Replication

Introduction

Replication streams writes from master to one or more replicas — read scaling and failover candidates. Replicas are async by default; a read immediately after write on replica may be stale ( eventual consistency ).

PSYNC partial resync avoids full RDB transfer on brief disconnect. Chain replication (replica of replica) adds lag — prefer star topology from master.

Never run slow commands or KEYS on replicas serving production reads without understanding lag impact.

Understanding the topic

Key concepts

• REPLICAOF host port — configure replica.
• Async replication — master doesn't wait replica ack (unless WAIT).
• INFO replication — lag, offset, role.
• READONLY on replica rejects writes.
• Replica serves reads to offload master.
• WAIT numreplicas timeout — sync wait rare.

flowchart LR
  Master -->|stream| Replica1
  Master -->|stream| Replica2

Step-by-step explanation

1. Replica connects sending REPLCONF.
2. Full sync: master BGSAVE RDB to replica.
3. Partial sync: backlog offset catch-up.
4. Master propagates command stream to replicas.
5. Replica applies commands in order on event loop.

Syntax reference

Common commands

• master_repl_offset vs slave repl offset — lag metric.
• READONLY default on replicas.
• Use WAIT only when strong need — hurts latency.

INFO replication
ROLE
REPLICAOF redis-master 6379
CONFIG SET replica-read-only yes

Informative example

Check replication lag during incident:

redis-cli INFO replication
# master_repl_offset:1000000
# slave_repl_offset:999500  ← 500 bytes behind

redis-cli -h replica.internal GET hot:key

Large lag — network, slow replica CPU, or huge write burst. Don't read-your-writes from replica without routing logic.

Real-world use

Real-world use cases

• Read scaling for cache-heavy GET traffic.
• Hot standby for manual promotion.
• Analytics SCAN on replica spare master.
• Cross-AZ read local replica.
• Backup BGSAVE from replica node.

Best practices

• Monitor repl lag alert >10MB or seconds.
• Route sticky read-after-write to master.
• Dedicated replicas for backup not prod reads.
• min-replicas-to-write on master prevent split-brain writes.
• Test promote replica drill.
• Avoid KEYS on lagging replica.

Common mistakes

• Read-your-writes from replica after POST.
• Writing to replica — fails or breaks topology.
• Ignoring lag during incident debugging.
• Too many chained replicas.

Advanced interview questions

Summary

Master streams writes to replicas async.