escandell.

Administracio de sistemes

SaaS Portal & WP Connector

SaaS platform to centrally manage WordPress sites with security, backups, SEO, tickets, and billing.

Status
En produccio
Role
Desenvolupador principal i arquitecte

Context

Managing WordPress maintenance for multiple clients required connecting to each site individually to run updates, perform backups, verify security, and handle tickets. Without a unified tool, the process was slow, fragmented, and difficult to scale.

What it does

  • Security audit with 0-100 score, 10 categories and automatic hardening
  • Scheduled backups with 4 types (full, db, files, no-media), retention, restoration and cloud storage via SSH+rsync
  • Advanced SEO with Lighthouse audit (Node.js worker) and broken link detection
  • Ticket system with Kanban, priorities and bidirectional Portal-Plugin WordPress sync
  • Billing with automatic calculations, PDF generation and status tracking
  • Autonomous AI Agent (nanobot) with maintenance actions, human approval via Telegram
  • Audit-service to capture leads without needing to install the plugin
  • Content AI for post generation, alt-text and internal links

Architecture

Microservices system on Docker Compose with 8 containers: FastAPI backend (20+ routers, 17 DB models, 5 background workers), PostgreSQL 15, nanobot, Lighthouse worker (Node.js), broken-links worker (Python), audit-service. The WordPress plugin connects each client site through 43 REST endpoints authenticated with RSA SHA256 signature.

Main difficulties

  • 502 error in production caused by an await outside an async function in FastAPI
  • 14 security vulnerabilities found during the audit: SQL injection (CVSS 9.8), path traversal (CVSS 9.8), IDOR, SSRF, CSRF, insecure cookies, running as root in Docker
  • Event loop timeouts when WordPress servers (ISP/Cloudflare) did not respond
  • SQLite to PostgreSQL migration without downtime, using custom migration scripts
  • Cryptographic PHP-Python communication with RSA signature for site tokens

Solutions

  • CSRF middleware based on cookies, SSRF protection (blocking private IPs and cloud metadata endpoints), rate limiting with slowapi
  • SQLAlchemy parameterized queries, filename sanitization, authorization verification on all endpoints
  • asyncio.to_thread + 20s timeout to avoid blocking the event loop
  • Per-script migration system for controlled DB schema evolution
  • RSA token system with cross-signing and verification between PHP and Python
  • Dockerfile with a non-root user, HTTP-only cookies with a configurable secure flag
  • LiteLLM proxy to decouple the AI provider from business logic

Learnings

  • In FastAPI, never mix await calls inside sync functions: tracking down a 502 error trace is extremely difficult
  • Security must be designed from day one: SQL injection, SSRF and path traversal are preventable with the right patterns
  • Background workers require idempotency and solid error handling to avoid cascading failures
  • An AI agent running in production demands a robust approval system for high-risk actions
  • SQLite is excellent for development, but PostgreSQL is a must-have in production