Built for Zero Downtime

All frontend assets are served from Cloudflare's global CDN with 300+ points of presence. A Cloudflare Worker sits at the edge and intercepts every /api/* request, routing it to the active backend. Static pages, the interactive map, Land & Power tool, and all UI components load instantly from cache — no backend dependency for initial render. The Worker also serves AI agent discovery files (llms.txt, openapi.json) directly from edge, so AI platforms like ChatGPT, Claude, and Gemini can discover and integrate DC Hub without hitting the origin.

The Python/Flask REST + MCP server runs on Railway with a 10-second health timeout at the CF Worker edge. The worker buffers POST request bodies so form submissions and authentication survive transient retries. X-DC-Worker-Version headers identify the build that served each request for monitoring. Auto-redeploy fires on every push to main in azmartone67/dchub-backend. When Railway is unreachable the CF Worker serves a 24h stale-cache fallback from KV so the site degrades gracefully instead of going dark.

All platform data — 20,000+ facility records, M&A transactions, capacity pipeline, news articles, user accounts, API keys, and alert configurations — is stored in Neon's serverless PostgreSQL. Neon provides provider-managed high availability with automatic failover, point-in-time recovery, and connection pooling. An automated nightly backup pipeline dumps the full database to Cloudflare R2 object storage with gzip compression and 30-day retention, providing an independent disaster recovery layer completely separate from the database provider. Railway connects via a managed connection pool (max 20). Write operations (discovery auto-approval, news sync, AI citation tracking) use idempotency keys on every INSERT so retries from the CF edge cannot create duplicates. The edge KV layer holds JSON responses up to 24h stale-while-revalidate so the public site survives transient Railway outages without showing 5xx pages.

Every data category on the platform — facilities, market intel, transactions, pipeline, news, announcements, chart data, and comparison metrics — has hardcoded fallback data embedded in the frontend. If every backend is down simultaneously, the map still loads with 12,000+ generated facility pins across 70+ global markets, the news feed shows cached articles, and the transaction table displays recent deals. Users never see a blank page. The map uses progressive loading: 2,000 facilities render instantly, with remaining records loading in background batches. A 30-second API timeout prevents hung connections.