An in-memory, partition-tolerant distributed cache node designed for low-latency key-value storage. Implements consistent hashing with virtual nodes to minimize re-sharding overhead and utilizes a decentralized gossip membership protocol for automatic node discovery.

Architectural Overview

code

Client Request -> Hash Ring Lookup -> Shard Leader (gRPC) -> In-Memory LRU
                      |
           Gossip Heartbeat Protocol (UDP)
                      |
               Peer Replicas (Sync)

Core Components

Consistent Hashing Ring: Employs a 32-bit CRC hash ring with 150 virtual nodes per physical instance to ensure uniform key distribution across cluster boundaries.
Gossip Protocol: Periodic UDP heartbeats evaluate node reachability with adjustable suspicion intervals, preventing split-brain states during transient network partitions.
gRPC Interface: High-throughput multiplexed RPC endpoints provide sub-millisecond retrieval and mutation operations with protocol buffers serialization.
LRU Eviction Engine: Thread-safe doubly-linked list with hash map indexing, guaranteeing O(1) reads, updates, and memory-bounded evictions.

Benchmarks

Throughput: 120,000 requests/sec per shard (single m6i.xlarge node).
p99 Latency: 0.42ms on local VPC networks.
Failover Convergence: Under 1.2 seconds across 7 cluster nodes.

Build and Run

code

# Build binary
go build -o bin/cache-node src/main.go
 
# Start primary node
./bin/cache-node --port=8080 --peers="10.0.0.2:8080,10.0.0.3:8080"