Skip to content

Tempo

By enabling this toolset, HolmesGPT will be able to fetch trace information from Tempo to debug performance related issues, like high latency in your application.

Proxying through Grafana

This is the recommended approach because we intend to add more capabilities to the toolset that are only available with Grafana.

Prerequisites

A Grafana service account token with the following permissions:

  • Basic role -> Viewer
  • Data sources -> Reader

Check out this video on creating a Grafana service account token.

Getting Grafana URL

You can find the Grafana URL required for Tempo in your Grafana cloud account settings.

Obtaining the datasource UID

You may have multiple Tempo data sources set up in Grafana. HolmesGPT uses a single Tempo datasource to fetch the traces and it needs to know the UID of this datasource.

A simple way to get the datasource UID is to access the Grafana API by running the following request:

# port forward if you are using Robusta's Grafana from your Kubernetes cluster
kubectl port-forward svc/robusta-grafana 3000:80
# List the Tempo data sources
curl -s -u <username>:<password> http://localhost:3000/api/datasources | jq '.[] | select(.type == "tempo")'

This will return something like:

{
    "id": 3,
    "uid": "klja8hsa-8a9c-4b35-1230-7baab22b02ee",
    "orgId": 1,
    "name": "Tempo",
    "type": "tempo",
    "typeName": "Tempo",
    "typeLogoUrl": "/public/app/plugins/datasource/tempo/img/tempo_icon.svg",
    "access": "proxy",
    "url": "http://tempo-query-frontend.tempo:3100",
    "user": "",
    "database": "",
    "basicAuth": false,
    "isDefault": false,
    "jsonData": {
        "tlsSkipVerify": true
    },
    "readOnly": false
}

In this case, the Tempo datasource UID is klja8hsa-8a9c-4b35-1230-7baab22b02ee.

Configuration (Grafana Proxy)

Add the following to ~/.holmes/config.yaml. Create the file if it doesn't exist:

toolsets:
  grafana/tempo:
    enabled: true
    config:
      api_key: <your grafana service account token>
      url: <your grafana url> # e.g. https://acme-corp.grafana.net
      grafana_datasource_uid: <the UID of the tempo data source in Grafana>

After making changes to your configuration, run: 

holmes toolset refresh

To test, run:

holmes ask "The payments DB is very slow, check tempo for any trace data"

holmes:
  toolsets:
    grafana/tempo:
      enabled: true
      config:
        api_key: <your grafana API key>
        url: <your grafana url> # e.g. https://acme-corp.grafana.net
        grafana_datasource_uid: <the UID of the tempo data source in Grafana>

Direct Connection

The toolset can directly connect to a Tempo instance without proxying through a Grafana instance. This is done by not setting the grafana_datasource_uid field. Not setting this field makes HolmesGPT assume that it is directly connecting to Tempo.

Configuration (Direct Connection)

Add the following to ~/.holmes/config.yaml. Create the file if it doesn't exist:

toolsets:
  grafana/tempo:
    enabled: true
    config:
      url: http://tempo.monitoring
      headers:
        X-Scope-OrgID: "<tenant id>" # Set the X-Scope-OrgID if tempo multitenancy is enabled

After making changes to your configuration, run: 

holmes toolset refresh


holmes:
  toolsets:
    grafana/tempo:
      enabled: true
      config:
        url: http://tempo.monitoring
        headers:
          X-Scope-OrgID: "<tenant id>" # Set the X-Scope-OrgID if tempo multitenancy is enabled

Example Usage

Finding Slow Traces

homles ask "Find traces where the payment service is taking longer than 1 second"

Holmes will use TraceQL to search for slow operations: 

{resource.service.name="payment" && duration > 1s}

Analyzing Errors

homles ask "Show me traces with HTTP 500 errors in the frontend service"

Holmes will search using: 

{resource.service.name="frontend" && span.http.status_code = 500}

Capabilities

Tool Name Description
tempo_fetch_traces_comparative_sample Fetches statistics and samples of fast/slow/typical traces for performance analysis
tempo_search_traces_by_query Search traces using TraceQL query language (recommended)
tempo_search_traces_by_tags Search traces using logfmt-encoded tags (legacy)
tempo_query_trace_by_id Retrieve detailed trace information by trace ID
tempo_search_tag_names Discover available tag names across traces
tempo_search_tag_values Get all values for a specific tag
tempo_query_metrics_instant Compute a single TraceQL metric value across time range
tempo_query_metrics_range Get time series data from TraceQL metrics queries