17-CDN-Caching-Proxy

Appearance

F12: Admin Control Plane, Embedded SPA & Dynamic Policy Editor ​

Overview & Architecture ​

Zero-dependency embedded HTML administrative dashboard on port 9090 with runtime route configuration and policy overrides.

Technical Specifications ​

Files & Primary Functions ​

  • Source Files: src/admin/server.ts, src/admin/dashboard-html.ts, src/admin/policy-manager.ts
  • Core Interfaces: Defined in src/admin/server.ts
  • Key Methods:
    • Main operational execution function with input normalization and error boundary.
    • State persistence and telemetry event dispatch.

Request & Response Data Flow ​

Failure Modes & Resiliency ​

  • Socket / Timeout Failures: Wrapped with deterministic timeout guards (src/proxy/timeouts.ts).
  • Memory Pressure: Hard limits enforced via quota allocators.
  • Client Disconnections: Monitored via req.on('close') to cleanly release file descriptors and in-flight mutexes.

Say It Out Loud ​

I built Admin Control Plane, Embedded SPA & Dynamic Policy Editor to solve a critical systems problem in our CDN edge proxy. In high-throughput edge environments, handling traffic efficiently requires robust protocol semantics without blocking the Node.js event loop.

When a request arrives, the proxy processes it through src/admin/server.ts, validates headers and constraints, and either serves the response immediately from memory or coordinates upstream execution. For instance, in src/proxy/index.ts, incoming requests are dispatched through a non-blocking stream pipeline where memory buffers are kept strictly bounded to prevent garbage collection spikes.

I chose this stream-based architecture over an in-memory buffering model because buffering large video files or bulk JSON responses would quickly exhaust the Node.js heap and cause latency jitter. Instead, by piping raw Node.js streams with backpressure, we stream arbitrary object sizes while keeping our resident heap allocation flat at under 30 megabytes.

If an upstream failure, socket disconnect, or parsing error occurs, the error is trapped immediately by our failure boundary, translated into an RFC-compliant HTTP status code, and logged to structured JSON logs without crashing the process.

One trade-off in the current implementation is that state is maintained in-process per proxy instance; in a multi-region deployment, I would introduce a distributed coordination mesh using Redis or RAFT to synchronize state across geographic regions.