1️⃣ Durable Databases (ACID-compliant)

If you are using a database that supports ACID transactions:

• A = Atomicity → Either the whole transaction succeeds or none of it does
• C = Consistency → Database moves from one valid state to another
• I = Isolation → Transactions don’t interfere
• D = Durability → Once a transaction commits, it survives crashes

Recovery Mechanisms:

1. Write-Ahead Logging (WAL)

   • Example: PostgreSQL, MySQL InnoDB
   • All changes are first written to a log on disk before applying to the data file
   • On crash → replay the log to recover committed transactions

2. Checkpoints

   • Periodically flush data to disk
   • Combined with WAL → can recover to last committed state efficiently

Outcome:

• Records that were committed before crash → recovered automatically
• Uncommitted records → rolled back

---

2️⃣ Replication & Redundancy

• Multi-AZ / Multi-Region replication ensures another copy exists if one system fails

• Example:

  • Redis Streams, Kafka, or databases like Cassandra/DynamoDB
• If primary fails → replicas have the data → system recovers without losing records

---

3️⃣ Backups

• Periodic database backups allow recovery of old records if corruption occurs

• Techniques:

  • Full backup
  • Incremental / differential backup
• Restore → recover lost data up to last backup

---

4️⃣ Crash Recovery in In-Memory Systems

• In-memory databases (e.g., Redis) require persistence configuration:

  1. RDB snapshots → periodic saves to disk
  2. AOF (Append-Only File) → logs every write operation

• On restart → replay AOF or load snapshot to recover records

---

5️⃣ Application-Level Recovery

• Some systems implement idempotency and retries:

  • If a failure occurs mid-operation → application retries safely without duplicating data

• Helps recover transient failures in distributed systems

---

✅ Key Takeaways

1. Committed data in durable systems can be recovered automatically after a crash.
2. Uncommitted data is usually lost unless special techniques are used.
3. Replication and backups improve resilience and reduce data loss.

4. Crash recovery mechanisms depend on:

   • Database type (ACID vs NoSQL)
   • Logging (WAL, AOF)
   • Replication (single vs multi-region)

---

💡 Analogy:

• Committed record = money deposited in a bank → survives a bank system crash
• Uncommitted record = money on the counter before deposit → lost if system fails

---