Go: Sub-15ms Dynamic Edge Router & Bang Engine

System Endpoint: https://go.shagoto.me • Ecosystem: shagoto.me • Architecture: Cloudflare Workers (V8 Isolates) + Hono + KV Cache + D1

The Constraint

Traditional URL shorteners and redirect services rely on centralized database architectures and heavy server-side frameworks. Every redirect incurs cold start penalties, cross-region TLS handshakes, and blocking analytics writes that delay page transitions by 200–600ms.

For go.shagoto.me, the primary engineering directive was ultra-low-latency edge execution:

• Global Sub-15ms Redirects: Requests must resolve at the nearest edge Point of Presence (PoP) out of 300+ worldwide data centers without hitting centralized origin servers.
• Non-Blocking Telemetry: Analytics recording (IP-less geocoding, user agent parsing, referral source, timestamp) must run strictly off the critical path, adding $0\text{ ms}$ overhead to the HTTP redirect.
• CLI-First Ergonomics: First-class management through Linux shell pipelines via standard curl commands, supporting instant link creation, updates, and format negotiation.

---

The Execution

go.shagoto.me is built using the Hono web framework running on Cloudflare Workers, combining an ultra-fast in-memory routing engine ($\sim 14\text{ KB}$ runtime overhead) with a hybrid two-tier storage layer.

1. Hybrid KV + D1 Storage Architecture

To achieve deterministic sub-15ms execution while preserving full SQL auditability, storage is split into complementary tiers:

1. Hot Path (Workers KV): Global key-value cache with sub-10ms read latencies. The edge worker resolves incoming slugs (/cv, /thesis, /meet) directly from KV cache without querying the relational layer.
2. Cold & Relational Tier (Cloudflare D1): Serverless SQLite acting as the authoritative source of truth. Handles structured metadata, link updates, access control verification, and click telemetry streams.
3. Cache Invalidation: When a link destination is updated via API or CLI, the worker writes the change to D1 and simultaneously flushes and updates the corresponding KV key across all edge locations.

2. Parameterized “Bang” Routing Engine

To eliminate lookups entirely for standardized destinations, go.shagoto.me implements deterministic pattern-matched “bang” routes evaluated in V8 memory with 0ms database overhead:

import { Hono } from 'hono';

const app = new Hono();

// Algorithmic Bang Rules (Zero Storage Overhead)
const BANG_PATTERNS: Record<string, (arg: string) => string> = {
  gh: (repo) => `https://github.com/shagoto-sarkar/${repo}`,
  arxiv: (id) => `https://arxiv.org/abs/${id}`,
  doi: (doi) => `https://doi.org/${doi}`,
  yt: (query) => `https://www.youtube.com/results?search_query=${encodeURIComponent(query)}`
};

app.get('/:prefix/:arg', async (c) => {
  const { prefix, arg } = c.req.param();
  const rule = BANG_PATTERNS[prefix.toLowerCase()];

  if (rule) {
    const destination = rule(arg);
    // Asynchronous non-blocking telemetry dispatch
    c.executionCtx.waitUntil(recordTelemetry(c, `bang:${prefix}`, destination));
    return c.redirect(destination, 302);
  }

  return c.notFound();
});

3. Non-Blocking Edge Telemetry via ctx.waitUntil()

Click telemetry is decoupled completely from the visitor’s HTTP response. Using the Cloudflare Worker execution context (c.executionCtx.waitUntil), the worker schedules an asynchronous background promise to write the telemetry payload to D1 after the HTTP 302/301 response stream has terminated.

4. Dynamic In-Memory SVG QR Code Matrix

For physical-to-digital workflows (e.g., conference presentations, academic posters, CV headers), appending .qr to any slug (e.g., https://go.shagoto.me/cv.qr) instructs the worker to dynamically compute the QR Reed-Solomon error correction matrix and stream an inline scalable SVG directly from V8 memory with zero native image library dependencies.

---

Engineering Takeaways

By treating URL routing as an edge-native problem and leveraging non-blocking isolate threads, go.shagoto.me delivers instantaneous personal link navigation while gathering rich operational telemetry without degrading visitor experience.