RDB Snapshots

Introduction

RDB persistence writes a compact binary point-in-time snapshot (dump.rdb) — fast restarts and efficient backups at the cost of losing writes since last snapshot. BGSAVE forks a child process to write disk while master serves commands.

save directives in redis.conf trigger snapshots after N changes in M seconds. fork() on large memory instances causes latency spikes via copy-on-write pressure — schedule BGSAVE off-peak.

Combine RDB for backup cadence with AOF for finer recovery point objective.

Understanding the topic

Key concepts

• SAVE — blocking snapshot; avoid production.
• BGSAVE — background fork child writes RDB.
• dump.rdb — single file restore point.
• save 900 1 style rules in config.
• LASTSAVE — Unix time of last success.
• Partial sync possible after restart with replication.

flowchart LR
  Writes --> Memory
  Memory -->|fork| RDB
  Memory -->|append| AOF

Step-by-step explanation

1. Trigger via save rule, BGSAVE command, or shutdown.
2. fork() creates child with shared memory pages.
3. Child iterates keyspace writing compressed RDB.
4. COW — parent writes duplicate pages on mutation.
5. Complete RDB atomically renamed into place.

Syntax reference

Common commands

• BGSAVE not SAVE in prod.
• Monitor latest_fork_usec in INFO.
• Copy RDB to S3 after BGSAVE for DR.

BGSAVE
LASTSAVE
CONFIG GET save
# redis.conf: save 900 1  save 300 10  save 60 10000

Informative example

Manual backup before migration — verify RDB file:

redis-cli BGSAVE
# wait for OK
redis-cli LASTSAVE
ls -lh /var/lib/redis/dump.rdb
aws s3 cp dump.rdb s3://backups/redis/$(date +%F).rdb

Stop writes or use REPLICAOF snapshot on replica for cold backup without master fork load.

Real-world use

Real-world use cases

• Nightly backup to object storage.
• Fast cold start after planned restart.
• Disaster recovery baseline.
• Clone staging from prod RDB (scrub PII).
• Compliance point-in-time archive.

Best practices

• Run BGSAVE on replica when possible.
• Leave RAM headroom for COW during fork.
• Automate S3 upload with encryption.
• Test restore quarterly — redis-check-rdb.
• Pair with AOF for tighter RPO.
• Alert on failed BGSAVE in logs.

Common mistakes

• SAVE on production master.
• No off-site copy — datacenter loss = data loss.
• Assuming RDB captures last second of writes.
• Fork failure from overcommit disabled.

Advanced interview questions

Summary

RDB = point-in-time binary snapshot.