Exported metrics have a domain instead of the full path

[vdombrovski]

ISSUE TYPE

• Bug Report

COMPONENT NAME

API / metrology

CLOUDSTACK VERSION

4.17.1.0

CONFIGURATION

OS / ENVIRONMENT

Irrelevant

SUMMARY

Prometheus metrics are grouped by domain, however the domain name provided isn't the full path. This leads to some duplicate labels whenever your domain names aren't unique

STEPS TO REPRODUCE

1. Create a domain /ROOT/CUSTOMER_A/CLOUD
2. Create a domain /ROOT/CUSTOMER_B/CLOUD
3. Enabled prometheus metrics inside UI configuration and restart management
4. Gather Prometheus metrics (here demonstrated using curl): curl "http://127.0.0.1:9595/metrics"

EXPECTED RESULTS

Either a full path or the parent is reported

cloudstack_domain_resource_count{domain="CLOUD", path="/ROOT/CUSTOMER_A/CLOUD", type="cpu"} 0
cloudstack_domain_resource_count{domain="CLOUD", path="/ROOT/CUSTOMER_B/CLOUD", type="cpu"} 0

ACTUAL RESULTS

cloudstack_domain_resource_count{domain="CLOUD", type="cpu"} 0
cloudstack_domain_resource_count{domain="CLOUD", type="cpu"} 0

[vdombrovski]

Looks like changing domain.getName() into domain.getPath() should be sufficient to get the proper domain names:

cloudstack/plugins/integrations/prometheus/src/main/java/org/apache/cloudstack/metrics/PrometheusExporterImpl.java

Line 371 in 4e2f461

metricsList.add(new ItemPerDomainResourceLimit(cpuLimit, domain.getName(), Resource.ResourceType.cpu.getName()));

[yadvr]

@DaanHoogland Is this critical/blocker issue?

[DaanHoogland]

no, not as far I I can see. we can skip it if there isn´t a simple fix.

[vdombrovski]

I've detailed the proposed fix above. I'd like to know if it is OK having this breaking change to be more accurate on the reported labels, or whether it's better to add another "path" label to the existing metric without changing the name label.

[DaanHoogland]

To tell you the truth @vdombrovski , I don´t know how this is used and whether it would break anything. It seems alright to me, but in my experience even changing log messages can break things (like monitorring systems)

[DaanHoogland]

that said, I´d say go for it.

[vdombrovski]

I've issued a fix in the above PR, please let me know if this change looks OK.