本文炒冷饭.说实话,一直挺看好Thrift,支持的语言又多,代码写的有很清晰,效率又不低,为啥研究Protocol Buffer的人那么多.不管那么多了....
Thrift中的对象序列化是我很看好的东西,他用compiler+类库,让你高效的完成任务,而且可以少犯错误.试想,有谁可以保证自己设计的对象,不会再改变呢?数据库的schema改了,你可以改改查询语句,但是如果你对象改了,之前序列化好的东西,有时候就很难搞回来了.(哎.....)
废话不说,看Thrift里面怎么搞的.
1. Thrift支持的数据类型
Thrift支持的数据类型定义在TProtocol.h这个头文件中,有一个TType的枚举:
enum TType {
T_STOP = 0,
T_VOID = 1,
T_BOOL = 2,
T_BYTE = 3,
T_I08 = 3,
T_I16 = 6,
T_I32 = 8,
T_U64 = 9,
T_I64 = 10,
T_DOUBLE = 4,
T_STRING = 11,
T_UTF7 = 11,
T_STRUCT = 12,
T_MAP = 13,
T_SET = 14,
T_LIST = 15,
T_UTF8 = 16,
T_UTF16 = 17
};
2. Thrift对各种数据类型的读写
Thrift把对象序列化抽象成TProtocol这样一个抽象类,这个类的成员非常多,但是思路很明显,就是对各种数据类型的读写操作:
class TProtocol {
public:
virtual ~TProtocol() {}
uint32_t writeMessageBegin(const std::string& name,
const TMessageType messageType,
const int32_t seqid);
uint32_t writeMessageEnd();
uint32_t writeFieldBegin(const char* name,
const TType fieldType,
const int16_t fieldId) ;
uint32_t writeFieldEnd();
uint32_t writeFieldStop();
//写各种类型的数据
uint32_t writeBool(const bool value);
uint32_t writeByte(const int8_t byte);
uint32_t writeI16(const int16_t i16);
uint32_t writeStructBegin(const char* name);
uint32_t writeStructEnd();
//此处省略若干行
uint32_t readMessageBegin(std::string& name,
TMessageType& messageType,
int32_t& seqid);
uint32_t readMessageEnd();
uint32_t readFieldBegin(std::string& name,
TType& fieldType,
int16_t& fieldId);
uint32_t readFieldEnd() ;
//读各种类型的数据
uint32_t readBool(bool& value);
uint32_t readI16(int16_t& i16) ;
uint32_t readStructBegin(std::string& name) ;
uint32_t readStructEnd();
//此处省略若干行
};
每一种数据类型都一对read/write方法,另外Message,Field也有read/write方法.
网友可能很奇怪,为啥抽象类的方法不是虚的.......其实我这里代码省略了很多,之前0.5.0版本的thrift里面,这些方法都是虚的,对TProtocol的实现都重写了这些方法;0.6.0里面,直接的read/write方法都不是虚的,但是添加了额外的虚函数:
//比如说对list的读
virtual uint32_t readListBegin_virt(TType& elemType,
uint32_t& size) = 0;
virtual uint32_t readListEnd_virt() = 0;
//另外有
uint32_t readListBegin(TType& elemType, uint32_t& size) {
T_VIRTUAL_CALL();
return readListBegin_virt(elemType, size);
}
uint32_t readListEnd() {
T_VIRTUAL_CALL();
return readListEnd_virt();
}
其实和直接使用虚函数是一样的.
OK,接口看完了,这就去看对接口的实现,我们来看TBinaryProtocolT是怎么实现的.这里多说几句,Thrift里面实现了好多种序列化,如果你觉得这种序列化不好,可以去重新实现一个上面说的那个接口,就可以工作了:-D
TBinaryProtocolT里面我们看一两个具有代表性的,int32_t/map的读写:
template <class Transport_>
uint32_t TBinaryProtocolT<Transport_>::writeI32(const int32_t i32) {
//把数字转化成网络字节序
int32_t net = (int32_t)htonl(i32);
//然后写入到transport
this->trans_->write((uint8_t*)&net, 4);
//返回写入数据的大小
return 4;
}
template <class Transport_>
uint32_t TBinaryProtocolT<Transport_>::readI32(int32_t& i32) {
uint8_t b[4];
this->trans_->readAll(b, 4);
i32 = *(int32_t*)b;
//读取四个字节,转为本地字节序
i32 = (int32_t)ntohl(i32);
//返回读出数据的大小
return 4;
}
template <class Transport_>
uint32_t TBinaryProtocolT<Transport_>::writeMapBegin(const TType keyType,
const TType valType,
const uint32_t size) {
uint32_t wsize = 0;
//写入一个byte的key类型TType
wsize += writeByte((int8_t)keyType);
//写入一个byte的value类型TType
wsize += writeByte((int8_t)valType);
//再写入元素的个数,int32_t的
wsize += writeI32((int32_t)size);
return wsize;
}
template <class Transport_>
uint32_t TBinaryProtocolT<Transport_>::readMapBegin(TType& keyType,
TType& valType,
uint32_t& size) {
int8_t k, v;
uint32_t result = 0;
int32_t sizei;
//读的时候也是类似,读取key的类型,value的类型,还有元素的个数
result += readByte(k);
keyType = (TType)k;
result += readByte(v);
valType = (TType)v;
result += readI32(sizei);
if (sizei < 0) {
throw TProtocolException(TProtocolException::NEGATIVE_SIZE);
} else if (this->container_limit_ && sizei > this->container_limit_) {
throw TProtocolException(TProtocolException::SIZE_LIMIT);
}
size = (uint32_t)sizei;
return result;
}
可以看到,代码内聚很强,也很易懂.float/double的序列化是通过强转成uint32_t/uint64_t来实现的,string么,先去写一个大小,然后才是内容.list和set都是类似的~~
对于field的read/write都是直接写该filed的类型信息,而field那么就忽略掉了,因为二进制序列化用不到那些东西,只有json这样的文本序列化才能用到:-D,有兴趣的可以去看看JSON Protocol的实现
3. 代码生成
如果让你手写对象的序列化,反序列化,你肯定要抱怨了,因为那样出错的机会非常大.Thrift和Protocol Buffer都给你提供了编译器,写好IDL之后,可以用编译器生成好代码~~这样可以保证不会出错.以UserProfile为例:
struct UserProfile {
1: i32 uid,
2: string name,
3: string blurb
}
生成代码: thrift-0.6.0.exe --gen cpp UserProfile.thrift
这样,thrift会在gen-cpp文件夹内生成好UserProfile的代码,我们只想看UserProfile类的read和write是怎么实现的:
uint32_t UserProfile::write(::apache::thrift::protocol::TProtocol* oprot) const {
uint32_t xfer = 0;
//write的代码比较简单,就是按照顺序
//把field的类型和fieldid和value写进去
xfer += oprot->writeStructBegin("UserProfile");
xfer += oprot->writeFieldBegin("uid", ::apache::thrift::protocol::T_I32, 1);
xfer += oprot->writeI32(this->uid);
xfer += oprot->writeFieldEnd();
xfer += oprot->writeFieldBegin("name", ::apache::thrift::protocol::T_STRING, 2);
xfer += oprot->writeString(this->name);
xfer += oprot->writeFieldEnd();
xfer += oprot->writeFieldBegin("blurb", ::apache::thrift::protocol::T_STRING, 3);
xfer += oprot->writeString(this->blurb);
xfer += oprot->writeFieldEnd();
xfer += oprot->writeFieldStop();
xfer += oprot->writeStructEnd();
return xfer;
}
uint32_t UserProfile::read(::apache::thrift::protocol::TProtocol* iprot) {
uint32_t xfer = 0;
std::string fname;
::apache::thrift::protocol::TType ftype;
int16_t fid;
xfer += iprot->readStructBegin(fname);
using ::apache::thrift::protocol::TProtocolException;
while (true)
{
//read的时候,每次都是先读出来field的类型和fieldid
xfer += iprot->readFieldBegin(fname, ftype, fid);
if (ftype == ::apache::thrift::protocol::T_STOP) {
break;
}
switch (fid)
{
//然后查看当前反序列化的fieldid和读出来的field是不是同一个类型的
//如果是就反序列化
//不是就skip....
//类型就是靠之前说的TType
case 1:
if (ftype == ::apache::thrift::protocol::T_I32) {
xfer += iprot->readI32(this->uid);
this->__isset.uid = true;
} else {
xfer += iprot->skip(ftype);
}
break;
case 2:
if (ftype == ::apache::thrift::protocol::T_STRING) {
xfer += iprot->readString(this->name);
this->__isset.name = true;
} else {
xfer += iprot->skip(ftype);
}
break;
case 3:
if (ftype == ::apache::thrift::protocol::T_STRING) {
xfer += iprot->readString(this->blurb);
this->__isset.blurb = true;
} else {
xfer += iprot->skip(ftype);
}
break;
default:
xfer += iprot->skip(ftype);
break;
}
xfer += iprot->readFieldEnd();
}
xfer += iprot->readStructEnd();
return xfer;
}
使用这个类也是比较简单的,Thrift的transport对读写操作做了一定的抽象,你可以读写网络端口,文件,内存等,我们这里用内存:
typedef unsigned long uint32_t;
typedef unsigned char uint8_t;
uint32_t bufferSize = 64*1024;
uint8_t *buffer = (uint8_t*)malloc(bufferSize);
boost::shared_ptr<TMemoryBuffer> _write(new TMemoryBuffer(buffer,bufferSize,TMemoryBuffer::TAKE_OWNERSHIP));
TProtocol *protowrite = new TBinaryProtocol(_write);
UserProfile _userProfile;
//在这里修改_userProfile的属性
//....
//序列化
uint32_t writeSize = _userProfile.write(protowrite);
boost::shared_ptr<TMemoryBuffer> _read(new TMemoryBuffer(buffer,bufferSize));
TProtocol *protoread = new TBinaryProtocol(_read);
UserProfile _userProfile2;
_userProfile2.read(protoread);
asser(_userProfile == _userProfile2);
4. Thrift向后兼容性的实现
看生成好的read代码可以看到有一个skip的方法(当他类型不一样的时候),这个就是向后兼容性实现的关键.一个对象,难免被改来改去,改不要紧,field id要变化,不要一个field id用到死..当然你继续用也没问题,比如你之前field id 1 type int32_t,之后还是这样,读是读出来了,有可能逻辑不对了.....来看看skip的实现.
//策略就是,读到什么抛弃什么
template <class Protocol_>
uint32_t skip(Protocol_& prot, TType type)
{
switch (type) {
case T_BOOL:
{
bool boolv;
return prot.readBool(boolv);
}
case T_BYTE:
{
int8_t bytev;
return prot.readByte(bytev);
}
//此处省略若干行
//因为代码都是类似的
//对结构体,map/list/set之类的skip操作是比较复杂的
case T_STRUCT:
{
uint32_t result = 0;
std::string name;
int16_t fid;
TType ftype;
result += prot.readStructBegin(name);
while (true) {
result += prot.readFieldBegin(name, ftype, fid);
if (ftype == T_STOP) {
break;
}
result += skip(prot, ftype);
result += prot.readFieldEnd();
}
result += prot.readStructEnd();
return result;
}
}
OK,至此,Thrift对象序列化的代码基本上就看的差不多了.因为我们不会去用Thrift的Service,所以那部分代码没看过....
Thrift千好万好,但是如果你数据写好了,schema丢失了.....那就不好玩了
PS:很早之前看过Thrift,但是没写篇文章总结一下.这篇也算是了了心愿.