Database & Storage Deep Dive
B-Tree vs LSM-Tree, replication strategies, consistency models, CAP theorem practical guide
Technical Reference & Key Concepts
Indexing Strategies
- B-Tree: Default in most databases. Good for equality + range queries. Tradeoff: Write amplification.
- LSM-Tree: Used in Cassandra, RocksDB. Sequential writes, great for write-heavy workloads. Tradeoff: Compaction overhead.
- Hash Index: O(1) point lookups. No range queries. Good for key-value workloads.
- Inverted Index: Maps terms->documents. For full-text search (Elasticsearch).
Replication
- Single-leader: One primary writes, replicas read. Simple but failover is manual.
- Multi-leader: Multiple primaries. Good for multi-datacenter. Tradeoff: Conflict resolution.
- Leaderless: Quorum-based (Dynamo, Cassandra). High availability. Tradeoff: Stale reads.
Consistency Models
- Linearizability: Strongest. Reads return most recent write. Hard to achieve in distributed systems.
- Serializability: Transactions appear to run sequentially. Not the same as linearizability.
- Snapshot Isolation: Each transaction sees a consistent snapshot. Prevents dirty reads.