How We Built a High‑Performance Ad Monitoring Service with OpenResty

Part.1 Initial State

At first the ad monitoring service was a simple client‑side reporting system that recorded exposure and click events, but it suffered from inaccurate data, limited processing capacity, burst‑traffic failures (Redis memory pressure), complex deployment, and low development efficiency.

Part.2 Architecture Based on OpenResty

We created ADMonitor to provide a stable, reliable, and highly available ad monitoring service.

2.1 Design Idea

The new workflow generates a unique monitoring link for each ad, attaches it to the original customer link, and makes both exposure and click requests depend on ADMonitor. Clients request exposure in parallel, while click requests are redirected through ADMonitor for a second‑hop jump.

2.2 Technical Selection

OpenResty was chosen because it operates at the network layer, offers powerful regex‑based routing, provides Lua coroutine‑driven event loops, and benefits from LuaJIT’s just‑in‑time compilation for high performance.

2.3 Implementation

2.3.1 Data Collection

Lua scripts handle request inspection, filtering, and pushing data to shared memory, avoiding extra network hops. Optimizations include enabling lua_code_cache, configuring DNS resolvers, tuning epoll settings (multi_accept, accept_mutex, worker_connections), and setting keepalive parameters.

2.3.2 Data Processing

Collected logs undergo ETL and counting. ETL normalizes logs, extracts meaningful fields, filters noise, and redirects data. Counting stores metrics in Redis and FluxDB, using connection pooling and clustering for reliability and time‑series query efficiency.

2.3.3 Data Archiving

Full data is archived via the company’s big‑data pipeline: online Flink and offline Hive ingest logs, then store them in Elasticsearch. Kafka handles streaming, and automated scripts monitor failures.

2.4 Other OpenResty Use Cases

Splitter: Lua scripts query RedisCluster for NodeJS server load, then perform circuit‑break, downgrade, and rate‑limit, reporting metrics to InfluxDB.

Lightweight Web Firewall: Using lua_resty_waf library with IP/URL/UA filtering and CC protection, extended to push alerts to InfluxDB.

Part.3 Summary

ADMonitor, built on OpenResty, delivers high availability, immediate response, decoupled modules, and robust data guarantees. Performance metrics show >99.9% exposure success and >99.8% click success, with throughput and latency meeting strict standards.