Redis Clustering, Replication & High Availability

Redis High Availability, Replication & Cluster Sharding (16,384 Slots)

1. Master-Replica Async ReplicationRead Scalability

Master accepts all writes asynchronously and streams updates to one or more replicas via replication backlog buffer & offsets.

Core Technical Mechanics

Asynchronous replication: Master returns WRITE success immediately to client without waiting for replicas.
Partial Resynchronization (PSYNC): Uses master_repl_offset and replication_id to sync missed stream entries after brief network split.
Full Resynchronization: Master generates RDB snapshot and streams to replica when offset gap is too large.

Command & Configuration Reference

replicaof 192.168.1.100 6379
INFO replication # Check master_link_status & offset

Master-Replica Replication Mechanics

Redis uses asynchronous primary-replica replication to provide read scaling and data redundancy.

+---------------------------------------------------------------+
|  Master Node (Primary) -- Handles 100% of Writes              |
|  - Appends to Replication Backlog Buffer (repl-backlog-size)   |
+-------------------------------+-------------------------------+
                                | Asynchronous Replication Stream
                                v (PSYNC)
+-------------------------------+-------------------------------+
|  Replica 1 (Read-Only)        |  Replica 2 (Read-Only)        |
+-------------------------------+-------------------------------+

Full Synchronization vs Partial Synchronization (`PSYNC`)

Partial Synchronization (PSYNC master_repl_id offset): When a replica temporarily disconnects, it reconnects and requests a partial sync. If the missing byte range exists in the Master's in-memory Replication Backlog Buffer (repl-backlog-size, default $1\text{ MB}$ ), the master transmits only the missing delta bytes without disk I/O.
Full Synchronization: If the replica has been disconnected longer than the backlog buffer capacity, the Master executes a BGSAVE to write a full RDB snapshot to disk, streams the RDB file over the network to the replica, and the replica wipes its existing database to load the RDB file.

Sentinel Architecture (High Availability)

Redis Sentinel is an external monitoring and failover management system operating alongside Redis instances:

Subjective Down (SDOWN): A single Sentinel node loses ping response from a Master for down-after-milliseconds.
Objective Down (ODOWN): Multiple Sentinel nodes confirm SDOWN state by reaching quorum agreement (quorum 2).
Failover & Election: Sentinels vote for a Leader Sentinel via Raft-like voting. The Leader Sentinel selects the healthiest Replica (lowest replication lag, highest offset) and promotes it to Master (replicaof no one).

Redis Cluster Sharding (16,384 Hash Slots)

Redis Cluster scales writes horizontally across multiple master nodes using 16,384 Hash Slots:

$\text{Slot Index} = \text{CRC16}(\text{key}) \pmod{16384}$

MOVED Redirect: If a client sends a command to Node A for a key belonging to Node B, Node A responds with -MOVED 14328 10.0.0.5:6379. Smart clients (Lettuce, Jedis) intercept MOVED responses and update local slot mapping tables automatically.
ASK Redirect: Returned during active cluster resharding when a slot is currently migrating between nodes.

Split-Brain Prevention (`min-replicas-to-write`)

To prevent a network-partitioned Master from accepting writes that will later be overwritten upon cluster recovery, configure:

# redis.conf
min-replicas-to-write 1
min-replicas-max-lag 10

If a Master loses all active replicas or replica lag exceeds 10 seconds, it immediately rejects incoming writes with a NOREPLICAS error.

Interview Questions

Q1. How does Redis replication work and what is the role of the Replication Backlog Buffer?

Redis replication is asynchronous. All write operations execute on the Master and are streamed asynchronously to Replicas. The Master maintains an in-memory ring buffer called the Replication Backlog Buffer (repl-backlog-size). If a replica briefly disconnects, it sends a PSYNC command containing its last offset. If the missing offsets are still present in the master's backlog buffer, the master sends only the missing byte delta (Partial Sync), avoiding a heavy full RDB disk snapshot.

Q2. What is the difference between Redis Sentinel and Redis Cluster?

Redis Sentinel provides High Availability (HA) and automatic failover for single-master deployments. All nodes store the complete dataset (no sharding), and write throughput is limited to one master. Redis Cluster provides both High Availability AND Horizontal Write Scaling by partitioning data across 16,384 Hash Slots mapped to multiple master nodes, enabling cluster storage and throughput to scale linearly across nodes.

Q3. How do Hash Tags `{...}` enable multi-key operations in a sharded Redis Cluster?

Multi-key Redis commands (MGET, MSET, Lua scripts) require all target keys to reside on the same cluster node. Hash Tags force Redis to compute the CRC16 slot index strictly using the string inside the braces {...} instead of the entire key name. For example, {user:1001}:profile and {user:1001}:orders yield the identical hash slot, enabling multi-key operations across co-located keys on a single node.

Master-Replica Replication Mechanics

Full Synchronization vs Partial Synchronization (PSYNC)