TaskMesh — Real-Time Collaborative Workspace

Collaborative tools are concurrency problems disguised as CRUD apps. TaskMesh separates writes from reads so the read path stays fast and consistent under multi-user load, uses domain events to keep notifications, audit, and analytics in sync, and chains audit entries so any tampering breaks the hash chain — a verifiable trail of who did what, when.

CQRS: CommandRepository validates via Zod, publishes domain events after writes, invalidates cache, and returns the entity; ReadRepository uses a cache-aside getOrSet pattern with L1 in-memory Map (LRU, 1000 entries) and L2 Redis with TTL. A custom Socket.IO server streams 26 server-to-client and 19 client-to-server events for presence, typing, and live board updates.

24 Prisma models across 9 enums: users/auth, workspaces/members/invitations, boards/columns, tasks/assignees/labels/comments/subtasks, sprints, activity logs, notifications, hash-chained AuditEntry, feature flags and rules, API keys, webhooks.

Real-time collaboration without data races required scoping every mutation to a workspace channel and using optimistic UI with server reconciliation. The audit chain had to be append-only and verifiable. The two-layer cache needed targeted invalidation on every write to avoid stale reads — each write publishes events that trigger cache invalidation.

9 BullMQ workers (email, audit, analytics, notification, webhook, cleanup, search-index, presence, feature-flag) handle side effects asynchronously. Kubernetes manifests include HPA (2-10), PDB, StatefulSets, and PVCs. Docker multi-stage builds with HEALTHCHECK. Six GitHub Actions jobs run lint, typecheck, test, build, docker, and security.

BullMQ retries with exponential backoff and dead-letter queues. Socket.IO reconnects with presence reconciliation. Cache misses fall through to the database. Database and Redis health checks gate requests.

Prometheus metrics across 11 metric types — HTTP request counts/duration/errors, cache hit ratios, WebSocket connection counts, BullMQ queue depth and job duration, rate limit hits, auth attempts, and feature flag evaluations.

CQRS over plain repositories for read-path scalability. BullMQ events over synchronous side effects for decoupling. Two-layer cache over single-layer for sub-ms reads. dnd-kit over native DnD for accessibility and touch support.