README.md

Webserv

HTTP/1.1 compatible web server in C++ using I/O multiplexing.

Motivation and Considerations

Webserv is crafted as part of the challenging curriculum at School 42, focusing on developing a HTTP server in C++ 98. Key requirements include:

• Non-blocking socket read/write using poll()(or equivalent).
• Ensure the server never blocks(non-blocking I/O is necessary).
• Ensure the server never hangs indefinitely.
• Ability to serve a fully static website.
• Support for GET, POST, and DELETE methods.
• Capability to listen to multiple ports. These elements combine to create a robust, efficient web server.

Technologies Used

• Languages: C++ 98, Makefile
• Web Technologies: HTML, CSS
• Additional Tools: Python
• Programming Concepts: Object Oriented Programming, Non-block I/O, I/O Multiplexing (kqueue), HTTP/1.1

Implementation Details

Architecture

Configuration

http {
	error_page 404 500 413 ./www/error/default.html;
	index index.html ;
	autoindex on;
	server {
		server_name localhost:4000;
		listen  4000;
		location / {
			alias ./www/example/;
			limit_except POST GET DELETE;
			
			location /box/ {
				limit_except GET;
			}
		}
		location /abc/ {
			alias ./www/example/;
		}
	}
	server {
		listen 42;
		client_max_body_size 100;
		location / {
			alias ./www/example/;
		}
	}
}

A Bottom-up Approach In the initialization process, a web server parses the configuration in a tree shape. This allows each location configuration to override specific settings like Nginx. Regex is partially implemented. (*, $) This is implemented in Config.{hpp, cpp} in less than 100 lines.

A Top-down Approach After making the tree, the configuration is being mapped into std::<map>(heap). This approach is to optimize the configuration access time of each HTTP request. The search is processed in several steps: 1. Port 2. Hostname 3. Location By using this method, the access time is reduced from $O(N)$ to $O(log N)$.(We could have reduced it to O(1) if we used std::<unordered_map>(hashmap), but C++98 doesn't have that STL.) This is implemented in ConfigMap.{hpp, cpp}

Event-Driven Architecture (I/O Multiplexing)

Flowchart

This is how this program flows. Server, client sockets, and children’s PID are registered to kqueue.

	class Server, Connection, CGI inherits class IEventHandler which has virtual void EventHandler(). When an event is caught, it calls EventHandler() thus calling the corresponding function in each class.

Server never falls into blocking I/O

• Server can process multiple requests simultaneously: We used pipe() to communicate with CGI processes. Whenever the buffer is full, the server registers an event(EVFILT_WRITE) to kqueue and moves on to another event in kqueue
• An idea from round-robin algorithm: Since we can queue events by using kqueue(epoll), the server slices its I/O. However, we did not set timeout to maximize throughput.

Key Features

• HTTP/1.1 Compliant: Supports GET, POST, and DELETE methods, along with features like HTTP pipelining and cookies.
• Nginx-like Configuration: Tree-shaped parsing for nested bracket {} configurations, resembling Nginx.
• Efficient Configuration Access: A sophisticated mapping system reduces access time from O(N) to O(log N).
• Unblocking I/O: Since the server should support multiple requests simultaneously, I/O never falls into blocking mode.

Challenges and Solutions

• Class Destruction: Implemented smart pointers for efficient object management, adapting to C++ 98's limitations.
• Forking and Multithreading: Refined our approach to meet project requirements, gaining insights into process creation and management in Linux.
• Parsing Complexity: Developed efficient parsing strategies for both configuration files and HTTP requests. For configuration parsing, we've used bitmask to make the logic as simple as possible. For HTTP request paring, we used Nekipelov’s httpparser and modified it to fit our usage. That helped us a lot. Thank you!

Installation and Usage

Ensure you have a macOS environment and a C++ 98 compiler.

1. Clone the repository: git clone [repository link]
2. Compile the project: make re at the root directory.
3. Run the server: ./webserv configs/example.conf

|

Results and Discussion

Performance

We achieved approximately 43% of NGINX's performance on the Siege test.

This project was a deep dive into web server mechanics, leveraging C++ with OOP and I/O multiplexing. Our journey was marked by constant problem-solving and code refinement, enhancing our collaboration and communication skills. Despite some initial challenges, the final product is a testament to our perseverance and commitment to learning.

Acknowledgements

Nginx

GitHub - nginx/nginx

RFC 2616

RFC 2616: Hypertext Transfer Protocol -- HTTP/1.1

HTTP Parser

GitHub - nekipelov/httpparser: HTTP request, response and urls parser

Kernel Queue

Kernel Queue: The Complete Guide On The Most Essential Technology For High-Performance I/O

License

MIT License