C confession on behalf of the mother

Confession on behalf of the mother C█ Micro Signal █: 138 • 0226 • 9370█ █ successful surrogacy package bag pack healthy sex surrogate █ ██ bag boy

A, Zipkin Introduction

What Zipkin that?

 Zipkin's official description: https://zipkin.apache.org/

 Zipkin is an open source distributed real-time data tracking system (Distributed Tracking System), based on Google Dapper papers from design, developed by the Twitter contribution. Its main function is to gather real-time monitoring data from various heterogeneous systems. Distributed tracking system there are other more mature realization, for example: Naver's Pinpoint, Apache's HTrace, Ali Hawkeye Tracing, Jingdong Hydra, Sina Watchman, the US group review of CAT, skywalking and so on.

Why Zipkin?

 As the business more complex, the system also will make various split, especially with the rise of micro-service architecture and container technology, a seemingly simple application, the background may have dozens or even hundreds of service support; front-end may need to request a service call to complete the last times; slow or when a request is not available, we do not know what is causing the back-office services, then we need to figure out how to quickly locate service point of failure, Zipkin distribution tracking system can be a good solution to this problem.

Zipkin some of the basic concepts?

Brave

Brave is a Java program to equip libraries provide equipment capacity for the Standard Servlet, Spring MVC, Http Client, JAX RS, Jersey, Resteasy and MySQL interface, you can simple configuration and code by writing, so based on these frameworks application Builder can report data to Zipkin. Meanwhile Brave also provides a very simple and standardized interfaces, in the above package can not meet the requirements of the time can be easily extended and customized.

Brave below is a configuration diagram of FIG. Brave trace information to send using reporter Zipkin the Collector.

Brave interceptors are mainly using Buried prior request and the request. For example Spingmvc monitor the use Interceptors, Mysql monitor the use statementInterceptors. Similarly Dubbo monitoring is the use of com.alibaba.dubbo.rpc.Filter producers and consumers to filter requests.

traceId

A global request only a traceId. Used to find a few requests for the same link in the flood of requests. For example servlet server receives a user request, dubbo call service, and returns the results to the user, only a whole link traceId. Begins with the user request, the user receives the results at the end.

spanId

A link every request has a spanId. For example a rpc, again there will be a separate sql spanId belonging traceId.

cs

Clent Sent time of the request initiated by the client, such as dubbo calls end before the start of the implementation of remote calls.

cr

Client Receive time the client receives the request processed.

ss

Server Receive server finished processing logic of time.

sr

Server Receive server received a call end time of the request.

sr - cs = 请求在网络上的耗时
ss - sr = 服务端处理请求的耗时
cr - ss = 回应在网络上的耗时
cr - cs = 一次调用的整体耗时

Zipkin working process

 When a user initiates a call, Zipkin the client at the entrance to the entire call link generates a globally unique trace id, and generates a span id distributed for every call on this link. There may be between the span and the span nesting parent-child relationship, representative of the relationship between the downstream distributed call. Between the span and span can be a sibling relationship, the two representatives of the child under the current call is made. A span consisting of a set of trace, trace may be considered to consist of a root node, span several child nodes of a tree.

 Zipkin will trace the relevant information on the call link transfer and asynchronous time-consuming to report information about the current call to Zipkin Server at the end of each call boundary. Zipkin Server trace after receiving the information, it is stored. Zipkin the Web UI will then extracted by means of API trace information will be accessible from out of storage analysis and display.

back to the top

Two, Zipkin deployment and operation

Zipkin's github Address: https://github.com/apache/incubator-zipkin

Docker way

docker run -d -p 9411:9411 openzipkin/zipkin

Jar packet mode (JDK8)

curl -sSL https://zipkin.apache.org/quickstart.sh | bash -s
java -jar zipkin.jar

Note: Zipkin above methods are based on memory storage, data will be lost after the restart Zipkin, it is recommended to use a test environment. Zipkin supported memory types inMemory, MySql, Cassandra, and ElasticsSearch several ways. Environmental official recommended Cassandra and ElasticSearch.

back to the top

Three, Zipkin Dubbo and integration with the Springmvc

Above we build a good Zipkin server, the task now is how to request data generated in our system to submit Zipkin server to render on the UI.

1. pom.xml

        <!-- 使用 okhttp3 作为 reporter -->
        <dependency>
            <groupId>io.zipkin.reporter2</groupId> <artifactId>zipkin-sender-okhttp3</artifactId> <version>2.8.2</version> </dependency> <!-- brave 对 dubbo 的集成 --> <dependency> <groupId>io.zipkin.brave</groupId> <artifactId>brave-instrumentation-dubbo-rpc</artifactId> <version>5.6.3</version> </dependency> <!-- brave 对 mvc 的集成 --> <dependency> <groupId>io.zipkin.brave</groupId> <artifactId>brave-instrumentation-spring-webmvc</artifactId> <version>5.6.3</version> </dependency>

2. application.yml

zipkin:
  url: http://127.0.0.1:9411/api/v2/spans
  connectTimeout: 5000 readTimeout: 10000 # 取样率，指的是多次请求中有百分之多少传到zipkin。例如 1.0 是全部取样，0.5是 50% 取样 rate: 1.0f

3. ZipkinProperties.java

@Configuration
@ConfigurationProperties("zipkin")
public class ZipkinProperties {


    @Value("${spring.application.name}") private String serviceName; private String url; private Long connectTimeout; private Long readTimeout; private Float rate; /*getter and setter*/ }

Note: remember to add @EnableConfigurationProperties comment on the start of class SpringBoot to make @ConfigurationProperties ( "zipkin") entered into force Oh!

4. ZipkinConfig.java

@Configuration
public class ZipkinConfig { @Autowired private ZipkinProperties zipkinProperties; /** * 为了实现 dubbo rpc调用的拦截 * * @return */ @Bean public Tracing tracing() { Sender sender = OkHttpSender.create(zipkinProperties.getUrl()); AsyncReporter reporter = AsyncReporter.builder(sender) .closeTimeout(zipkinProperties.getConnectTimeout(), TimeUnit.MILLISECONDS) .messageTimeout(zipkinProperties.getReadTimeout(), TimeUnit.MILLISECONDS) .build(); Tracing tracing = Tracing.newBuilder() .localServiceName(zipkinProperties.getServiceName()) .propagationFactory(ExtraFieldPropagation.newFactory(B3Propagation.FACTORY, "shiliew")) .sampler(Sampler.create(zipkinProperties.getRate())) .spanReporter(reporter) .build(); return tracing; } /** * MVC Filter，为了实现 SpringMvc 调用的拦截 * @param tracing * @return */ @Bean public Filter tracingFilter(Tracing tracing) { HttpTracing httpTracing = HttpTracing.create(tracing); httpTracing.toBuilder() .serverParser(new HttpServerParser() { @Override public <Req> String spanName(HttpAdapter<Req, ?> adapter, Req req) { return adapter.path(req); } }) .clientParser(new HttpClientParser() { @Override public <Req> String spanName(HttpAdapter<Req, ?> adapter, Req req) { return adapter.path(req); } }).build(); return TracingFilter.create(httpTracing); } }

So, we put Zipkin and Dubbo and Springmvc integrated well:

Stepped pit below, encourage each other:

• Zipkin Server will be sure to generate data after the call, the registration information will not put up services.
• MVC 的拦截，span 的名字是以请求方式命名的，如下：
  

• 如果仍然想查看，某个请求路径的调用情况呢？
  

• github 源代码：https://github.com/JMCuixy/dubbo-demo

 

分类:  服务架构

一、Zipkin 介绍

Zipkin 是什么？

 Zipkin的官方介绍：https://zipkin.apache.org/

 Zipkin是一款开源的分布式实时数据追踪系统（Distributed Tracking System），基于 Google Dapper的论文设计而来，由 Twitter 公司开发贡献。其主要功能是聚集来自各个异构系统的实时监控数据。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。

为什么用 Zipkin？

 随着业务越来越复杂，系统也随之进行各种拆分，特别是随着微服务架构和容器技术的兴起，看似简单的一个应用，后台可能有几十个甚至几百个服务在支撑；一个前端的请求可能需要多次的服务调用最后才能完成；当请求变慢或者不可用时，我们无法得知是哪个后台服务引起的，这时就需要解决如何快速定位服务故障点，Zipkin分布式跟踪系统就能很好的解决这样的问题。

Zipkin的一些基本概念？

Brave

Brave 是用来装备 Java 程序的类库，提供了面向 Standard Servlet、Spring MVC、Http Client、JAX RS、Jersey、Resteasy 和 MySQL 等接口的装备能力，可以通过编写简单的配置和代码，让基于这些框架构建的应用可以向 Zipkin报告数据。同时 Brave 也提供了非常简单且标准化的接口，在以上封装无法满足要求的时候可以方便扩展与定制。

如下图是 Brave 的结构图。Brave 利用 reporter 向 Zipkin的 Collector 发送 trace 信息。

Brave 主要是利用拦截器在请求前和请求后分别埋点。例如 Spingmvc 监控使用 Interceptors，Mysql 监控使用 statementInterceptors。同理 Dubbo 的监控是利用 com.alibaba.dubbo.rpc.Filter 来过滤生产者和消费者的请求。

traceId

一次请求全局只有一个traceId。用来在海量的请求中找到同一链路的几次请求。比如servlet服务器接收到用户请求，调用dubbo服务，然后将结果返回给用户，整条链路只有一个traceId。开始于用户请求，结束于用户收到结果。

spanId

一个链路中每次请求都会有一个spanId。例如一次rpc，一次sql都会有一个单独的spanId从属于traceId。

cs

Clent Sent 客户端发起请求的时间，比如 dubbo 调用端开始执行远程调用之前。

cr

Client Receive 客户端收到处理完请求的时间。

ss

Server Receive 服务端处理完逻辑的时间。

sr

Server Receive 服务端收到调用端请求的时间。

sr - cs = 请求在网络上的耗时
ss - sr = 服务端处理请求的耗时
cr - ss = 回应在网络上的耗时
cr - cs = 一次调用的整体耗时

Zipkin的工作过程

 当用户发起一次调用时，Zipkin 的客户端会在入口处为整条调用链路生成一个全局唯一的 trace id，并为这条链路中的每一次分布式调用生成一个 span id。span 与 span 之间可以有父子嵌套关系，代表分布式调用中的上下游关系。span 和 span 之间可以是兄弟关系，代表当前调用下的两次子调用。一个 trace 由一组 span 组成，可以看成是由 trace 为根节点，span 为若干个子节点的一棵树。

 Zipkin 会将 trace 相关的信息在调用链路上传递，并在每个调用边界结束时异步的把当前调用的耗时信息上报给 Zipkin Server。Zipkin Server 在收到 trace 信息后，将其存储起来。随后 Zipkin 的 Web UI 会通过 API 访问的方式从存储中将 trace 信息提取出来分析并展示。

回到顶部

二、Zipkin的部署与运行

Zipkin的 github 地址：https://github.com/apache/incubator-zipkin

Docker 方式

docker run -d -p 9411:9411 openzipkin/zipkin

Jar 包方式（JDK8）

curl -sSL https://zipkin.apache.org/quickstart.sh | bash -s
java -jar zipkin.jar

注意：以上方式的 Zipkin 都是基于内存存储，Zipkin 重启后数据会丢失，建议测试环境使用。Zipkin 支持的存储类型有 inMemory、MySql、Cassandra、以及 ElasticsSearch 几种方式。正式环境推荐使用 Cassandra 和 ElasticSearch。

回到顶部

三、Zipkin 与 Dubbo 和 Springmvc 的集成

上面我们搭建好了 Zipkin 服务器，现在的任务就是如何把我们系统内产生的请求数据报送给 Zipkin 服务器，以便在 UI 上渲染出来。

1. pom.xml

        <!-- 使用 okhttp3 作为 reporter -->
        <dependency>
            <groupId>io.zipkin.reporter2</groupId> <artifactId>zipkin-sender-okhttp3</artifactId> <version>2.8.2</version> </dependency> <!-- brave 对 dubbo 的集成 --> <dependency> <groupId>io.zipkin.brave</groupId> <artifactId>brave-instrumentation-dubbo-rpc</artifactId> <version>5.6.3</version> </dependency> <!-- brave 对 mvc 的集成 --> <dependency> <groupId>io.zipkin.brave</groupId> <artifactId>brave-instrumentation-spring-webmvc</artifactId> <version>5.6.3</version> </dependency>

2. application.yml

zipkin:
  url: http://127.0.0.1:9411/api/v2/spans
  connectTimeout: 5000 readTimeout: 10000 # 取样率，指的是多次请求中有百分之多少传到zipkin。例如 1.0 是全部取样，0.5是 50% 取样 rate: 1.0f

3. ZipkinProperties.java

@Configuration
@ConfigurationProperties("zipkin")
public class ZipkinProperties {


    @Value("${spring.application.name}") private String serviceName; private String url; private Long connectTimeout; private Long readTimeout; private Float rate; /*getter and setter*/ }

注意：记得在 SpringBoot 的启动类上加上 @EnableConfigurationProperties 注解才能使 @ConfigurationProperties("zipkin") 生效哦！

4. ZipkinConfig.java

@Configuration
public class ZipkinConfig { @Autowired private ZipkinProperties zipkinProperties; /** * 为了实现 dubbo rpc调用的拦截 * * @return */ @Bean public Tracing tracing() { Sender sender = OkHttpSender.create(zipkinProperties.getUrl()); AsyncReporter reporter = AsyncReporter.builder(sender) .closeTimeout(zipkinProperties.getConnectTimeout(), TimeUnit.MILLISECONDS) .messageTimeout(zipkinProperties.getReadTimeout(), TimeUnit.MILLISECONDS) .build(); Tracing tracing = Tracing.newBuilder() .localServiceName(zipkinProperties.getServiceName()) .propagationFactory(ExtraFieldPropagation.newFactory(B3Propagation.FACTORY, "shiliew")) .sampler(Sampler.create(zipkinProperties.getRate())) .spanReporter(reporter) .build(); return tracing; } /** * MVC Filter，为了实现 SpringMvc 调用的拦截 * @param tracing * @return */ @Bean public Filter tracingFilter(Tracing tracing) { HttpTracing httpTracing = HttpTracing.create(tracing); httpTracing.toBuilder() .serverParser(new HttpServerParser() { @Override public <Req> String spanName(HttpAdapter<Req, ?> adapter, Req req) { return adapter.path(req); } }) .clientParser(new HttpClientParser() { @Override public <Req> String spanName(HttpAdapter<Req, ?> adapter, Req req) { return adapter.path(req); } }).build(); return TracingFilter.create(httpTracing); } }

如此，我们就把 Zipkin 和 Dubbo 以及 Springmvc 的集成做好了：

踩过的坑如下，共勉：

• Zipkin Server 一定要在调用后才会产生数据，不会先把服务的信息注册上去。
• MVC 的拦截，span 的名字是以请求方式命名的，如下：
  

• 如果仍然想查看，某个请求路径的调用情况呢？
  

• github 源代码：https://github.com/JMCuixy/dubbo-demo