I am using Spring Cloud Sleuth and Zipkin (via HTTP), by adding spring-cloud-starter-zipkin version 2.0.0.M6 to my dependencies (based on Spring Boot 2.0.0.RC1 and Spring Cloud Finchley M6).
I am using @Newspan annotation to mark a child span around some (expensive) operation. When the span information is sent to Zipkin, I notice that the timestamp and duration of the child span are missing. This leads to a strange rendering on Zipking side. However, when I create the child span by calling tracer#newChild, it works as expected.
Am I missing something? Would this be an issue with Sleuth 2.0.0.M6?
When I run the same code using Spring Boot 1.5.9 and Spring Cloud Edgware SR2, it behaves as expected.
Here's the JSON received on Zipkin side. The span named "child-span-with-annotation" is the one created using @NewSpan, whereas the span "childspanwithnewchild" is created using tracer#newChild.
[
{
"traceId": "b1c2636366c919be",
"id": "b1c2636366c919be",
"name": "get",
"timestamp": 1518495271073166,
"duration": 862032,
"annotations": [
{
"timestamp": 1518495271073166,
"value": "sr",
"endpoint": {
"serviceName": "sample-sleuth-app",
"ipv4": "---.---.---.---"
}
},
{
"timestamp": 1518495271935198,
"value": "ss",
"endpoint": {
"serviceName": "sample-sleuth-app",
"ipv4": "---.---.---.---"
}
}
],
"binaryAnnotations": [
{
"key": "ca",
"value": true,
"endpoint": {
"serviceName": "",
"ipv6": "::1",
"port": 51982
}
},
{
"key": "http.path",
"value": "/hello",
"endpoint": {
"serviceName": "sample-sleuth-app",
"ipv4": "---.---.---.---"
}
},
{
"key": "mvc.controller.class",
"value": "MyRestController",
"endpoint": {
"serviceName": "sample-sleuth-app",
"ipv4": "---.---.---.---"
}
},
{
"key": "mvc.controller.method",
"value": "sayHello",
"endpoint": {
"serviceName": "sample-sleuth-app",
"ipv4": "---.---.---.---"
}
}
]
},
{
"traceId": "b1c2636366c919be",
"id": "14be7ac6eafb0e01",
"name": "child-span-with-annotation",
"parentId": "b1c2636366c919be",
"binaryAnnotations": [
{
"key": "class",
"value": "MyService",
"endpoint": {
"serviceName": "sample-sleuth-app",
"ipv4": "---.---.---.---"
}
},
{
"key": "method",
"value": "expensiveOperation1",
"endpoint": {
"serviceName": "sample-sleuth-app",
"ipv4": "---.---.---.---"
}
}
]
},
{
"traceId": "b1c2636366c919be",
"id": "b34a4f910f27fdb4",
"name": "childspanwithnewchild",
"parentId": "b1c2636366c919be",
"timestamp": 1518495271479040,
"duration": 453747,
"binaryAnnotations": [
{
"key": "lc",
"value": "",
"endpoint": {
"serviceName": "sample-sleuth-app",
"ipv4": "---.---.---.---"
}
}
]
}
]
It's a bug - https://github.com/spring-cloud/spring-cloud-sleuth/issues/855 . I've fixed it ATM. A workaround is to start it manually either in each method that uses @NewSpan
by calling start()
method on current span (that doesn't scale too nicely)
@Autowired SpanCustomizer customizer;
@NewSpan
void foo() {
this.customizer.start();
}
You can also create a bean of SpanCreator
(you can check the fixed version here https://github.com/spring-cloud/spring-cloud-sleuth/blob/master/spring-cloud-sleuth-core/src/main/java/org/springframework/cloud/sleuth/annotation/DefaultSpanCreator.java)
class MySpanCreator implements SpanCreator {
private static final Log log = LogFactory.getLog(MySpanCreator.class);
private final Tracing tracer;
MySpanCreator(Tracing tracer) {
this.tracer = tracer;
}
@Override public Span createSpan(MethodInvocation pjp, NewSpan newSpanAnnotation) {
String name = StringUtils.isEmpty(newSpanAnnotation.name()) ?
pjp.getMethod().getName() : newSpanAnnotation.name();
String changedName = SpanNameUtil.toLowerHyphen(name);
if (log.isDebugEnabled()) {
log.debug("For the class [" + pjp.getThis().getClass() + "] method "
+ "[" + pjp.getMethod().getName() + "] will name the span [" + changedName + "]");
}
return this.tracer.tracer().nextSpan().name(changedName).start();
}
}
Notice the .start()
at the end of the method.