SGW

SGW is a simple non-blocking micro-service gateway built upon Netty, it stands between different types of http clients and all micro-service backends (mostly using faster RPC protocols).

The main purpose of this project is to build a protocol transition and request dispatching layer, and aims to handle large throughput (which results in using Netty as the base framework).

Other features include:

• Service discovery/load balancing
• Pooled connections
• Zuul style Filters
• Spring style router configuration

中文文档

Dependencies

• use Gradle to build project.
• Netty 4.1
• Thrift 0.10.0
• CuratorFramework

Target functions

0. Non-blocking IO

1. Dynamic Routers

   • Router configuration using "routing.yaml"

      thriftServices:
        - http: POST /echo # http请求
      	convertor: # 解析请求和生成响应的类
      	service: echoservice # 下游服务名（zookeeper中的名字）
          method: echo # 下游服务方法名
          clazz: # thrift生成的类名
        - http: ....
        ...
      xxxServices:
     
     

   • Spring style router configuration

      @ThriftRouter(http = {"POST", "/echo"}, service = "echoservice", method = "echo", args = EchoService.echo_args.class, result = EchoService.echo_result.class)
     public class EchoConvertor{
      	
      	@RequestParser
      	public Object[] parse(FullHttpRequest request) {
      		return new Object[] {request.content().toString(CharsetUtil.UTF_8)};
      	}
     
      	@ResponseGenerator
      	public String generate(String result) {
      		return result;
      	}
      }
     

   • Routing templates:

      	@RequestParser
         	public Object[] parse(FullHttpRequest request, @PathVar("id") int id) {
         		...
         	}
     

   • Routing configuration hot replacement (under development)

2. protocol conversion from HTTP to Thrift

   currently hard coded, plan to enable configuration on thrift protocols.

   • protocol：TMultiplexedProtocol, TCompactProtocol
   • transport： TFramedTransport

   Also planed to provide user interfaces to help support more protocols.

3. service discovery/load balancing

   • cache service node metadata in memory
   • receive Zookeeper update in real time
   • using round-robin load balancing strategy. (plan to support customization)

4. filters

   • preRouting filters: filter requests when requests arrive.
   • postRouting filters: filter responses before sending response.
   • routing filters: filter requests before sending to backend services.

5. circuit breaker (under development)

   behave like hystrix.

   • built purely on filters

Request - Response lifecycle

1. decoding http requests from network
2. pre-routing filters
3. Routing: find the matched backend service and use load balancing to get the target service node.
4. routing filter
5. convert http request to thrift request (@RequestParser)
6. get a service connection (either established or new) and send the encoded thrift request to the backend.
7. decode thrift response
8. convert thrift response back to http response (@ResponseGenerator)
9. post-routing filter
10. send http response back to client.

About Extension

Demo

sgw.demo.DemoServer