Design and Pitfalls of Internet App Long‑Connection Push Systems and Related Backend Practices

The article begins by distinguishing three connection models—long connections (e.g., a persistent socket opened by a Java or Flash client), half‑long connections used by web browsers via JavaScript or Flash plugins, and short connections that poll the server at intervals.

Client‑side handling methods include JavaScript setTimeout , JavaScript calling Flash/Java wrappers, and Node.js WebSocket libraries.

On the server side, three layers are described: an I/O processing layer that holds many connections, a logic processing layer, and a data storage layer.

Popular implementation stacks are compared: (a) Nginx + C/C++ modules (high performance, hard to develop), (b) Nginx + Lua + Redis (easy development with high‑performance I/O), (c) Node.js + Redis (simple, often using socket.io), and (d) Go + Redis (goroutine‑based, capable of handling hundreds of thousands of connections per machine, up to millions after optimization). Typical backend components include register_server, user‑server, push‑server, message‑server, and storage‑server.

Additional reading references are listed, such as Comet (HTTP long‑connection push) and security hardening guides.

Security advice covers application‑layer protections (injection, XSS, CSRF), network‑layer defenses (load balancing, firewalls, IP blacklisting, rate limiting with Redis counters, cookie‑based limits), and practical tools like fail2ban and hardware/software firewalls.

HTTP API design tips discuss RESTful conventions, idempotency, locking strategies, request signing, OAuth, rate limiting (IP‑based, token buckets), and the trade‑offs between stateful and stateless authentication.

Operations tooling for database administration is shared, mentioning phpMyAdmin, MongoDB admin, Redis admin, SQLyog, HeidiSQL, Navicat, and the importance of consistent naming conventions across code, forms, and database fields.

The article then compares various queue services: Redis (simple and versatile), ZeroMQ, RabbitMQ, Apache Kafka (high throughput), Beanstalkd, NSQ, and language‑specific solutions like Celery for Python. Recommendations emphasize using queues to off‑load unpredictable tasks such as push notifications.

Debugging and troubleshooting techniques are enumerated, including breakpoint usage, error logs, network diagnostics (ping, dig, traceroute, tcpdump, wireshark), log analysis, and code snippets such as: if(isset($_GET['debug'])) { register_shutdown_function(function(){ var_dump(error_get_last()); }); } These help capture fatal errors without modifying configuration.

For Web IM message systems, the discussion suggests a hybrid approach: real‑time socket.io pushes for online users, fallback to Redis or database queues for offline users, and strategies for scaling to millions of concurrent connections.

Finally, the article surveys NoSQL use cases, highlighting Redis, Memcached, MongoDB, and Hadoop/Hive for caching, high‑frequency data, and large‑scale analytics, while noting pitfalls such as Redis persistence errors and the need for proper memory allocation.