Add SkyWalking Java Agent self observability dashboard

[CzyerChen]

#12595

• Update the documentation to include this new feature.
• If it's UI related, attach the screenshots below.

• If this pull request closes/resolves/fixes an existing issue, replace the issue number. Closes [Feature] Java agent self-observability #12595.
• Update the CHANGES log.

[wu-sheng]

I think we need a UI PR for this?

[CzyerChen]

OK

[wu-sheng]

How your preview works without UI change? Is that possible?

[wu-sheng]

I asked for a precision change for the agent side about duration cost by interceptor.
The current ms seems not precious enough, which would end all in 0.

My proposal is using nanosecond in agent and storage, but convert to ms or s as this is agent level metric.

[CzyerChen]

I asked for a precision change for the agent side about duration cost by interceptor. The current ms seems not precious enough, which would end all in 0.

My proposal is using nanosecond in agent and storage, but convert to ms or s as this is agent level metric.

execution time storaged in ns and shown in ms.

[wu-sheng]

These are costs of all plugin executions in this minute, right? Could we add some tips to explain this metric and others?

Or do we have a well explained document?

[wu-sheng]

Please update the UI submodule. And I am going to merge apache/skywalking-java#716, could you add the e2e accordingly?

[wu-sheng]

Please update the UI submodule. And I am going to merge apache/skywalking-java#716, could you add the e2e accordingly?

Merged.

[CzyerChen]

Please update the UI submodule. And I am going to merge apache/skywalking-java#716, could you add the e2e accordingly?

OK

[wu-sheng]

Suggestion, as we have Java agent in e2e, we could bump up java agent version, and add more checks to save resource costs of new e2e.

[wankai123]

LGTM

[wu-sheng]

I think this is not always per minute. When you pick several hours/days duration, the unit would be hour or day. You could just remove per minute for these tips.

[wu-sheng]

There are some irrelevant test cases failing. Please recheck.

Besides #12622 (comment), I think this is good.

[wu-sheng]

@CzyerChen @weixiang1862 This feature should be a good blog material. We could introduce what we add and why these are important.

[wu-sheng]

Could you share why Kafka needs to be disabled?

[CzyerChen]

Could you share why Kafka needs to be disabled?

Kafka test cases introduce the producer and consumer in Kafka Basic\Kafka Profiling, and now the agent adds self observability metrics by default, which will affect the original use case.

[wu-sheng]

Why some new metrics would affect that?

I just want to be certain, we are bot breaking anything.

[weixiang1862]

Why some new metrics would affect that?

I just want to be certain, we are bot breaking anything.

Java agent so11y meters generated agent so11y service traffic which make the order of service list changes and generated new agent so11y traffic.

Original test cases has many service traffic actual and expected checks, it is big change to fix all service expected data, so disable SW_METER_ACTIVE by default and actived it in meter test case to validate so11y changes.

[wu-sheng]

So11y service is using the same service name and instance names, right?

I think we just build extra layer for each service?

[weixiang1862]

So11y service is using the same service name and instance names, right?

I think we just build extra layer for each service?

Yes, the SO11Y_JAVA_AGENT service name and instance name is the same as GENERAL.

[weixiang1862]

So11y service is using the same service name and instance names, right?

I think we just build extra layer for each service?

Metrics won't be effected, just service list values and order.