Dubbo + Zipkin + Brave实现全链路追踪
最近写了一个链路追踪Demo分享下,实现了链路追踪过程中数据的记录,还有能扩展的地方,后期再继续补充。
原理参考上面文章 《Dubbo链路追踪——生成全局ID(traceId)》
实现链路追踪的目的
- 服务调用的流程信息,定位服务调用链
- 记录调用入参及返回值信息,方便问题重现
- 记录调用时间线,代码重构及调优处理
- 调用信息统计
分布式跟踪系统还有其他比较成熟的实现,例如:Naver的Pinpoint、Apache的HTrace、阿里的鹰眼Tracing、京东的Hydra、新浪的Watchman,美团点评的CAT,skywalking等。 本次主要利用Dubbo数据传播特性扩展Filter接口来实现链路追踪的目的
重点主要是zipkin及brave使用及特性,当前brave版本为 5.2.0 为 2018年8月份发布的release版本 , zipkin版本为2.2.1 所需JDK为1.8
快速启动zipkin
下载最新的zipkin并启动
1wget -O zipkin.jar 'https://search.maven.org/remote_content?g=io.zipkin.java&a=zipkin-server&v=LATEST&c=exec' 2java -jar zipkin.jar
输入 http://localhost:9411/zipkin/ 进入WebUI界面如下 
核心源码
代码的初步版本:方便描述
1import brave.Span; 2import brave.Tracer; 3import brave.Tracing; 4import brave.propagation.*; 5import brave.sampler.Sampler; 6import com.alibaba.dubbo.common.Constants; 7import com.alibaba.dubbo.common.extension.Activate; 8import com.alibaba.dubbo.common.json.JSON; 9import com.alibaba.dubbo.common.logger.Logger; 10import com.alibaba.dubbo.common.logger.LoggerFactory; 11import com.alibaba.dubbo.remoting.exchange.ResponseCallback; 12import com.alibaba.dubbo.rpc.*; 13import com.alibaba.dubbo.rpc.protocol.dubbo.FutureAdapter; 14import com.alibaba.dubbo.rpc.support.RpcUtils; 15import zipkin2.codec.SpanBytesEncoder; 16import zipkin2.reporter.AsyncReporter; 17import zipkin2.reporter.Sender; 18import zipkin2.reporter.okhttp3.OkHttpSender; 19 20import java.io.IOException; 21import java.net.InetSocketAddress; 22import java.util.Map; 23import java.util.concurrent.Future; 24import java.util.concurrent.TimeUnit; 25 26/** 27 * Created with IntelliJ IDEA. 28 * 29 * @author: bakerZhu 30 * @description: 31 * @modifytime: 32 */ 33@Activate(group = {Constants.PROVIDER, Constants.CONSUMER}) 34public class TracingFilter implements Filter { 35 36 private static final Logger log = LoggerFactory.getLogger(TracingFilter.class); 37 38 private static Tracing tracing; 39 private static Tracer tracer; 40 private static TraceContext.Extractor<Map<String, String>> extractor; 41 private static TraceContext.Injector<Map<String, String>> injector; 42 43 static final Propagation.Getter<Map<String, String>, String> GETTER = 44 new Propagation.Getter<Map<String, String>, String>() { 45 @Override 46 public String get(Map<String, String> carrier, String key) { 47 return carrier.get(key); 48 } 49 50 @Override 51 public String toString() { 52 return "Map::get"; 53 } 54 }; 55 56 static final Propagation.Setter<Map<String, String>, String> SETTER = 57 new Propagation.Setter<Map<String, String>, String>() { 58 @Override 59 public void put(Map<String, String> carrier, String key, String value) { 60 carrier.put(key, value); 61 } 62 63 @Override 64 public String toString() { 65 return "Map::set"; 66 } 67 }; 68 69 static { 70 // 1 71 Sender sender = OkHttpSender.create("http://localhost:9411/api/v2/spans"); 72 // 2 73 AsyncReporter asyncReporter = AsyncReporter.builder(sender) 74 .closeTimeout(500, TimeUnit.MILLISECONDS) 75 .build(SpanBytesEncoder.JSON_V2); 76 // 3 77 tracing = Tracing.newBuilder() 78 .localServiceName("tracer-client") 79 .spanReporter(asyncReporter) 80 .sampler(Sampler.ALWAYS_SAMPLE) 81 .propagationFactory(ExtraFieldPropagation.newFactory(B3Propagation.FACTORY, "user-name")) 82 .build(); 83 tracer = tracing.tracer(); 84 // 4 85 // 4.1 86 extractor = tracing.propagation().extractor(GETTER); 87 // 4.2 88 injector = tracing.propagation().injector(SETTER); 89 } 90 91 92 93 public TracingFilter() { 94 } 95 96 @Override 97 public Result invoke(Invoker<?> invoker, Invocation invocation) throws RpcException { 98 99 100 RpcContext rpcContext = RpcContext.getContext(); 101 // 5 102 Span.Kind kind = rpcContext.isProviderSide() ? Span.Kind.SERVER : Span.Kind.CLIENT; 103 final Span span; 104 if (kind.equals(Span.Kind.CLIENT)) { 105 //6 106 span = tracer.nextSpan(); 107 //7 108 injector.inject(span.context(), invocation.getAttachments()); 109 } else { 110 //8 111 TraceContextOrSamplingFlags extracted = extractor.extract(invocation.getAttachments()); 112 //9 113 span = extracted.context() != null ? tracer.joinSpan(extracted.context()) : tracer.nextSpan(extracted); 114 } 115 116 if (!span.isNoop()) { 117 span.kind(kind).start(); 118 //10 119 String service = invoker.getInterface().getSimpleName(); 120 String method = RpcUtils.getMethodName(invocation); 121 span.kind(kind); 122 span.name(service + "/" + method); 123 InetSocketAddress remoteAddress = rpcContext.getRemoteAddress(); 124 span.remoteIpAndPort( 125 remoteAddress.getAddress() != null ? remoteAddress.getAddress().getHostAddress() : remoteAddress.getHostName(),remoteAddress.getPort()); 126 } 127 128 boolean isOneway = false, deferFinish = false; 129 try (Tracer.SpanInScope scope = tracer.withSpanInScope(span)){ 130 //11 131 collectArguments(invocation, span, kind); 132 Result result = invoker.invoke(invocation); 133 134 if (result.hasException()) { 135 onError(result.getException(), span); 136 } 137 // 12 138 isOneway = RpcUtils.isOneway(invoker.getUrl(), invocation); 139 // 13 140 Future<Object> future = rpcContext.getFuture(); 141 142 if (future instanceof FutureAdapter) { 143 deferFinish = true; 144 ((FutureAdapter) future).getFuture().setCallback(new FinishSpanCallback(span));// 14 145 } 146 return result; 147 } catch (Error | RuntimeException e) { 148 onError(e, span); 149 throw e; 150 } finally { 151 if (isOneway) { // 15 152 span.flush(); 153 } else if (!deferFinish) { // 16 154 span.finish(); 155 } 156 } 157 } 158 159 static void onError(Throwable error, Span span) { 160 span.error(error); 161 if (error instanceof RpcException) { 162 span.tag("dubbo.error_msg", RpcExceptionEnum.getMsgByCode(((RpcException) error).getCode())); 163 } 164 } 165 166 static void collectArguments(Invocation invocation, Span span, Span.Kind kind) { 167 if (kind == Span.Kind.CLIENT) { 168 StringBuilder fqcn = new StringBuilder(); 169 Object[] args = invocation.getArguments(); 170 if (args != null && args.length > 0) { 171 try { 172 fqcn.append(JSON.json(args)); 173 } catch (IOException e) { 174 log.warn(e.getMessage(), e); 175 } 176 } 177 span.tag("args", fqcn.toString()); 178 } 179 } 180 181 182 183 static final class FinishSpanCallback implements ResponseCallback { 184 final Span span; 185 186 FinishSpanCallback(Span span) { 187 this.span = span; 188 } 189 190 @Override 191 public void done(Object response) { 192 span.finish(); 193 } 194 195 @Override 196 public void caught(Throwable exception) { 197 onError(exception, span); 198 span.finish(); 199 } 200 } 201 // 17 202 private enum RpcExceptionEnum { 203 UNKNOWN_EXCEPTION(0, "unknown exception"), 204 NETWORK_EXCEPTION(1, "network exception"), 205 TIMEOUT_EXCEPTION(2, "timeout exception"), 206 BIZ_EXCEPTION(3, "biz exception"), 207 FORBIDDEN_EXCEPTION(4, "forbidden exception"), 208 SERIALIZATION_EXCEPTION(5, "serialization exception"),; 209 210 private int code; 211 212 private String msg; 213 214 RpcExceptionEnum(int code, String msg) { 215 this.code = code; 216 this.msg = msg; 217 } 218 219 public static String getMsgByCode(int code) { 220 for (RpcExceptionEnum error : RpcExceptionEnum.values()) { 221 if (code == error.code) { 222 return error.msg; 223 } 224 } 225 return null; 226 } 227 } 228}
- 构建客户端发送工具
- 构建异步reporter
- 构建tracing上下文
- 初始化injector 和 Extractor [tab]4.1 extractor 指数据提取对象,用于在carrier中提取TraceContext相关信息或者采样标记信息到TraceContextOrSamplingFlags 中 -4.2 injector 用于将TraceContext中的各种数据注入到carrier中,其中carrier一半是指数据传输中的载体,类似于Dubbo中Invocation中的attachment(附件集合)
- 判断此次调用是作为服务端还是客户端
- rpc客户端调用会从ThreadLocal中获取parent的 TraceContext ,为新生成的Span指定traceId及 parentId如果没有parent traceContext 则生成的Span为 root span
- 将Span绑定的TraceContext中 属性信息 Copy 到 Invocation中达到远程参数传递的作用
- rpc服务提供端 , 从invocation中提取TraceContext相关信息及采样数据信息
- 生成span , 兼容初次服务端调用
- 记录接口信息及远程IP Port
- 将创建的Span 作为当前Span (可以通过Tracer.currentSpan 访问到它) 并设置查询范围
- oneway调用即只请求不接受结果
- 如果future不为空则为 async 调用 在回调中finish span
- 设置异步回调,回调代码执行span finish() .
- oneway调用 因为不需等待返回值 即没有 cr (Client Receive) 需手动flush()
- 同步调用 业务代码执行完毕后需手动finish()
- 设置枚举类 与 Dubbo中RpcException保持对应
测试项
- Dubbo sync async oneway 测试
- RPC异常测试
- 普通业务异常测试
- 并发测试
配置方式
POM依赖添加
1<dependency> 2 <groupId>com.github.baker</groupId> 3 <artifactId>Tracing</artifactId> 4 <version>1.0-SNAPSHOT</version> 5</dependency>
资源目录根路径下添加tracing.properties文件
一次调用信息
调用链
调用成功失败汇总
zipkinHost 指定zipkin服务器IP:PORT 默认为localhost:9411 serviceName 指定应用名称 默认为trace-default
调用链: 
待扩展项
- 抽象数据传输(扩展Kafka数据传输)
- 调用返回值数据打印
- 更灵活的配置方式
赞赏支持
