TraceContext.getCorrelation() cannot obtain the value

[shining-stars-l]

Search before asking

• I had searched in the issues and found no similar issues.

Apache SkyWalking Component

Java Agent (apache/skywalking-java)

What happened

1. In Gateway, the filters use TraceContext.putCorrelation to place custom IDs.

String custom_request_id = request.getHeaders().getFirst("custom_request_id");
if (StringUtil.isNotEmpty(custom_request_id)) {
    TraceContext.putCorrelation("custom_request_id", custom_request_id);
}

2. In the subsequent user services, the custom ID cannot be obtained.

Optional<String> op = TraceContext.getCorrelation("custom_request_id");

What you expected to happen

I think it's because of Gateway's webflux. Because for regular services, such as when the user service calls TraceContext.putCorrelation, and then through Feign calls the order service, the order service can obtain it.

How to reproduce

• springboot：2.3.12.RELEASE
• gateway：2.2.9.RELEASE
• skywalking-java-agent：9.3.0

Anything else

No response

Are you willing to submit a pull request to fix on your own?

• Yes I am willing to submit a pull request on my own!

Code of Conduct

• I agree to follow this project's Code of Conduct

[wu-sheng]

If this gets lost, I think the trace should be broken too. It could be caused by Feign or some framework?

[shining-stars-l]

The gateway normally invokes the user service. After the Gateway executes TraceContext.putCorrelation, the user service is unable to obtain the correlation information through TraceContext.getCorrelation.

However, when the user service calls the order service, the order service is able to obtain the required information properly.

It should not be due to feign, but rather due to Gateway.

[wu-sheng]

TraceContext#correlation is propagated with the trace context. My point is, no matter whether it is a Feign issue or a Gateway plugin issue, if the correlation got lost from upstream, then the trace should be broken too.

[shining-stars-l]

The entire link call was normal without any loss. The traceId was passed throughout the process, and the APM interface could also display the entire link. Only the execution of TraceContext.getCorrelation was lost.

[wu-sheng]

Then, it means the place you wrote this, actually doesn't have tracing context. The trace context propagate in other way. Then this doesn't work.
You need to debug and find whether there is a tracing context in the current thread. Note, you need to learn to debug agent by JVM remote debugging(or other ways)

[shining-stars-l]

I wrote a plugin myself to solve it.