[CLI]Failed to query JVM instance metrics

[fgksgf]

Please answer these questions before submitting your issue.

• Why do you submit this issue?
• Question or discussion
• Bug
• Requirement
• Feature or performance improvement

---

Bug

This bug is detected, because there is no case when metrics' name starts with "instance", which leads to the scope is All instead of ServiceInstance.

[wu-sheng]

Oh, I don't know there is this kind of logic for query. Could you share why we did this? From my understanding, you only need to set inputs(entity mostly) base on users' input. What is your block?

[fgksgf]

Could you share why we did this?

Because I want to make it much easier for the user. The scope can be inferred from the metric name without user input. After all, the variety of scopes is limited.

Therefore, there are two types of solutions:

• Add a case
• Let users specify the scope

What do you think?

[wu-sheng]

I mean, if the user set --service and --instance, you know this is expected instance scope, right?
If they only give you --service with instance metrics name, you could still do this, the OAP would not return error, instead, it returns all 0 data.

[fgksgf]

Yes, but there are some special circumstances. For example, this query:
swctl metrics top 10 --name="endpoint_cpm" --service=="xxx", the user set --service but its scope is Endpoint. So I think inferring the scope from the metric name is effective.

[wu-sheng]

swctl metrics top 10 --name="endpoint_cpm" --service=="xxx", the user set --service but its scope is Endpoint. So I think inferring the scope from the metric name is effective.

Let's talk about top n separately. Have you taken a look at the UI configuration page? There is a typeOfMetric query like this,
https://github.com/apache/skywalking-query-protocol/blob/4be8e459b0ff56ed5d41a3b5c2173ba83668831e/metrics-v2.graphqls#L138
You are using name referencing is very dangerous, and users can't follow, as the OAL script basically could be named by anyone. UI is configurable and not sensitive about this. So, the scope is really necessary in some cases.

---

About the metrics naming, it is just related to the OAL variable name,

skywalking/oap-server/server-bootstrap/src/main/resources/oal/core.oal

Line 24 in 4aa9f30

service_resp_time = from(Service.latency).longAvg();

Even we could change it at any minute, we can't afford CLI breaks due to this.

[fgksgf]

You are right. Get it. I will let users specify the scope.

[wu-sheng]

I think only top N requires that, others could follow the auto-scope logic easily