What does the Sentinel system consist of?

The Redis Sentinel system consists of three main components that work together, providing monitoring, failover, and configuration management for Redis.

#1. Sentinels (the watchers)

These are separate processes (Sentinel nodes) that:

• watch the state of the master and all its replicas,
• exchange information with each other,
• jointly decide when there's been a failure,
• trigger a failover when needed.

Main role: to be the "watchers" over the Redis instances.

A minimum of 3 Sentinels is needed for fault tolerance (a quorum).

#2. The Redis master and replicas

These are the Redis servers themselves, which Sentinel monitors.

• The master is the primary source of data.
• Replicas are copies that receive data from the master. Sentinel tracks their availability and the connections between them.

Main role: to hold the data Sentinel is watching over.

#3. Clients (applications using Redis)

Sentinel doesn't just manage Redis, it also serves as the entry point for clients:

• clients connect not directly to the Redis master, but through Sentinel,
• Sentinel tells them who the current master is,
• during a failover, clients are automatically redirected to the new master.

Main role: to get up-to-date information about the master.

#Sentinel's additional internal mechanisms

• Monitoring: regular PING requests to every Redis node.
• Voting (a quorum): several Sentinels agreeing on a failure.
• Failover: electing a new replica and reassigning roles.
• Configuration: forwarding notifications to clients (+switch-master, +failover-end, etc.) over Pub/Sub.

#Summary

The Sentinel system = (several Sentinel processes) + a set of Redis servers (master + replicas) + clients that know how to listen to Sentinels.

It forms a self-healing loop: Sentinel watches → detects a failure → picks a new role → notifies the clients.