Java to Go: From Locks to Lock-Free

A legacy Java TCP chat server had a fundamental flaw: shared mutable state protected by locks. Rebuilt from scratch in Go using single-owner goroutines and message passing instead.

GoConcurrencyCSPDockerPrometheus

Problem → Solution

Problem	Java	Go
Synchronization	Mutex-heavy, deadlock risk	No locks anywhere
I/O Model	Thread-per-connection, blocking	Goroutines, non-blocking channels
Shared State	Concurrent mutations, race conditions	Single Registry goroutine (sequential)
Backpressure	Threads block on queues	Drop slow clients, unblock writer
Race Safety	Manual review, hard to verify	`go test -race` (zero detections)

Architecture: Registry Goroutine as Single Source of Truth

Client 1 (read) ──┐
Client 1 (write)──┤
                  ├──> [Registry] <──────> [Shared State]
Client N (read) ──┤                        (users, messages)
Client N (write)──┘

All mutations flow through Registry via channels (CSP).
No concurrent access. No locks. No races.

Features: Register username, broadcast, whisper (/w), list users (/users).

Evidence: Zero Data Races

go test -race passes with zero detections across all state transitions
Test coverage: duplicate rejection, user list consistency, whisper routing
Prometheus metrics: connected clients, message throughput, latency (p50/p95/p99)
Grafana dashboard auto-loads with docker-compose up --build

View on GitHub →

Previous Project← Conway's Game of Life: Parallel vs Distributed

Next ProjectConcurrent Order Book →