Backup & Recovery

Introduction

Untested backups are Schrödinger's disaster recovery — RDB snapshots, AOF files, and managed cloud snapshots each need quarterly restore drills verifying key counts and application smoke tests.

Best practice: BGSAVE on replica, encrypt, upload to object storage with versioning, document RTO/RPO. Point-in-time for Redis alone is approximate — combine with application-level audit if needed.

Restore to isolated instance first — never overwrite prod without validation.

Understanding the topic

Key concepts

• RDB — point-in-time binary snapshot.
• AOF — replay write log on restore.
• Managed automatic snapshots (ElastiCache).
• Off-site immutable copy (S3 Object Lock).
• redis-check-rdb / redis-check-aof validation.
• RTO/RPO documented per service tier.

Step-by-step explanation

1. Schedule BGSAVE or rely on cloud snapshot.
2. Copy artifact to durable storage.
3. Validate with check tools.
4. Restore to temp Redis instance.
5. Compare DBSIZE, spot keys, run app tests.
6. Cutover or discard.

Syntax reference

Common commands

• Test restore ≠ listing backup exists.
• Encrypt at rest in S3.
• Scrub PII when cloning prod to staging.

redis-cli BGSAVE
redis-check-rdb dump.rdb
# Restore test
redis-server --dbfilename dump.rdb --dir /tmp/restore-test
redis-cli -p 6379 DBSIZE

Informative example

Automated backup script outline:

#!/bin/bash
redis-cli -h replica.internal BGSAVE
sleep 30
scp replica:/var/lib/redis/dump.rdb ./backup-$(date +%F).rdb
redis-check-rdb backup-$(date +%F).rdb
aws s3 cp backup-$(date +%F).rdb s3://dr/redis/ --sse AES256

Run from replica. Alert if redis-check-rdb fails. Keep 30-day retention minimum for compliance tiers.

Real-world use

Real-world use cases

• Disaster recovery regional outage.
• Accidental FLUSHDB recovery from last RDB.
• Staging refresh from sanitized prod.
• Compliance audit trail of backups.
• Pre-upgrade snapshot rollback point.

Best practices

• Quarterly restore drill with checklist.
• Backup from replica not master.
• Encrypt and version off-site copies.
• Separate AWS account for backup bucket.
• Document who approves prod restore.
• Monitor backup job success alerts.

Common mistakes

• Backups never restored in test.
• Only on-primary BGSAVE under load.
• No encryption on backup bucket.
• Restoring FLUSHALL over prod without isolation.

Advanced interview questions

Summary

Automate RDB/snapshot to off-site storage.