How Databases Guarantee Durability
Medium Priority9 min readUpdated July 4, 2026
AI Mock Interview
Practice this topic in a realistic system design interview
Durability means that once a database says a transaction is committed, that transaction should not disappear after a failure.
That sounds simple, but real systems fail in many different ways. The database process can crash. The operating system can restart before changed pages are written to disk. A disk can fail. A replica can fall behind. Someone can delete the wrong rows.
No single feature protects against all of those failures.
Databases build durability in layers:
- Write-ahead logging protects committed data after a local crash.
- Flushing makes sure important log records actually reach durable storage.
- Checkpointing makes crash recovery faster.
- Replication protects against losing a machine or disk.
- Backups and point-in-time recovery protect against bad deletes, corruption, bad deploys, and late-discovered mistakes.
Durability is not one switch. It is a stack of safeguards.
This chapter explains how those safeguards work together.