Uber: Operating Resilient, Fault-Tolerant, High-Performance Clus...