RelayCore Engine · Case Study · Łukasz Kubok

┌─────────────────┐ ┌──────────────────────┐ │ Ingress Traffic │ ───> │ Raft Consensus Group │ └─────────────────┘ └──────────────────────┘ │ │ ▼ ▼ ┌─────────────────┐ ┌──────────────────────┐ │ Memory Queue │ <───> │ Dispatcher Worker #8 │ └─────────────────┘ └──────────────────────┘

Problem

Mission-critical asynchronous payment processing and telemetry webhook dispatches require at-least-once delivery guarantees even in the face of sudden node restarts, network partitions, and noisy neighbor spikes.

Role

Designed and implemented the core message broker, consensus coordinator, and backpressure scheduling pipeline.

Architecture & Stack

Raft Consensus Group: 5-node consensus protocol managing partition elections and dynamic leader assignment with sub-100ms failover.
Predictive Backpressure Scheduler: Rate-limits external third-party webhook sinks using adaptive sliding token buckets.
Write-Ahead Logging (WAL): Disk-backed durable WAL in Go paired with memory-mapped queues for microsecond dispatcher throughput.

Outcome & Metrics

Benchmarked at 100,000 messages/sec dispatch rate on modest 4-core worker nodes with p99 latency under 1ms.
0 message loss during chaos injection tests simulating partition failures and killed leaders.