Back to Projects
Case StudyPrivate product architecture

Distributed Task Queue System

Designed a coordinator-free queue where PostgreSQL remains the source of truth and Redis handles fast lease acquisition and scheduling.

GoRedisKubernetes

Impact

40K+ daily jobs · P99 < 200ms

Source

Private code

Repos

3 linked services

Repository Shape

Worker Service
Private
Scheduler API
Private
Ops Manifests
Private

Private repositories are represented through architecture notes, impact, and design tradeoffs instead of source links.

Role

Backend system design, queue semantics, observability, and production hardening across the worker service, scheduler API, and deployment manifests.

Architecture

  • PostgreSQL stores durable job state, retry history, idempotency keys, and tenant ownership.
  • Redis Lua scripts acquire leases atomically so multiple workers can compete without double execution.
  • Stateless Kubernetes workers heartbeat active leases and reclaim abandoned jobs through reconciliation.

Highlights

  • 40K+ daily jobs with P99 latency below 200ms.
  • Tenant-hash sharding prevents noisy-neighbor job bursts from starving other tenants.
  • Dead-letter queue and exponential backoff make poison jobs inspectable without blocking healthy work.

Constraints

Private Code

Cannot publish source because the implementation belongs to a private product. Operational notes and design decisions are shareable as a case study.