Clustering

Clustering in ALTCHA Sentinel allows you to deploy multiple instances that share data across nodes, enabling high availability for mission-critical applications.

Core Principles

• Kubernetes is the recommended environment for multi-instance deployments.
• All nodes connect directly to a Redis cluster. Deploying a highly available Redis setup is strongly recommended.
• Worker nodes are stateless and do not require persistent storage.

Clustering Modes

Mode	High Availability	Performance
Worker-Pool	No	Yes
Distributed	Yes	No
Hybrid	Yes	Yes

Worker-Pool Clustering

Recommended size: 3 or more nodes

In Worker-Pool mode, a single master node manages the database, while stateless worker nodes handle traffic. Workers fetch data directly from the master. While all nodes are traffic-capable, external traffic should be routed through a load balancer to the workers.

Characteristics

• Consistency: Maintained, as all data comes from the master.
• Availability: If the master node fails, workers can serve stale (cached) data and sync once the master recovers.
• Partition Tolerance: Workers continue operating during network partitions using cached data.
• Data Loss Risk: High — all data resides on the master. A failure can result in permanent loss.

Worker-Pool Deployment

Distributed Clustering

Recommended size: 5 to 11 nodes

Distributed Clustering provides fault tolerance and strong consistency using the Raft consensus algorithm via RQLite, a proven open-source solution.

All changes require consensus, which limits write throughput to around 200 operations per second. However, this typically does not impact Sentinel’s performance due to in-memory caching on each node. Request logging, which performs frequent writes, should be disabled to avoid bottlenecks.

Characteristics

• Consistency: Guaranteed by Raft consensus.
• Availability: Operates as long as a majority of nodes are functional.
• Partition Tolerance: Survives network partitions if a majority can communicate.
• Data Loss Risk: Low — data is replicated and writes are durable.

Distributed Deployment (Raft Protocol)

Hybrid Clustering

Recommended size: 5 or more nodes

Hybrid mode combines the strong consistency of distributed clustering with the scalability of stateless workers. Workers connect to any node in the distributed cluster and automatically failover to another if a connection is lost.

This mode provides high availability from the distributed layer and scalability through the worker layer.

As in Worker-Pool mode, external traffic should be routed only to the worker nodes.

Support

Need help with clustering? Contact support for configuration guidance.

Clustering features are available exclusively to Enterprise customers.