SpringBoot+Zipkin（定制AsyncHttpClient）

引言

由于原有业务服务中使用的是第三方的HttpClient组件调用业务服务接口，而SpringCloud-Sleuth本身没有对AsyncHttpClient提供类似于ClientRequestInterceptor和ClientResponseInterceptor，所以需要我们定制，而ServerRequestInterceptor和ServerResponseInterceptor是在BraveServletFilter部分直接指定就好了（这在任何client技术下都可以复用）。

为了不影响现有业务服务代码逻辑的修改和侵入，保证原有业务服务系统的稳定性和减少人员成本的投入，亟需一种轻量的方式加入少量的代码达到技术实现目标。

分析

SpringCloud-Sleuth默认提供的是基于RestTemplate和javax.servlet.HttpServletRequest的拦截器实现链路追踪，代码如下：

public class TraceRestTemplateInterceptor extends AbstractTraceHttpRequestInterceptor
		implements ClientHttpRequestInterceptor {

	private final ErrorParser errorParser;

	public TraceRestTemplateInterceptor(Tracer tracer, HttpSpanInjector spanInjector,
			HttpTraceKeysInjector httpTraceKeysInjector, ErrorParser errorParser) {
		super(tracer, spanInjector, httpTraceKeysInjector);
		this.errorParser = errorParser;
	}

	@Override
	public ClientHttpResponse intercept(HttpRequest request, byte[] body,
			ClientHttpRequestExecution execution) throws IOException {
		publishStartEvent(request);
		return response(request, body, execution);
	}

	private ClientHttpResponse response(HttpRequest request, byte[] body,
			ClientHttpRequestExecution execution) throws IOException {
		try {
			return new TraceHttpResponse(this, execution.execute(request, body));
		} catch (Exception e) {
			if (log.isDebugEnabled()) {
				log.debug("Exception occurred while trying to execute the request. Will close the span [" + currentSpan() + "]", e);
			}
			this.errorParser.parseErrorTags(currentSpan(), e);
			finish();
			throw e;
		}
	}

}

 通过上面的代码我们可以发现它重写了intercept方法，接收HttpRequest参数。同时，它继承了AbstractTraceHttpRequestInterceptor抽象拦截器。

为了搞清楚内部是如何将多个请求进行串联起来的，延着这个思路，我们分析一下其抽象拦截器的代码：

abstract class AbstractTraceHttpRequestInterceptor {

	protected static final Log log = LogFactory.getLog(MethodHandles.lookup().lookupClass());

	protected final Tracer tracer;
	protected final HttpSpanInjector spanInjector;
	protected final HttpTraceKeysInjector keysInjector;

	protected AbstractTraceHttpRequestInterceptor(Tracer tracer,
			HttpSpanInjector spanInjector, HttpTraceKeysInjector keysInjector) {
		this.tracer = tracer;
		this.spanInjector = spanInjector;
		this.keysInjector = keysInjector;
	}

	/**
	 * Enriches the request with proper headers and publishes
	 * the client sent event
	 */
	protected void publishStartEvent(HttpRequest request) {
		URI uri = request.getURI();
		String spanName = getName(uri);
		Span newSpan = this.tracer.createSpan(spanName);
		this.spanInjector.inject(newSpan, new HttpRequestTextMap(request));
		addRequestTags(request);
		newSpan.logEvent(Span.CLIENT_SEND);
		if (log.isDebugEnabled()) {
			log.debug("Starting new client span [" + newSpan + "]");
		}
	}

	private String getName(URI uri) {
		// The returned name should comply with RFC 882 - Section 3.1.2.
		// i.e Header values must composed of printable ASCII values.
		return SpanNameUtil.shorten(uriScheme(uri) + ":" + uri.getRawPath());
	}

	private String uriScheme(URI uri) {
		return uri.getScheme() == null ? "http" : uri.getScheme();
	}

	/**
	 * Adds HTTP tags to the client side span
	 */
	protected void addRequestTags(HttpRequest request) {
		this.keysInjector.addRequestTags(request.getURI().toString(),
				request.getURI().getHost(),
				request.getURI().getPath(),
				request.getMethod().name(),
				request.getHeaders());
	}

	/**
	 * Close the current span and log the client received event
	 */
	public void finish() {
		if (!isTracing()) {
			return;
		}
		currentSpan().logEvent(Span.CLIENT_RECV);
		this.tracer.close(this.currentSpan());
	}

	protected Span currentSpan() {
		return this.tracer.getCurrentSpan();
	}

	protected boolean isTracing() {
		return this.tracer.isTracing();
	}

}

没错，这就是整个链路追踪请求拦截器的核心之处，在构造方法中会接收三个参数，分别是：Tracer,HttpSpanInjector,HttpTraceKeysInjector。