protobuf 、protostuff、json 经过nginx的gizp压缩前后的大小比较

公司一个cs架构项目接口，使用json传输数据大小接近10M，简单的研究了一下protobuf 、protostuff，以及针对这个接口场景 使用protobuf 、protostuff、json序列化nginx的gzip压缩传输后文件大小的比较。

1.protobuf简单使用

下载protoc-3.5.1-win32.zip 和 protobuf-java-3.5.1.zip 分别解压

把 protoc-3.5.1-win32.zip里的protoc.exe拷贝到 E:\develop\protobuf-3.5.1\src （不考也行）

E:\develop\protobuf-3.5.1\src\protoc.exe 是protoc.exe 所在目录

E:\develop\protobuf-3.5.1\examples 是 .proto所在的目录

进入要编译的.proto文件目录（E:\develop\protobuf-3.5.1\examples） 执行一下命令

E:\develop\protobuf-3.5.1\examples>E:\develop\protobuf-3.5.1\src\protoc.exe --j

va_out=. addressbook.proto

会在当前目录生成AddressBookProtos.java

还可使用maven插件使每次打包的时候自动把.proto文件编译成.java文件

这里没有这么做，使用protocol buffer主要的工作是要另外编写.proto文件

对应现有系统改的比较大所以没有使用protobuffer方案。

.proto还是比较简单，可以自行学习

下面给出一下例子，只是为了忘记时候可以回忆一下，可以略过

syntax = "proto3";

option java_package = "com.*.*.*.*.rcp.protobuf";

option java_outer_classname = "XXXXProto";

import public "FltBookResult.proto";

import "ResultBase.proto";

message XXXXInfo {

/**

* <p>

* Description:航段唯一 ID 号

* </p>

*/

string soflSeqNr = 1;

.......

FltBookResult fltBookResult = 96;

ResultBase resultBase = 97;

}

2.prostostuff的使用

结合springmvc的Converter

import com.google.common.base.Stopwatch;

import io.protostuff.LinkedBuffer;

import io.protostuff.ProtobufIOUtil;

import io.protostuff.Schema;

import io.protostuff.runtime.RuntimeSchema;

import net.jpountz.lz4.LZ4BlockOutputStream;

import org.apache.commons.io.IOUtils;

import org.slf4j.Logger;

import org.slf4j.LoggerFactory;

import org.springframework.http.HttpInputMessage;

import org.springframework.http.HttpOutputMessage;

import org.springframework.http.MediaType;

import org.springframework.http.converter.AbstractHttpMessageConverter;

import org.springframework.http.converter.HttpMessageNotReadableException;

import org.springframework.http.converter.HttpMessageNotWritableException;

import org.xerial.snappy.SnappyOutputStream;

 

import java.io.IOException;

import java.io.InputStream;

import java.io.OutputStream;

import java.nio.charset.Charset;

import java.util.zip.GZIPOutputStream;

 

/**

* <p>

* An {@code HttpMessageConverter} that can read and write

* <a href="https://developers.google.com/protocol-buffers/">Google Protocol Buffer (Protobuf)</a> messages using the

* open-source <a href="http://http://www.protostuff.io">Protostuff library</a>. Its advantage over native Protobuf

* serialization and deserialization is that it can work with normal {@code POJO}s, as compared to the native

* implementation that requires the objects to implement the {@code Message} interface from the Protobuf Java library.

* This allows applications to use Protobuf with existing classes instead of having to re-generate them using the

* Protobuf compiler.

* </p>

* <p>

* Supports the {@code application/x-protobuf} media type. Regular Spring MVC application clients can use this as the

* media type for the {@code Accept} and {@code Content-Type} HTTP headers for exchanging information as Protobuf

* messages.

* </p>

*/

public class ProtostuffHttpMessageConverter extends AbstractHttpMessageConverter<Object> {

private Logger logger = LoggerFactory.getLogger(getClass());

public static final Charset DEFAULT_CHARSET = Charset.forName("UTF-8");

public static final MediaType MEDIA_TYPE = new MediaType("application", "x-protobuf", DEFAULT_CHARSET);

public static final MediaType MEDIA_TYPE_LZ4 = new MediaType("application", "x-protobuf-lz4", DEFAULT_CHARSET);

public static final MediaType MEDIA_TYPE_GZIP = new MediaType("application", "x-protobuf-gzip", DEFAULT_CHARSET);

public static final MediaType MEDIA_TYPE_SNAPPY = new MediaType("application", "x-protobuf-snappy",

DEFAULT_CHARSET);

 

/**

* Construct a new instance.

*/

public ProtostuffHttpMessageConverter() {

super(MEDIA_TYPE, MEDIA_TYPE_LZ4, MEDIA_TYPE_GZIP, MEDIA_TYPE_SNAPPY);

}

 

/**

* {@inheritDoc}

*/

@Override

public boolean canRead(final Class<?> clazz, final MediaType mediaType) {

return canRead(mediaType);

}

 

/**

* {@inheritDoc}

*/

@Override

public boolean canWrite(final Class<?> clazz, final MediaType mediaType) {

return canWrite(mediaType);

}

 

/**

* {@inheritDoc}

*/

@Override

protected Object readInternal(final Class<?> clazz, final HttpInputMessage inputMessage)

throws IOException, HttpMessageNotReadableException {

if (MEDIA_TYPE.isCompatibleWith(inputMessage.getHeaders().getContentType())) {

final Schema<?> schema = RuntimeSchema.getSchema(clazz);

 

final Object value = schema.newMessage();

 

try (final InputStream stream = inputMessage.getBody()) {

ProtobufIOUtil.mergeFrom(stream, value, (Schema<Object>) schema);

 

return value;

}

}

 

throw new HttpMessageNotReadableException(

"Unrecognized HTTP media type " + inputMessage.getHeaders().getContentType().getType() + ".");

}

 

/**

* {@inheritDoc}

*/

@Override

protected boolean supports(final Class<?> clazz) {

// Should not be called, since we override canRead/canWrite.

throw new UnsupportedOperationException();

}

 

/**

* {@inheritDoc}

*/

@Override

protected void writeInternal(final Object o, final HttpOutputMessage outputMessage)

throws IOException, HttpMessageNotWritableException {

if(logger.isDebugEnabled()){

logger.info("Current type: {}", outputMessage.getHeaders().getContentType());

}

Stopwatch stopwatch = Stopwatch.createStarted();

OutputStream stream = null;

 

try {

stream = outputMessage.getBody();

if (MEDIA_TYPE.isCompatibleWith(outputMessage.getHeaders().getContentType())) {

} else if (MEDIA_TYPE_GZIP.isCompatibleWith(outputMessage.getHeaders().getContentType())) {

stream = new GZIPOutputStream(stream);

} else if (MEDIA_TYPE_LZ4.isCompatibleWith(outputMessage.getHeaders().getContentType())) {

stream = new LZ4BlockOutputStream(stream);

} else if (MEDIA_TYPE_SNAPPY.isCompatibleWith(outputMessage.getHeaders().getContentType())) {

stream = new SnappyOutputStream(stream);

} else {

throw new HttpMessageNotWritableException(

"Unrecognized HTTP media type " + outputMessage.getHeaders().getContentType().getType() + ".");

}

 

ProtobufIOUtil.writeTo(stream, o, RuntimeSchema.getSchema((Class<Object>) o.getClass()),

LinkedBuffer.allocate());

stream.flush();

} finally {

IOUtils.closeQuietly(stream);

}

if (logger.isDebugEnabled()){

logger.info("Output spend {}", stopwatch.toString());

}

 

}

}

springmvc配置

 

使用httpclient调用时候使用httpBase.addHeader("Accept","application/x-protobuf"); //

Schema<***Dto> schema = RuntimeSchema.getSchema(***Dto.class);

***Dto value = schema.newMessage();

ProtobufIOUtil.mergeFrom(result1, value, schema);

注意的地方，

1、服务不能直接返回map或者list，如果是map或者list要经过多层封装

2、不能返回带有泛型的实体如：DataGridBean<T> ，在服务器返回的时候没问题，

反序列化成实体时候报错，估计是

Schema<***Dto> schema = RuntimeSchema.getSchema(***Dto.class);

Schema<***Dto<T>> schema = RuntimeSchema.getSchema(***Dto<T>.class);//这里不能这样写，还没找到解决办法，所以只能另外写一个实体，不要使用泛型，把具体类型写进去

后来补充：对于上述问题静态已经解决了,要在声明Schema的时候使用Schema<? extends DataGridBean>

ProtobufIOUtil第三个参数schema强转一下(Schema<DataGridBean<XXXDto>>)schema如下：

// DataGridBean<XXXDto> o = new DataGridBean<XXXDto>();

// Schema<? extends DataGridBean> schema = RuntimeSchema.getSchema(o.getClass());

Schema<? extends DataGridBean> schema1 = RuntimeSchema.getSchema(DataGridBean.class);

DataGridBean<FlightGoccInfoDto> dataGridBean = schema1.newMessage();

ProtobufIOUtil.mergeFrom(result1, dataGridBean, (Schema<DataGridBean<XXXDto>>)schema1);

 

3.比较大小

只针对生产出现问题需要优化的一种常景

protostuff大小约为json的一半

protostuff

json

4.经过nginx gzip压缩后大小比较

经过fiddler抓包后 发现经过gzip压缩后大小相差不是很大

protostuff 52k左右 压缩了将近10倍

json 60k左右 压缩了将近20倍

5.在此过程中遇到到问题

5.1 使用postman调用接口的过程，postman显示的时候解压后的大小，还没找到压缩前的大小

5.2 使用httpclient4.3 返回的response会把返回类型为gzip，x-gzip自动封装成GzipDecompressingEntity

然后remove Content-Encoding 导致使用旧的判断方法时候一直都是进入非gzip的分支

使我一度怀疑nginx的gzip不起效果

最后经过debug查看源码以及使用fildder4抓包的才确定确实有压缩

ResponseContentEncoding.java