全链路spring cloud sleuth+zipkin

一、About ZipKin

please google

二、 Demo Scene

三、 Result Display

四、Prepare

1、soft version

kafka：2.10-0.10.2.0
zokeeper：3.4.10
elasticsearch：5.2.2
jdk：1.8
spring boot：1.5.3.RELEASE
sprign cloud：Dalston.RELEASE
rabbit mq：3.6.9

2、install

kafka+zookeeper
elasticsearch
rabbit mq
mysql

3、create four spring cloud project

web-api、user-api、order-api、zipkin
ps:

why i will create zipkin project use spring boot by myself not use  zipkin.jar from http://zipkin.io/,actually，zipkin.jar is a spring boot project，check it's dependency lib you will find it din't use spring-cloud-sleuth-stream,but i will send trace info to kafka for zipkin server collector ,so i must use spring-cloud-sleuth-stream in my service

and the message send to kafka is a  sleuth.span object and use kafka default serialized，but zipkin.jar only receive zipkin.span and json or thrift encode,so it‘s not matching,That's the reason i create zipkin server

but if you use rabbit mq，that's no problem.

4、configuration

4.1、the service web-api、user-api、order-api config part like:
pom.xml

<dependency>    <groupId>org.springframework.cloud</groupId>    <artifactId>spring-cloud-sleuth-zipkin-stream</artifactId></dependency><dependency>    <groupId>org.springframework.cloud</groupId>    <artifactId>spring-cloud-starter-stream-kafka</artifactId></dependency>

application.properties

spring.sleuth.sampler.percentage=1.0spring.cloud.stream.kafka.binder.brokers=10.20.1.11:9092,10.20.1.12:9092spring.cloud.stream.kafka.binder.zkNodes=10.20.1.11:2181,10.20.1.12:2181

4.2、the zipkinconfig part like:

pom.xml

<!-- the  first one dependency below，In principle, there is no need，beause sleuth-zipkin-stream 1.5.3 will  Introduce zipkin version1.19 Automaticly，but 1.19 only support  elasticsearch version 2.X --><dependency><groupId>io.zipkin.java</groupId><artifactId>zipkin</artifactId><version>1.24.0</version></dependency><dependency><groupId>org.springframework.cloud</groupId><artifactId>spring-cloud-sleuth-zipkin-stream</artifactId></dependency><dependency><groupId>org.springframework.cloud</groupId><artifactId>spring-cloud-starter-stream-kafka</artifactId></dependency><dependency><groupId>io.zipkin.java</groupId><artifactId>zipkin-autoconfigure-storage-elasticsearch-http</artifactId><version>1.24.0</version><optional>true</optional></dependency>

application.properties

#kafka configspring.sleuth.enabled=falsespring.sleuth.sampler.percentage=1.0spring.cloud.stream.kafka.binder.brokers=10.20.1.11:9092,10.20.1.12:9092spring.cloud.stream.kafka.binder.zkNodes=10.20.1.11:2181,10.20.1.12:2181#elasticsearch configzipkin.storage.type=elasticsearchzipkin.storage.elasticsearch.hosts=10.20.1.11:9200,10.20.1.12:9200zipkin.storage.elasticsearch.cluster=elasticsearchzipkin.storage.elasticsearch.index=zipkinzipkin.storage.elasticsearch.index-shards=5zipkin.storage.elasticsearch.index-replicas=1

ZipKin Server Startup class configuration

@SpringBootApplication//@EnableZipkinServer //this is used by interface receive trace info@EnableZipkinStreamServer //can be used kafka,rabbitpublic class ZkingApplication {    public static void main(String[] args) {        SpringApplication.run(ZkingApplication.class, args);    }}

五、Demo DownLoad

click me

by the way,spring cloud is a pretty boy,i like its combination of terseness and elegance

六、补充

如果kafka没有启动，spring boot会启动失败，这个异常处理设计的真是缺德

/**
 * 1、修改背景
 * 因kafka节点没有启动 在spring boot启动时初始化outputBindingLifecycle、inputBindingLifecycle
 * 两个bean时候连接kafka失败，向外抛出了异常直到EmbeddedWebApplicationContext类
 * 捕获处理，处理方式为：stopAndReleaseEmbeddedServletContainer()导致整个应用停止启动
 * 2、修改方案
 * 干预上面两个bean的初始化，在连接kafka异常时，将异常处理掉，不向上层抛出
 * 3、修改步骤
 * 3.1、使用自定义MyBindingLifecycle的bean将BindingServiceConfiguration中的两个bean初始化替换掉
 * 3.2、在自定bean中启动线程MyBindingThread来控制两个bean的初始化
 * 4、解决启动问题之后，实际上kafka还是没有连接的，此时向kafka发送span时会失败，默认的处理方案是捕获到异常之后使用
 *   handleError处理，再次发送新的span，这就导致循环发送
 *   参见：ErrorHandlingTaskExecutor中的execute方法
 *   catch (Throwable t)
 *   {
 *        ErrorHandlingTaskExecutor.this.errorHandler.handleError(t);
 *     }
 * 5、解决方案
 * 重写ErrorHandler的handleError方法
 * 6、跟踪代码发现
 * 跟踪发现ErrorHandler对线对象是在SourcePollingChannelAdapterFactoryBean初始化时候设置的
 * spca.setErrorHandler(this.pollerMetadata.getErrorHandler());
 * 进一步发现是在pollerMetadata对象中，所以需要在pollerMetadata对象初始化时候做修改
 * 7、修改步骤
 * 自定义MyPollerMetadata且需要@Configuration，重写handleError方法如下
 * @author zhangdingxin、yangxi
 */