ID Allocation
Globally unique IDs are pre-generated and buffered across multiple layers for low-latency
assignment. Each node holds a local pool replenished asynchronously — no coordinator in the hot
path.
pre-buffered
base62
async refill
Read Path
Immutable short-link mappings resolved through a three-tier cache hierarchy — in-process LRU,
global Redis, then persistent storage as fallback. P99 redirects resolve in under 5 ms.
in-process LRU
redis global
< 5ms p99
Scalability
Stateless services orchestrated with Kubernetes. Horizontal pod autoscaling responds to
request rate. No shared mutable state between instances — any node can handle any request.
Kubernetes
stateless
HPA