Jaeger 2.6 - Change service_name for prometheus exporter?

[wansors]

Hi,

I'm trying to deploy multiple Jaeger instances (one per namespace) in the same K8s cluster for a multi-tenant application.

Current Behavior:
The Search tab works as expected, showing traces per tenant.

However, the Monitor (APM) tab aggregates metrics from all Jaeger instances.

Root Cause:
After inspecting Prometheus metrics, I noticed that the service_name is the same across all Jaeger instances. This causes the Monitor tab to display combined data.

Attempted Fix:
I tried setting a unique service.name via the Jaeger configuration:

telemetry:
  resource:
    service.name: jaeger-tenant01

But this only changes the service name for Jaeger's own metrics (port 8888), not the application metrics(port 8889).
Changing the deployment name on each NS work, however I want to have the same name for each deployment.

Question:
Is there a way to configure the Prometheus exporter (port 8889) to use a unique service_name per tenant? I couldn’t find this in the documentation.

Any help would be greatly appreciated!

Thanks!

[dosubot[bot]]

Hi @wansors! I'm Dosu and I’m helping the jaeger team.

Based on the available documentation and configuration files, there is currently no documented way to configure the Prometheus exporter (port 8889) in Jaeger to use a unique service_name per tenant for application metrics. The service.name setting in the telemetry section only affects Jaeger's own metrics (typically on port 8888), not the application metrics exposed on port 8889. This behavior is consistent across the example configuration files, such as config-spm.yaml, config-spm-elasticsearch.yaml, and config-kafka-ingester.yaml, where the service.name is set but does not impact the application metrics exporter on port 8889.

There are also no relevant discussions or issues in the Jaeger repository addressing this specific need.

At this time, changing the deployment name per namespace remains the only effective workaround to differentiate metrics in Prometheus, even though you prefer to keep the deployment names the same. If you require strict separation of metrics by tenant, you may need to rely on Kubernetes labels, Prometheus relabeling, or other Prometheus-side mechanisms to distinguish between instances.

If you have further constraints or requirements, please share them, and I can help look for alternative approaches.

To reply, just mention my name @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}   

[yurishkuro]

1. you can deploy prometheus per namespace, same as Jaeger, to avoid overlaps
2. you can use OTEL processors in the ingestion path to rewrite service names in the spans to something like {namespace}/{service}
3. I believe there is a config option in Jaeger that allows you to add additional params to outgoing Prometheus query (statically per Jaeger installation) that allows additional filtering of the metrics.

[wansors]

Hi, thanks for your response.

After reviewing the code for the different options available, I found that I can use namespaces with both the Prometheus exporter and metric_backend. I successfully implemented namespace separation for a single Prometheus instance, though it required a small adjustment since Jaeger uses a default namespace called "traces_span_metrics" for exported metrics.

Here's the working configuration:

...
    jaeger_storage:
        metric_backends:
          some_metrics_storage:
            prometheus:
              endpoint: http://prometheus-service:80
             ...
              metric_namespace: tenant01_traces_span_metrics
...
    exporters:
      jaeger_storage_exporter:
        trace_storage: some_storage
      prometheus:
        endpoint: "0.0.0.0:8889"
        namespace: tenant01