The JusDB Database Engineering Blog

Explore the trade-offs between database normalization and denormalization for optimal performance. Learn when to normalize for data integrity and when to denormalize for query speed.

Twelve schema design patterns that survive production scale: primary key strategy, normalization trade-offs, partitioning, sharding boundaries, and indexing rules.

Eliminate stale cache reads and thundering-herd stampedes by streaming MySQL binlog changes to Valkey via Debezium CDC. Full Java setup with embedded engine and XFetch.

Design multi-tenant databases — compare schema-per-tenant, row-level-security, and database-per-tenant approaches

Design multi-region database topologies for global latency and DR. Compares active-passive, Aurora Global Database replication, and application-level read routing patterns.

FoundationDB powers Apple iCloud and Snowflake's metadata layer with ACID transactions across distributed nodes. Learn how its ordered key-value core, layer architecture, and simulation testing deliver unmatched reliability.

Redis Streams and Kafka both handle millions of events per second, but they solve different problems. Compare latency, retention, delivery guarantees, and operational overhead to choose the right architecture.

Compare database sharding strategies: hash sharding for even distribution, range sharding for locality, and directory-based for flexibility. When to shard and alternatives.

Compare three PostgreSQL multi-tenant patterns: shared schema with RLS, schema-per-tenant, and database-per-tenant. Includes hash partitioning for shared schema at scale.

TiDB 8.0 is the most production-ready version of the distributed SQL database yet. With improved HTAP, stronger MySQL compatibility, and Resource Groups, it's compelling for teams outgrowing single-node MySQL.

OpenSearch (Apache 2.0) and Elasticsearch (EL2/SSPL) have diverged. Compare licensing, security features, AI/ML capabilities, and migration complexity for 2026.

TiDB uses Raft consensus protocol through TiKV to replicate every region across three or more nodes. Here's how leader election, log replication, and region management work in practice.