基于Dubbo埋点的分布式调用跟踪系统

package com.gxl.core.tracing; import org.slf4j.Logger; import org.slf4j.LoggerFactory; import org.springframework.context.ApplicationContext; import org.springframework.context.support.ClassPathXmlApplicationContext; import com.alibaba.dubbo.common.Constants; import com.alibaba.dubbo.common.extension.Activate; import com.alibaba.dubbo.rpc.Filter; import com.alibaba.dubbo.rpc.Invocation; import com.alibaba.dubbo.rpc.Invoker; import com.alibaba.dubbo.rpc.Result; import com.alibaba.dubbo.rpc.RpcContext; import com.alibaba.dubbo.rpc.RpcException; import com.alibaba.dubbo.rpc.RpcInvocation; import redis.clients.jedis.JedisCluster; /** * 拦截RPC请求进行调用追踪和收集数据 * * @author gaoxianglong */ @SuppressWarnings("resource") @Activate(group = { Constants.PROVIDER, Constants.CONSUMER }) public class TraceFilter implements Filter { private TraceHandler traceHandler; private static TraceDao traceDao; public static JedisCluster jedisCluster; /* 采样率计数器 */ private Integer samplingCounter = 1; /* 采样率 */ public static Integer samplingNum; private static Logger logger = LoggerFactory.getLogger(TraceFilter.class); public TraceFilter() { traceHandler = new TraceHandler(); } static { try { ApplicationContext context = new ClassPathXmlApplicationContext("classpath:tracing-context.xml"); traceDao = (TraceDao) context.getBean("traceDao"); jedisCluster = (JedisCluster) context.getBean("jedisCluster"); } catch (Exception e) { logger.error("error", e); } } @Override public Result invoke(Invoker<?> invoker, Invocation invocation) throws RpcException { logger.info("hashCode-->" + this.hashCode() + "\tthreadID-->" + Thread.currentThread().getId()); final long BEFORE_TIME = System.currentTimeMillis(); RpcInvocation rpcInvocation = (RpcInvocation) invocation; /* 获取当前Span过程的状态记录信息 */ RpcContext context = RpcContext.getContext(); final boolean IS_CONSUMER = context.isConsumerSide(); final boolean IS_PROVIDER = context.isProviderSide(); final String HOST = context.getLocalHost(); final Integer PORT = context.getLocalPort(); final String SERVICE_NAME = context.getUrl().getServiceInterface(); String METHOD_NAME = context.getMethodName(); /* 从ThreadLocal中获取当前线程的Trace上下文信息 */ TraceBean traceBean = traceHandler.getTrace().get(); /* 判断是服务提供方还是调用方 */ if (IS_CONSUMER) { /* 如果ThreadLocal中不包含当前线程的Trace上下文信息则意味着是根调用,需要创建TraceID */ if (null == traceBean) { synchronized (samplingCounter) { if (samplingCounter != samplingNum) { logger.info("serviceName-->" + SERVICE_NAME + "不需要采样"); rpcInvocation.setAttachment("isSampling", "N"); /* 根据请求数递增采样率计数器 */ samplingCounter++; return invoker.invoke(rpcInvocation); } /* 重置采样率计数器 */ samplingCounter = 1; logger.info("serviceName-->" + SERVICE_NAME + "需要采样"); rpcInvocation.setAttachment("isSampling", "Y"); traceBean = traceHandler.createTracer(HOST, PORT, SERVICE_NAME, METHOD_NAME); } } else { /* 检测是否需要采样 */ if (!consumerIsSampling(traceBean, rpcInvocation)) return invoker.invoke(rpcInvocation); final Long TRACE_ID = traceBean.getTraceId(); /** * 重设当前线程的调用链上下文信息 * * 一次请求的Trace通过TraceID串联起来,服务之间的依赖关系、调用顺序 * 通过parentSpanID和SpanID保证, * 服务提供方才会将Trace上下文信息存储在当前线程的ThreadLocal中,SpanID每次都是递增1的, * 而parentSpanID则是取它的上一个Span过程的SpanID,这样就可以明确服务调用的调用顺序和依赖关系 */ traceBean = traceHandler.createTracer(TRACE_ID, traceHandler.getSpanID(TRACE_ID, jedisCluster), traceBean.getSpanId(), HOST, PORT, SERVICE_NAME, METHOD_NAME); } traceBean.setEventType(0); /* 将Trace上下文信息设置到Invocation中 */ setAnnotation(rpcInvocation, traceBean); } if (IS_PROVIDER) { /* 检测是否需要采样 */ if (!providerIsSampling(rpcInvocation)) return invoker.invoke(rpcInvocation); /* 从Invocation中获取调用方传递过来的调用链上下文信息 */ final Long TRACE_ID = Long.parseLong(rpcInvocation.getAttachment("traceID")); final Integer SPAN_ID = Integer.parseInt(rpcInvocation.getAttachment("spanID")); final Integer PARENT_SPAN_ID = Integer.parseInt(rpcInvocation.getAttachment("parentSpanID")); traceBean = traceHandler.createTracer(TRACE_ID, SPAN_ID, PARENT_SPAN_ID, HOST, PORT, SERVICE_NAME, METHOD_NAME); traceBean.setEventType(1); } /* 前置数据收集 */ beforeDataCollect(traceBean, BEFORE_TIME); Result result = invoker.invoke(rpcInvocation); /* 后置数据收集 */ afterDataCollect(traceBean, System.currentTimeMillis()); return result; } /** * 设置需要传递给服务提供方的调用链上下文信息 * * @author gaoxianglong */ public void setAnnotation(RpcInvocation invocation, TraceBean trace) { invocation.setAttachment("traceID", String.valueOf(trace.getTraceId())); invocation.setAttachment("spanID", String.valueOf(trace.getSpanId())); invocation.setAttachment("parentSpanID", String.valueOf(trace.getParentSpanId())); } /** * 前置数据收集 * * @author gaoxianglong */ public void beforeDataCollect(TraceBean trace, long time) { /* 事件类型,0为调用方,1为提供方 */ if (0 == trace.getEventType()) { traceDao.insertClientSendTime(trace, time); } else { traceDao.insertServerReceiveTime(trace, time); /* 将Trace上下文信息设置在ThreadLocal中 */ traceHandler.getTrace().set(trace); } } /** * 后置数据收集 * * @author gaoxianglong */ public void afterDataCollect(TraceBean trace, long time) { if (0 == trace.getEventType()) { traceDao.insertClientReceiveTime(trace, time); } else { traceDao.insertServerSendTime(trace, time); traceHandler.getTrace().remove(); } /* 如果是根服务调用则删除存放在Redis中的一次请求的Span记录 */ if (trace.isRootSpan()) { jedisCluster.del(String.valueOf(trace.getTraceId())); } } /** * 服务调用方是否采样检测 * * @author gaoxianglong */ public boolean consumerIsSampling(TraceBean traceBean, RpcInvocation rpcInvocation) { boolean result = false; if ("N".equals(traceBean.getIsSampling())) { rpcInvocation.setAttachment("isSampling", "N"); } else { rpcInvocation.setAttachment("isSampling", "Y"); result = true; } return result; } /** * 服务提供方是否采样检测 * * @author gaoxianglong */ public boolean providerIsSampling(RpcInvocation rpcInvocation) { boolean result = false; TraceBean traceBean = traceHandler.createTracer(); if ("N".equals(rpcInvocation.getAttachment("isSampling"))) { traceBean.setIsSampling("N"); } else { traceBean.setIsSampling("Y"); result = true; } traceHandler.getTrace().set(traceBean); return result; } }