Explain Aurora-style internals: WAL, MVCC, replication, recovery

Database Internals Deep Dive

You are interviewing for a cloud database engine team (Aurora-like). Answer conceptually and with trade-offs.

Questions

1. Architecture : How does a cloud-native database architecture (separating compute from distributed storage) differ from a traditional single-node database using local disks? What are the benefits and costs?
2. Replication model : Compare replication approaches:
   • physical vs logical replication
   • statement-based vs row-based vs log-based
   • synchronous vs asynchronous
   • quorum-based durability in distributed storage
   When would you choose each, and what failure modes do they address?
3. WAL / redo log : What is the role of WAL (write-ahead logging) in durability and crash recovery? How do redo/undo relate to WAL? What must be forced to durable storage on commit?
4. Crash recovery : Walk through a typical recovery procedure after a crash (phases, checkpoints, replay). What changes when storage is distributed?
5. MVCC : Explain MVCC fundamentals:
   • how reads avoid blocking writes
   • snapshot visibility rules (high level)
   • what metadata is stored per row/version
6. High-concurrency writes : Under heavy write load, how do lock conflicts happen and how can an engine reduce contention (e.g., latches vs locks, optimistic CC, partitioning/hotspot mitigation)?
7. Trade-offs : Discuss consistency vs latency vs throughput trade-offs in replication and commit paths (e.g., quorum write, 1-RTT vs 2-RTT commit, read-your-writes, failover behavior).