SpringCloud多线程链路追踪

Spring Cloud项目，如果采用sleuth实现链路追踪，使用线程池时建议采用Configuration配置方式实例化线程池Bean，

这样可以用到sleuth链路追踪，输出日志有traceId和spanId，方便我们查看整个执行链路日志定位问题。

定义

有2种Executor可以定义

• Spring的ThreadPoolTaskExecutor
• JDK的ThreadPoolExecutor

import org.springframework.context.annotation.Bean;
import org.springframework.context.annotation.Configuration;
import org.springframework.scheduling.concurrent.ThreadPoolTaskExecutor;

import java.util.concurrent.Executor;
import java.util.concurrent.ThreadPoolExecutor;

@Configuration
public class ExecutorConfig {

    /**
     * 使用Spring的ThreadPoolTaskExecutor自定义Executor
     * @return
     */
    @Bean
    public Executor customExecutor() {
        ThreadPoolTaskExecutor executor = new ThreadPoolTaskExecutor();
        executor.setCorePoolSize(5);
        executor.setMaxPoolSize(20);
        executor.setQueueCapacity(50);
        executor.setKeepAliveSeconds(60);
        executor.setThreadNamePrefix("CustomExecutor-");

        // rejection-policy：当pool已经达到max size的时候，如何处理新任务
        // CALLER_RUNS：不在新线程中执行任务，而是有调用者所在的线程来执行
        executor.setRejectedExecutionHandler(new ThreadPoolExecutor.CallerRunsPolicy());
        executor.initialize();
        return executor;
    }
	
	 /**
     * JDK的ThreadPoolExecutor自定义Executor
     * @return
     */
    @Bean
    public Executor jdkCustomExecutor() {
    	ThreadPoolExecutor executor = new ThreadPoolExecutor(5, 20, 60, TimeUnit.SECONDS, new ArrayBlockingQueue<Runnable>(60));
    	executor.setThreadFactory(new CustomizableThreadFactory("CustomExecutor-"));
    	executor.setRejectedExecutionHandler(new ThreadPoolExecutor.CallerRunsPolicy());
    	return executor;
    }
}

使用

有2种使用Executor方式，

• 直接注入Executor，手动调用execute
• 使用@Async注解（注意@Aysnc注解是采用AOP实现，同一个类内部方法间调用不会生效）

	//方式1，直接用Executor
	@Resource(name = "customExecutor")
    private Executor executor;
	
	public void methodA(){
		executor.execute(() -> {
			//do something
		});
	}
	
	//方式2，使用@Async注解，注意项目要有@EnableAsync注解
	@Async("customExecutor")
	public void methodB(){
		//do something
	}

能生成traceId和spanId的原因

1. 引入sleuth包后org.springframework.cloud.sleuth.instrument.async.ExecutorBeanPostProcessor类会处理ThreadPoolTaskExecutor和ThreadPoolExecutor，生成代理类。

ThreadPoolTaskExecutor对应的代理类是org.springframework.cloud.sleuth.instrument.async.LazyTraceThreadPoolTaskExecutor

JDK的ThreadPoolExecutor对应的代理类是org.springframework.cloud.sleuth.instrument.async.LazyTraceExecutor

2. Runnable和Callable也会被sleuth包裹成代理类。

execute执行的Runnable被包裹成org.springframework.cloud.sleuth.instrument.async.SpanContinuingTraceRunnable

@Async注解执行的Callable被包裹成org.springframework.cloud.sleuth.instrument.async.SpanContinuingTraceCallable

3. SpanContinuingTraceRunnable和SpanContinuingTraceCallable生成链路追踪traceId和spanId。

日志

2020-05-13 08:58:42.569  INFO [abc-edf-service,aefefc33ebb8b0f9,aefefc33ebb8b0f9,false] 9192 --- [ XNIO-2 task-14] c.a.z.account.controller.TestController  : testExecutor请求接收: []
2020-05-13 08:58:42.600  INFO [abc-edf-service,aefefc33ebb8b0f9,aefefc33ebb8b0f9,false] 9192 --- [chMsgExecutor-2] c.a.z.account.controller.TestController  : ThreadPoolTaskExecutor execute
2020-05-13 08:58:42.602  INFO [abc-edf-service,aefefc33ebb8b0f9,7dd6028c7b79ed63,false] 9192 --- [chMsgExecutor-1] c.a.z.a.component.async.CustomTask       : cn.akeparking.zbcloud.account.component.async.CustomTask

上面是使用Spring的ThreadPoolTaskExecutor不同使用方式的日志，有个问题：手动使用方式日志有链路发起方一样的traceId和spanId；@Async注解方式日志只有一样的traceId，spanId是新生成的。

原因：@Async注解有个切面org.springframework.cloud.sleuth.instrument.async.TraceAsyncAspect，会拦截@Asyc注解的方法，重新生成spanId。

那么使用JDK的ThreadPoolExecutor自定义Executor呢？

日志如下，

2020-08-21 18:24:30.541  INFO [abc-edf,ffff4056ccd39dea,ffff4056ccd39dea,false] 12884 --- [afka-listener-1] c.h.s.s.c.s.l.TopicMessageListener       : receive {"content":"content9999","custom_id":"9999","key":9999}
2020-08-21 18:24:30.578  INFO [abc-edf,ffff4056ccd39dea,ffff4056ccd39dea,false] 12884 --- [ustomExecutor-1] c.h.s.concurrent.ExecutorDifferentUse    : manual execute
2020-08-21 18:24:30.607  INFO [abc-edf,becef15f1baafe8b,becef15f1baafe8b,false] 12884 --- [ustomExecutor-2] c.h.s.concurrent.ExecutorDifferentUse    : @Async execute

手动使用方式日志有链路发起方一样的traceId和spanId；@Async注解方式日志traceId和spanId都是新的。

原因：
JDK的ThreadPoolExecutor的代理类org.springframework.cloud.sleuth.instrument.async.LazyTraceExecutor没有重新实现submit方法，而@Async注解是被包装成Callable，需要用submit方法；所以@Async注解方式最终输出的是切面TraceAsyncAspect重新生成traceId和spanId，不能关联上链路发起方。

结论

不管是Spring的ThreadPoolTaskExecutor还是JDK的ThreadPoolExecutor，建议采用直接注入Executor，手动调用execute方式，这样可以避免@Async注解不能同一个类内部调用和traceId或spanId与链路发起方不一致的问题。