Spring Cloud Sleuth 3.1.4：tracer.currentSpan()返回NULL引发NPE问题求助

问题原因 #

在WebFlux的异步处理流程中，你在chain.filter(exchange).then()的回调里调用tracer.currentSpan()时，原请求的Span已经结束并被关闭。tracer.currentSpan()依赖当前线程的上下文，而异步回调的线程上下文已丢失Span信息，因此返回NULL引发空指针异常。

方案一：提前保存Span上下文 #

在调用chain.filter(exchange)之前（此时Span仍处于活跃状态）获取并保存Span上下文，后续在异步回调中直接使用该上下文：

@Configuration
public class ResponseFilter {

    @Autowired
    Tracer tracer;
    
    @Autowired
    FilterUtils filterUtils;
    
    final Logger logger = LoggerFactory.getLogger(ResponseFilter.class);
    
    @Bean
    public GlobalFilter postGlobalFilter() {
        return (exchange, chain) -> {
            // 提前获取活跃的Span上下文，避免回调中丢失
            SpanContext spanContext = tracer.currentSpan() != null ? tracer.currentSpan().context() : null;
            
            return chain.filter(exchange).then(Mono.fromRunnable(() -> {
                if (spanContext != null) {
                    String traceId = spanContext.traceIdString();
                    logger.debug("Adding the correlation id to the outbound headers. {}", traceId);
                    exchange.getResponse().getHeaders().add(FilterUtils.CORRELATION_ID, traceId);
                    logger.debug("Completing outgoing request for {}.", exchange.getRequest().getURI());
                } else {
                    logger.debug("No active span found for outgoing request {}", exchange.getRequest().getURI());
                }
            }));
        };
    }
}

方案二：通过Reactor上下文获取Trace信息 #

Spring Cloud Sleuth会将Trace上下文存入Reactor的Context中，直接从上下文取出Trace信息，更符合响应式编程模型：

@Configuration
public class ResponseFilter {

    @Autowired
    FilterUtils filterUtils;
    
    final Logger logger = LoggerFactory.getLogger(ResponseFilter.class);
    
    @Bean
    public GlobalFilter postGlobalFilter() {
        return (exchange, chain) -> {
            return chain.filter(exchange)
                    .then(Mono.deferContextual(contextView -> {
                        // 从Reactor上下文取出TraceContext
                        TraceContext traceContext = contextView.getOrDefault(TraceContext.class, null);
                        if (traceContext != null) {
                            String traceId = traceContext.traceIdString();
                            logger.debug("Adding the correlation id to the outbound headers. {}", traceId);
                            exchange.getResponse().getHeaders().add(FilterUtils.CORRELATION_ID, traceId);
                            logger.debug("Completing outgoing request for {}.", exchange.getRequest().getURI());
                        } else {
                            logger.debug("No trace context found in reactor context for outgoing request {}", exchange.getRequest().getURI());
                        }
                        return Mono.empty();
                    }));
        };
    }
}

注意事项 #

• 两种方案都加入了空值判断，避免无Span场景（如健康检查接口）再次抛出NPE。
• 方案二更适合复杂的异步响应式场景，能保证Trace信息在Reactor流中正确传递。

内容的提问来源于stack exchange，提问作者Toji