使用感受
之前对Thrift的理解有点不准确,使用之后发现Thrift比想象中的要简单得多。
Thrift做的事情就是跨语言的分布式RPC,通过编写.thrift文件声明接口类和方法,客户端调用定义的方法,Server端实现定义的接口。虽然的确RPC是需要网络请求,但不像Netty这种NIO网络编程库(还要关注很多数据传输中的细节,比如数据如何序列化、如何在字节数组里建立结构、如何在两端解析字节数组、如何处理Handler里的事件状态、如何把多个Handler按顺序串起来),Thrift掩盖了数据传输这件事情,开发者使用的时候就是纯纯的RPC的使用感受。基本使用
Thrift使用起来几乎没有任何门槛,可以看这篇HelloWorld的文章,虽然有点老,但是看完之后基本使用起来没有任何障碍了。
官方给出的这个例子更加全面些,全面在.thrift文件里可以声明的东西列的更全些。
下面看看两个.thrift的定义:shared.thrift
/**
* This Thrift file can be included by other Thrift files that want to share
* these definitions.
*/
namespace java com.baidu.mordor.sink.service
struct SharedStruct {
1: i32 key
2: string value
}
service SharedService {
SharedStruct getStruct(1: i32 key)
}
tutorial.thrift
/**
* The first thing to know about are types. The available types in Thrift are:
*
* bool Boolean, one byte
* byte Signed byte
* i16 Signed 16-bit integer
* i32 Signed 32-bit integer
* i64 Signed 64-bit integer
* double 64-bit floating point value
* string String
* binary Blob (byte array)
* map<t1,t2> Map from one type to another
* list<t1> Ordered list of one type
* set<t1> Set of unique elements of one type
*
* Did you also notice that Thrift supports C style comments?
*/
// Just in case you were wondering... yes. We support simple C comments too.
/**
* Thrift files can reference other Thrift files to include common struct
* and service definitions. These are found using the current path, or by
* searching relative to any paths specified with the -I compiler flag.
*
* Included objects are accessed using the name of the .thrift file as a
* prefix. i.e. shared.SharedObject
*/
include "shared.thrift"
/**
* Thrift files can namespace, package, or prefix their output in various
* target languages.
*/
namespace java com.baidu.mordor.sink.service
/**
* Thrift lets you do typedefs to get pretty names for your types. Standard
* C style here.
*/
typedef i32 MyInteger
/**
* Thrift also lets you define constants for use across languages. Complex
* types and structs are specified using JSON notation.
*/
const i32 INT32CONSTANT = 9853
const map<string,string> MAPCONSTANT = {'hello':'world', 'goodnight':'moon'}
/**
* You can define enums, which are just 32 bit integers. Values are optional
* and start at 1 if not supplied, C style again.
*/
enum Operation {
ADD = 1,
SUBTRACT = 2,
MULTIPLY = 3,
DIVIDE = 4
}
/**
* Structs are the basic complex data structures. They are comprised of fields
* which each have an integer identifier, a type, a symbolic name, and an
* optional default value.
*
* Fields can be declared "optional", which ensures they will not be included
* in the serialized output if they aren't set. Note that this requires some
* manual management in some languages.
*/
struct Work {
1: i32 num1 = 0,
2: i32 num2,
3: Operation op,
4: optional string comment,
}
/**
* Structs can also be exceptions, if they are nasty.
*/
exception InvalidOperation {
1: i32 what,
2: string why
}
/**
* Ahh, now onto the cool part, defining a service. Services just need a name
* and can optionally inherit from another service using the extends keyword.
*/
service Calculator extends shared.SharedService {
void ping(),
i32 add(1:i32 num1, 2:i32 num2),
i32 calculate(1:i32 logid, 2:Work w) throws (1:InvalidOperation ouch),
/**
* This method has a oneway modifier. That means the client only makes
* a request and does not listen for any response at all. Oneway methods
* must be void.
*/
oneway void zip()
}
Thrift通过IDL(接口定义语言),在.thrift文件里声明接口类和方法,声明struct结构、const、Exception等,还可以include别的.thrift文件,这套语法与C非常相似。通过编写IDL和generate代码,做到了不同语言之间的RPC,客户端实现接口类和使用结构类的时候非常简单好用。
上面例子的代码可以从官方下载到,可以放到本地看一下他的使用,非常简单。
Thrift重要组件
Thrift API里三个重要组成部分:Protocal,Transport,Server。
Protocal定义了消息如何序列化。常见的是TBinaryProtocol,TJSONProtocol,TCompactProtocol。
Transport定义了消息在客户端和服务端如何通信。常见的是TSocket,TFramedTransport,TNonblockingTransport等。
TServer说明
Thrift实现的几种不同的TServer。对于Java而言,版本按0.9.0为准:TSimpleServer在sever端只有一个I/O阻塞的单线程,每次只接受并服务一个客户端,适合测试使用,不能用于线上服务。
TNonblockingServer修改了TSimpleServer里阻塞的缺点,借助NIO里的Selector实现非阻塞I/O,允许多个客户端连接并且客户端可以使用select()选择。但是处理消息和select()的是同一个线程,当有大量客户端连接的时候,性能是不理想的。
THsHaServer(半同步半异步server)在以上基础上,使用一个单独线程来处理网络I/O,一个worker线程池来处理消息。好处是只要有空闲worker线程,消息可以被及时、并行处理,吞吐量会大一些。
TThreadedSelectorServer,与THsHaServer的区别是处理网络I/O也是多线程了,它维护两个线程池,一个负责网络I/O,一个负责数据处理。优点是当网络I/O是瓶颈的情况下,性能比THsHaServer更好。
TThreadPoolServer有一个专用的线程来接收connections,连接被建立后,会从ThreadPoolExecutor里取一个工作线程来负责这次连接,直到连接断开后线程回到线程池里,且线程池大小可配。也就是说,并发性的大小可根据服务器设定,如果不介意开很多线程的话,TThreadPoolServer是个还不错的选择。
全文完 :)