Work

Scalable search proxy for e-commerce (in progress)

The problem

High-traffic stores depend on fast, reliable product search. When applications query the search backend directly, traffic spikes reach it unfiltered and the cluster is exposed to the network. Teams also have no single place to enforce access rules or tune queries. The result is uneven performance and more operational risk.

Our approach

A lightweight, horizontally scalable proxy tier sits between the application and the search backend:

• Traffic distribution: a load balancer spreads requests across autoscaled proxy instances.
• Protection: per-client rate limits, authentication and request filtering keep abusive or malformed traffic away from the backend.
• Query control: queries are validated against an allow-list, normalized and rewritten, so only safe, efficient queries reach the cluster.
• Caching: repeated queries are served from cache, which cuts backend load and response times.
• Resilience: timeouts and circuit breakers keep a slow backend from cascading into an outage.
• Reduced exposure: the search cluster sits in private subnets, reachable only from the proxy tier.

Users -> CDN / WAF -> Load balancer -> Proxy tier (autoscaled)
                                         |-- rate limit / auth / filter
                                         |-- cache
                                         '-> Search cluster (private)

Compute choices

Proxy nodes run on AWS EC2 T4g (Graviton, ARM) instances, which are cost-efficient for bursty, lightweight request handling. Heavier steady-state work runs on EC2 M6a (AMD) general-purpose instances. Autoscaling policies across both keep costs proportional to traffic.

Status

Build and load testing are underway. We'll publish throughput, latency and cost figures here when they're complete.