zipkin

zipkin为分布式链路调用监控系统，聚合各业务系统调用延迟数据，达到链路调用监控跟踪。

architecture

如图，在复杂的调用链路中假设存在一条调用链路响应缓慢，如何定位其中延迟高的服务呢？

• 日志： 通过分析调用链路上的每个服务日志得到结果
• zipkin：使用zipkin的web UI可以一眼看出延迟高的服务

如图所示，各业务系统在彼此调用时，将特定的跟踪消息传递至zipkin,zipkin在收集到跟踪信息后将其聚合处理、存储、展示等，用户可通过web UI方便 获得网络延迟、调用链路、系统依赖等等。

zipkin主要涉及四个组件 collector storage search web UI

• Collector接收各service传输的数据
• Cassandra作为Storage的一种，也可以是mysql等，默认存储在内存中，配置cassandra可以参考这里
• Query负责查询Storage中存储的数据,提供简单的JSON API获取数据，主要提供给web UI使用
• Web 提供简单的web界面

install

执行如下命令下载jar包

wget -O zipkin.jar 'https://search.maven.org/remote_content?g=io.zipkin.java&a=zipkin-server&v=LATEST&c=exec'

其为一个spring boot功能，直接运行jar

nohup java -jar zipkin.jar & 

访问 http://ip:9411

terminology

使用zipkin涉及几个概念

• Span:基本工作单元，一次链路调用(可以是RPC，DB等没有特定的限制)创建一个span，通过一个64位ID标识它， span通过还有其他的数据，例如描述信息，时间戳，key-value对的(Annotation)tag信息，parent-id等,其中parent-id 可以表示span调用链路来源，通俗的理解span就是一次请求信息

• Trace:类似于树结构的Span集合，表示一条调用链路，存在唯一标识

• Annotation: 注解,用来记录请求特定事件相关信息(例如时间)，通常包含四个注解信息

  cs - Client Start,表示客户端发起请求

  sr - Server Receive,表示服务端收到请求

  ss - Server Send,表示服务端完成处理，并将结果发送给客户端

  cr - Client Received,表示客户端获取到服务端返回信息

• BinaryAnnotation:提供一些额外信息，一般已key-value对出现

概念说完，来看下完整的调用链路

上图表示一请求链路，一条链路通过Trace Id唯一标识，Span标识发起的请求信息，各span通过parent id 关联起来，如图

整个链路的依赖关系如下:

完成链路调用的记录后，如何来计算调用的延迟呢，这就需要利用Annotation信息

sr-cs 得到请求发出延迟

ss-sr 得到服务端处理延迟

cr-cs 得到真个链路完成延迟

brave

作为各调用链路，只需要负责将指定格式的数据发送给zipkin即可，利用brave可快捷完成操作。

首先导入jar包pom.xml

<parent>
        <groupId>org.springframework.boot</groupId>
        <artifactId>spring-boot-starter-parent</artifactId>
        <version>1.3.6.RELEASE</version>
    </parent>



    <!-- https://mvnrepository.com/artifact/io.zipkin.brave/brave-core -->
    <dependencies>

        <dependency>
            <groupId>org.springframework.boot</groupId>
            <artifactId>spring-boot-starter-web</artifactId>
        </dependency>
        <dependency>
            <groupId>org.springframework.boot</groupId>
            <artifactId>spring-boot-starter-aop</artifactId>
        </dependency>

        <dependency>
            <groupId>org.springframework.boot</groupId>
            <artifactId>spring-boot-starter-actuator</artifactId>
        </dependency>

        <dependency>
            <groupId>io.zipkin.brave</groupId>
            <artifactId>brave-core</artifactId>
            <version>3.9.0</version>
        </dependency>
        <!-- https://mvnrepository.com/artifact/io.zipkin.brave/brave-http -->
        <dependency>
            <groupId>io.zipkin.brave</groupId>
            <artifactId>brave-http</artifactId>
            <version>3.9.0</version>
        </dependency>
        <dependency>
            <groupId>io.zipkin.brave</groupId>
            <artifactId>brave-spancollector-http</artifactId>
            <version>3.9.0</version>
        </dependency>
        <dependency>
            <groupId>io.zipkin.brave</groupId>
            <artifactId>brave-web-servlet-filter</artifactId>
            <version>3.9.0</version>
        </dependency>

        <dependency>
            <groupId>io.zipkin.brave</groupId>
            <artifactId>brave-okhttp</artifactId>
            <version>3.9.0</version>
        </dependency>

        <!-- https://mvnrepository.com/artifact/org.slf4j/slf4j-api -->
        <dependency>
            <groupId>org.slf4j</groupId>
            <artifactId>slf4j-api</artifactId>
            <version>1.7.13</version>
        </dependency>
        <dependency>
            <groupId>org.apache.httpcomponents</groupId>
            <artifactId>httpclient</artifactId>
            <version>4.5.1</version>
        </dependency>

    </dependencies>

利用spring boot创建工程

Application.java

package com.lkl.zipkin;

import org.springframework.boot.SpringApplication;
import org.springframework.boot.autoconfigure.SpringBootApplication;

/**
 *
 * Created by liaokailin on 16/7/27.
 */
@SpringBootApplication
public class Application {


    public static void main(String[] args) {
        SpringApplication app = new SpringApplication(Application.class);
        app.run(args);


    }
}

建立controller对外提供服务

HomeController.java

RestController
@RequestMapping("/")
public class HomeController {

    @Autowired
    private OkHttpClient client;

    private  Random random = new Random();

    @RequestMapping("start")
    public String start() throws InterruptedException, IOException {
        int sleep= random.nextInt(100);
        TimeUnit.MILLISECONDS.sleep(sleep);
        Request request = new Request.Builder().url("http://localhost:9090/foo").get().build();
        Response response = client.newCall(request).execute();
        return " [service1 sleep " + sleep+" ms]" + response.body().toString();
    }

HomeController中利用OkHttpClient调用发起http请求。在每次发起请求时则需要通过brave记录Span信息，并异步传递给zipkin 作为被调用方(服务端)也同样需要完成以上操作.

ZipkinConfig.java

package com.lkl.zipkin.config;

import com.github.kristofa.brave.Brave;
import com.github.kristofa.brave.EmptySpanCollectorMetricsHandler;
import com.github.kristofa.brave.SpanCollector;
import com.github.kristofa.brave.http.DefaultSpanNameProvider;
import com.github.kristofa.brave.http.HttpSpanCollector;
import com.github.kristofa.brave.okhttp.BraveOkHttpRequestResponseInterceptor;
import com.github.kristofa.brave.servlet.BraveServletFilter;
import okhttp3.OkHttpClient;
import org.springframework.beans.factory.annotation.Autowired;
import org.springframework.context.annotation.Bean;
import org.springframework.context.annotation.Configuration;

/**
 * Created by liaokailin on 16/7/27.
 */
@Configuration
public class ZipkinConfig {

    @Autowired
    private ZipkinProperties properties;


    @Bean
    public SpanCollector spanCollector() {
        HttpSpanCollector.Config config = HttpSpanCollector.Config.builder().connectTimeout(properties.getConnectTimeout()).readTimeout(properties.getReadTimeout())
                .compressionEnabled(properties.isCompressionEnabled()).flushInterval(properties.getFlushInterval()).build();
        return HttpSpanCollector.create(properties.getUrl(), config, new EmptySpanCollectorMetricsHandler());
    }


    @Bean
    public Brave brave(SpanCollector spanCollector){
        Brave.Builder builder = new Brave.Builder(properties.getServiceName());  //指定state
        builder.spanCollector(spanCollector);
        builder.traceSampler(Sampler.ALWAYS_SAMPLE);
        Brave brave = builder.build();
        return brave;
    }

    @Bean
    public BraveServletFilter braveServletFilter(Brave brave){
        BraveServletFilter filter = new BraveServletFilter(brave.serverRequestInterceptor(),brave.serverResponseInterceptor(),new DefaultSpanNameProvider());
        return filter;
    }

    @Bean
    public OkHttpClient okHttpClient(Brave brave){
        OkHttpClient client = new OkHttpClient.Builder()
                .addInterceptor(new BraveOkHttpRequestResponseInterceptor(brave.clientRequestInterceptor(), brave.clientResponseInterceptor(), new DefaultSpanNameProvider()))
                .build();
        return client;
    }
}

• SpanCollector 配置收集器

• Brave 各工具类的封装,其中builder.traceSampler(Sampler.ALWAYS_SAMPLE)设置采样比率，0-1之间的百分比

• BraveServletFilter 作为拦截器，需要serverRequestInterceptor,serverResponseInterceptor 分别完成sr和ss操作

• OkHttpClient 添加拦截器，需要clientRequestInterceptor,clientResponseInterceptor 分别完成cs和cr操作,该功能由 brave中的brave-okhttp模块提供，同样的道理如果需要记录数据库的延迟只要在数据库操作前后完成cs和cr即可，当然brave提供其封装。

以上还缺少一个配置信息ZipkinProperties.java

package com.lkl.zipkin.config;

import org.springframework.boot.context.properties.ConfigurationProperties;
import org.springframework.context.annotation.Configuration;

/**
 * Created by liaokailin on 16/7/28.
 */
@Configuration
@ConfigurationProperties(prefix = "com.zipkin")
public class ZipkinProperties {

    private String serviceName;

    private String url;

    private int connectTimeout;

    private int readTimeout;

    private int flushInterval;

    private boolean compressionEnabled;

    public String getUrl() {
        return url;
    }

    public void setUrl(String url) {
        this.url = url;
    }

    public int getConnectTimeout() {
        return connectTimeout;
    }

    public void setConnectTimeout(int connectTimeout) {
        this.connectTimeout = connectTimeout;
    }

    public int getReadTimeout() {
        return readTimeout;
    }

    public void setReadTimeout(int readTimeout) {
        this.readTimeout = readTimeout;
    }

    public int getFlushInterval() {
        return flushInterval;
    }

    public void setFlushInterval(int flushInterval) {
        this.flushInterval = flushInterval;
    }

    public boolean isCompressionEnabled() {
        return compressionEnabled;
    }

    public void setCompressionEnabled(boolean compressionEnabled) {
        this.compressionEnabled = compressionEnabled;
    }

    public String getServiceName() {
        return serviceName;
    }

    public void setServiceName(String serviceName) {
        this.serviceName = serviceName;
    }
}

则可以在配置文件application.properties中配置相关信息

com.zipkin.serviceName=service1
com.zipkin.url=http://110.173.14.57:9411
com.zipkin.connectTimeout=6000
com.zipkin.readTimeout=6000
com.zipkin.flushInterval=1
com.zipkin.compressionEnabled=true
server.port=8080

那么其中的service1即完成，同样的道理，修改配置文件(调整com.zipkin.serviceName,以及server.port)以及controller对应的方法构造若干服务

service1 中访问http://localhost:8080/start需要访问http://localhost:9090/foo,则构造server2提供该方法

server2配置

com.zipkin.serviceName=service2
com.zipkin.url=http://110.173.14.57:9411
com.zipkin.connectTimeout=6000
com.zipkin.readTimeout=6000
com.zipkin.flushInterval=1
com.zipkin.compressionEnabled=true


server.port=9090

controller方法

    @RequestMapping("foo")
    public String foo() throws InterruptedException, IOException {
        Random random = new Random();
        int sleep= random.nextInt(100);
        TimeUnit.MILLISECONDS.sleep(sleep);
        Request request = new Request.Builder().url("http://localhost:9091/bar").get().build();  //service3
        Response response = client.newCall(request).execute();
        String result = response.body().string();
        request = new Request.Builder().url("http://localhost:9092/tar").get().build();  //service4
        response = client.newCall(request).execute();
       result += response.body().string();
        return " [service2 sleep " + sleep+" ms]" + result;
    }

在server2中调用server3和server4中的方法

方法分别为

 @RequestMapping("bar")
    public String bar() throws InterruptedException, IOException {  //service3 method
        Random random = new Random();
        int sleep= random.nextInt(100);
        TimeUnit.MILLISECONDS.sleep(sleep);
        return " [service3 sleep " + sleep+" ms]";
    }

    @RequestMapping("tar")
    public String tar() throws InterruptedException, IOException { //service4 method
        Random random = new Random();
        int sleep= random.nextInt(1000);
        TimeUnit.MILLISECONDS.sleep(sleep);
        return " [service4 sleep " + sleep+" ms]";
    }

将工程修改后编译成jar形式

执行

nohup java -jar server4.jar &
nohup java -jar server3.jar &
nohup java -jar server2.jar &
nohup java -jar server1.jar &

访问http://localhost:8080/start后查看zipkin的web UI

点击条目可以查看具体的延迟信息

服务之间的依赖为

brave 源码

以上完成了基本的操作，下面将从源码角度来看下brave的实现

首先从SpanCollector来入手

 @Bean
    public SpanCollector spanCollector() {
        HttpSpanCollector.Config config = HttpSpanCollector.Config.builder().connectTimeout(properties.getConnectTimeout()).readTimeout(properties.getReadTimeout())
                .compressionEnabled(properties.isCompressionEnabled()).flushInterval(properties.getFlushInterval()).build();
        return HttpSpanCollector.create(properties.getUrl(), config, new EmptySpanCollectorMetricsHandler());
    }

从名称上看HttpSpanCollector是基于http的span收集器,因此超时配置是必须的，默认给出的超时时间较长，flushInterval表示span的传递 间隔，实际为定时任务执行的间隔时间.在HttpSpanCollector中覆写了父类方法sendSpans

@Override
  protected void sendSpans(byte[] json) throws IOException {
    // intentionally not closing the connection, so as to use keep-alives
    HttpURLConnection connection = (HttpURLConnection) new URL(url).openConnection();
    connection.setConnectTimeout(config.connectTimeout());
    connection.setReadTimeout(config.readTimeout());
    connection.setRequestMethod("POST");
    connection.addRequestProperty("Content-Type", "application/json");
    if (config.compressionEnabled()) {
      connection.addRequestProperty("Content-Encoding", "gzip");
      ByteArrayOutputStream gzipped = new ByteArrayOutputStream();
      try (GZIPOutputStream compressor = new GZIPOutputStream(gzipped)) {
        compressor.write(json);
      }
      json = gzipped.toByteArray();
    }
    connection.setDoOutput(true);
    connection.setFixedLengthStreamingMode(json.length);
    connection.getOutputStream().write(json);

    try (InputStream in = connection.getInputStream()) {
      while (in.read() != -1) ; // skip
    } catch (IOException e) {
      try (InputStream err = connection.getErrorStream()) {
        if (err != null) { // possible, if the connection was dropped
          while (err.read() != -1) ; // skip
        }
      }
      throw e;
    }
  }
}

可以看出最终span信息是通过HttpURLConnection实现的，同样道理就可以推理brave对brave-spring-resttemplate-interceptors模块的实现， 只是换了一种http封装。

Brave

 @Bean
    public Brave brave(SpanCollector spanCollector){
        Brave.Builder builder = new Brave.Builder(properties.getServiceName());  //指定state
        builder.spanCollector(spanCollector);
        builder.traceSampler(Sampler.ALWAYS_SAMPLE);
        Brave brave = builder.build();
        return brave;
    }

Brave类包装了各种工具类

public Brave build() {
            return new Brave(this);
        }

创建一个Brave

private Brave(Builder builder) {
        serverTracer = ServerTracer.builder()
                .randomGenerator(builder.random)
                .spanCollector(builder.spanCollector)
                .state(builder.state)
                .traceSampler(builder.sampler).build();

        clientTracer = ClientTracer.builder()
                .randomGenerator(builder.random)
                .spanCollector(builder.spanCollector)
                .state(builder.state)
                .traceSampler(builder.sampler).build();

        localTracer = LocalTracer.builder()
                .randomGenerator(builder.random)
                .spanCollector(builder.spanCollector)
                .spanAndEndpoint(SpanAndEndpoint.LocalSpanAndEndpoint.create(builder.state))
                .traceSampler(builder.sampler).build();
        
        serverRequestInterceptor = new ServerRequestInterceptor(serverTracer);
        serverResponseInterceptor = new ServerResponseInterceptor(serverTracer);
        clientRequestInterceptor = new ClientRequestInterceptor(clientTracer);
        clientResponseInterceptor = new ClientResponseInterceptor(clientTracer);
        serverSpanAnnotationSubmitter = AnnotationSubmitter.create(SpanAndEndpoint.ServerSpanAndEndpoint.create(builder.state));
        serverSpanThreadBinder = new ServerSpanThreadBinder(builder.state);
        clientSpanThreadBinder = new ClientSpanThreadBinder(builder.state);
    }

封装了*Tracer,*Interceptor,*Binder等

其中 serverTracer当服务作为服务端时处理span信息，clientTracer当服务作为客户端时处理span信息

Filter

BraveServletFilter是http模块提供的拦截器功能，传递serverRequestInterceptor,serverResponseInterceptor,spanNameProvider等参数 其中spanNameProvider表示如何处理span的名称，默认使用method名称,spring boot中申明的filter bean 默认拦截所有请求

@Override
    public void doFilter(ServletRequest request, ServletResponse response, FilterChain filterChain) throws IOException, ServletException {

        String alreadyFilteredAttributeName = getAlreadyFilteredAttributeName();
        boolean hasAlreadyFilteredAttribute = request.getAttribute(alreadyFilteredAttributeName) != null;

        if (hasAlreadyFilteredAttribute) {
            // Proceed without invoking this filter...
            filterChain.doFilter(request, response);
        } else {

            final StatusExposingServletResponse statusExposingServletResponse = new StatusExposingServletResponse((HttpServletResponse) response);
            requestInterceptor.handle(new HttpServerRequestAdapter(new ServletHttpServerRequest((HttpServletRequest) request), spanNameProvider));

            try {
                filterChain.doFilter(request, statusExposingServletResponse);
            } finally {
                responseInterceptor.handle(new HttpServerResponseAdapter(new HttpResponse() {
                    @Override
                    public int getHttpStatusCode() {
                        return statusExposingServletResponse.getStatus();
                    }
                }));
            }
        }
    }

首先来看requestInterceptor.handle方法，

 public void handle(ServerRequestAdapter adapter) {
        serverTracer.clearCurrentSpan();
        final TraceData traceData = adapter.getTraceData();

        Boolean sample = traceData.getSample();
        if (sample != null && Boolean.FALSE.equals(sample)) {
            serverTracer.setStateNoTracing();
            LOGGER.fine("Received indication that we should NOT trace.");
        } else {
            if (traceData.getSpanId() != null) {
                LOGGER.fine("Received span information as part of request.");
                SpanId spanId = traceData.getSpanId();
                serverTracer.setStateCurrentTrace(spanId.traceId, spanId.spanId,
                        spanId.nullableParentId(), adapter.getSpanName());
            } else {
                LOGGER.fine("Received no span state.");
                serverTracer.setStateUnknown(adapter.getSpanName());
            }
            serverTracer.setServerReceived();
            for(KeyValueAnnotation annotation : adapter.requestAnnotations())
            {
                serverTracer.submitBinaryAnnotation(annotation.getKey(), annotation.getValue());
            }
        }
    }

其中serverTracer.clearCurrentSpan()清除当前线程上的span信息，调用ThreadLocalServerClientAndLocalSpanState中的

  @Override
    public void setCurrentServerSpan(final ServerSpan span) {
        if (span == null) {
            currentServerSpan.remove();
        } else {
            currentServerSpan.set(span);
        }
    }

currentServerSpan为ThreadLocal对象

private final static ThreadLocal<ServerSpan> currentServerSpan = new ThreadLocal<ServerSpan>() {

回到ServerRequestInterceptor#handle()方法中final TraceData traceData = adapter.getTraceData()

 @Override
    public TraceData getTraceData() {
        final String sampled = serverRequest.getHttpHeaderValue(BraveHttpHeaders.Sampled.getName());
        if (sampled != null) {
            if (sampled.equals("0") || sampled.toLowerCase().equals("false")) {
                return TraceData.builder().sample(false).build();
            } else {
                final String parentSpanId = serverRequest.getHttpHeaderValue(BraveHttpHeaders.ParentSpanId.getName());
                final String traceId = serverRequest.getHttpHeaderValue(BraveHttpHeaders.TraceId.getName());
                final String spanId = serverRequest.getHttpHeaderValue(BraveHttpHeaders.SpanId.getName());

                if (traceId != null && spanId != null) {
                    SpanId span = getSpanId(traceId, spanId, parentSpanId);
                    return TraceData.builder().sample(true).spanId(span).build();
                }
            }
        }
        return TraceData.builder().build();
    }

其中 SpanId span = getSpanId(traceId, spanId, parentSpanId) 将构造一个SpanId对象

 private SpanId getSpanId(String traceId, String spanId, String parentSpanId) {
        return SpanId.builder()
            .traceId(convertToLong(traceId))
            .spanId(convertToLong(spanId))
            .parentId(parentSpanId == null ? null : convertToLong(parentSpanId)).build();
   }

将traceId,spanId,parentId关联起来，其中设置parentId方法为

public Builder parentId(@Nullable Long parentId) {
      if (parentId == null) {
        this.flags |= FLAG_IS_ROOT;
      } else {
        this.flags &= ~FLAG_IS_ROOT;
      }
      this.parentId = parentId;
      return this;
    }

如果parentId为空为根节点，则执行this.flags |= FLAG_IS_ROOT ,因此后续在判断节点是否为根节点时，只需要执行(flags & FLAG_IS_ROOT) == FLAG_IS_ROOT即可.

构造完SpanId后看

    serverTracer.setStateCurrentTrace(spanId.traceId, spanId.spanId,
                        spanId.nullableParentId(), adapter.getSpanName());

设置当前Span

 public void setStateCurrentTrace(long traceId, long spanId, @Nullable Long parentSpanId, @Nullable String name) {
        checkNotBlank(name, "Null or blank span name");
        spanAndEndpoint().state().setCurrentServerSpan(
            ServerSpan.create(traceId, spanId, parentSpanId, name));
    }

ServerSpan.create创建Span信息

 static ServerSpan create(long traceId, long spanId, @Nullable Long parentSpanId, String name) {
        Span span = new Span();
        span.setTrace_id(traceId);
        span.setId(spanId);
        if (parentSpanId != null) {
            span.setParent_id(parentSpanId);
        }
        span.setName(name);
        return create(span, true);
    }

构造了一个包含Span信息的AutoValue_ServerSpan对象

通过setCurrentServerSpan设置到当前线程上

继续看serverTracer.setServerReceived()方法

public void setServerReceived() {
        submitStartAnnotation(zipkinCoreConstants.SERVER_RECV);
    }

为当前请求设置了server received event

void submitStartAnnotation(String annotationName) {
        Span span = spanAndEndpoint().span();
        if (span != null) {
            Annotation annotation = Annotation.create(
                currentTimeMicroseconds(),
                annotationName,
                spanAndEndpoint().endpoint()
            );
            synchronized (span) {
                span.setTimestamp(annotation.timestamp);
                span.addToAnnotations(annotation);
            }
        }
    }

在这里为Span信息设置了Annotation信息,后续的

 for(KeyValueAnnotation annotation : adapter.requestAnnotations())
            {
                serverTracer.submitBinaryAnnotation(annotation.getKey(), annotation.getValue());
            }

设置了BinaryAnnotation信息，adapter.requestAnnotations()在构造HttpServerRequestAdapter时已完成

 @Override
    public Collection<KeyValueAnnotation> requestAnnotations() {
        KeyValueAnnotation uriAnnotation = KeyValueAnnotation.create(
                TraceKeys.HTTP_URL, serverRequest.getUri().toString());
        return Collections.singleton(uriAnnotation);
    }

以上将Span信息(包括sr)存储在当前线程中，接下来继续看BraveServletFilter#doFilter方法的finally部分

 responseInterceptor.handle(new HttpServerResponseAdapter(new HttpResponse() {
                    @Override  //获取http状态码
                    public int getHttpStatusCode() {
                        return statusExposingServletResponse.getStatus();
                    }
                }));

handle方法

 public void handle(ServerResponseAdapter adapter) {
        // We can submit this in any case. When server state is not set or
        // we should not trace this request nothing will happen.
        LOGGER.fine("Sending server send.");
        try {
            for(KeyValueAnnotation annotation : adapter.responseAnnotations())
            {
                serverTracer.submitBinaryAnnotation(annotation.getKey(), annotation.getValue());
            }
            serverTracer.setServerSend();
        } finally {
            serverTracer.clearCurrentSpan();
        }
    }

首先配置BinaryAnnotation信息，然后执行serverTracer.setServerSend,在finally中清除当前线程中的Span信息(不管前面是否清楚成功,最终都将执行该不走)，ThreadLocal中的数据要做到有始有终

看serverTracer.setServerSend()

public void setServerSend() {
        if (submitEndAnnotation(zipkinCoreConstants.SERVER_SEND, spanCollector())) {
            spanAndEndpoint().state().setCurrentServerSpan(null);
        }
    }

终于看到spanCollector收集器了，说明下面将看是收集Span信息,这里为ss注解

boolean submitEndAnnotation(String annotationName, SpanCollector spanCollector) {
        Span span = spanAndEndpoint().span();
        if (span == null) {
          return false;
        }
        Annotation annotation = Annotation.create(
            currentTimeMicroseconds(),
            annotationName,
            spanAndEndpoint().endpoint()
        );
        span.addToAnnotations(annotation);
        if (span.getTimestamp() != null) {
            span.setDuration(annotation.timestamp - span.getTimestamp());
        }
        spanCollector.collect(span);
        return true;
    }

首先获取当前线程中的Span信息，然后处理注解信息，通过annotation.timestamp - span.getTimestamp()计算延迟, 调用spanCollector.collect(span)进行收集Span信息，那么Span信息是同步收集的吗？肯定不是的，接着看

调用spanCollector.collect(span)则执行FlushingSpanCollector中的collect方法

@Override
  public void collect(Span span) {
    metrics.incrementAcceptedSpans(1);
    if (!pending.offer(span)) {
      metrics.incrementDroppedSpans(1);
    }
  }

首先进行的是metrics统计信息，可以自定义该SpanCollectorMetricsHandler信息收集各指标信息,利用如grafana等展示信息

pending.offer(span)将span信息存储在BlockingQueue中，然后通过定时任务去取出阻塞队列中的值，偷偷摸摸的上传span信息

定时任务利用了Flusher类来执行，在构造FlushingSpanCollector时构造了Flusher类

 static final class Flusher implements Runnable {
    final Flushable flushable;
    final ScheduledExecutorService scheduler = Executors.newScheduledThreadPool(1);

    Flusher(Flushable flushable, int flushInterval) {
      this.flushable = flushable;
      this.scheduler.scheduleWithFixedDelay(this, 0, flushInterval, SECONDS);
    }

    @Override
    public void run() {
      try {
        flushable.flush();
      } catch (IOException ignored) {
      }
    }
  }

创建了一个核心线程数为1的线程池，每间隔flushInterval秒执行一次Span信息上传，执行flush方法

@Override
  public void flush() {
    if (pending.isEmpty()) return;
    List<Span> drained = new ArrayList<Span>(pending.size());
    pending.drainTo(drained);
    if (drained.isEmpty()) return;

    int spanCount = drained.size();
    try {
      reportSpans(drained);
    } catch (IOException e) {
      metrics.incrementDroppedSpans(spanCount);
    } catch (RuntimeException e) {
      metrics.incrementDroppedSpans(spanCount);
    }
  }

首先将阻塞队列中的值全部取出存如集合中，最后调用reportSpans(List<Span> drained)抽象方法，该方法在AbstractSpanCollector得到覆写

@Override
  protected void reportSpans(List<Span> drained) throws IOException {
    byte[] encoded = codec.writeSpans(drained);
    sendSpans(encoded);
  }

转换成字节流后调用sendSpans抽象方法发送Span信息，此时就回到一开始说的HttpSpanCollector通过HttpURLConnection实现的sendSpans方法。

more about is here