New Relic integrations with the Google Cloud Platform (GCP) include one that reports Gemini Enterprise data to New Relic. This document explains how to activate the GCP Gemini Enterprise integration and describes the data it reports.
Features
Gemini Enterprise is Google Cloud's platform for building and running enterprise search and AI agent applications, built on Discovery Engine. New Relic's Gemini Enterprise integration collects request, latency, and session metrics for apps (engines), agents, the tools those agents invoke, and data connectors, along with resource counts and quota consumption per location and per user.
Activate integration
Important
This integration is available only when you connect your GCP project with Workload Identity Federation. It isn't available for service account or user account connections.
To enable the integration, follow standard procedures to connect your GCP service to New Relic with Workload Identity Federation.
Polling frequency
New Relic integrations query your GCP services according to a polling interval that varies by integration. The polling frequency for Gemini Enterprise is 5 minutes. The resolution is 1 data point every minute.
Workload Identity Federation
Find and use data
After you enable the integration, your Gemini Enterprise resources appear as entities in the New Relic entity explorer. To see dashboards and manage services, go to one.newrelic.com > All capabilities > Infrastructure > GCP.
New Relic collects Gemini Enterprise metrics available in GCP Cloud Monitoring as dimensional metrics in the Metric event type, for example:
FROM Metric SELECT count(*) WHERE metricName LIKE 'gcp.discoveryengine.%' FACET metricNameEntities
Metric data
Engine metrics
The integration reports the following metrics for each Gemini Enterprise engine.
Metric name | Unit | Description |
|---|---|---|
| Count | Total number of requests received by a Gemini Enterprise app (engine). The metric breaks requests down by public API method, canonical gRPC response code, response code class, query type, and model. It spans all traffic an engine receives, including non-agent paths (search, answer, and assist). |
| Milliseconds | Distribution of end-to-end request latency for a Gemini Enterprise app (engine), measured at the assistant server request boundary. The metric breaks latency down by public API method, canonical gRPC response code, response code class, query type, and model. |
| Milliseconds | Distribution of latency from request receipt to the first streamed non-thought answer token, for a Gemini Enterprise app (engine). Google Cloud emits this metric only for requests that produce at least one answer token. The metric breaks latency down by public API method, canonical gRPC response code, response code class, query type, and model. |
| Milliseconds | Distribution of latency from request receipt to the first streamed token (thought or answer), for a Gemini Enterprise app (engine). Google Cloud emits this metric only for requests that produce at least one streamed token. The metric breaks latency down by public API method, canonical gRPC response code, response code class, query type, and model. |
Agent metrics
The integration reports the following metrics for each Gemini Enterprise agent.
Metric name | Unit | Description |
|---|---|---|
| Count | Number of sessions created for each Gemini Enterprise agent. |
| Count | Number of sessions where at least one tool was invoked. |
| Milliseconds | Distribution of total latency of Gemini Enterprise agent invocations over a 1-minute window. |
| Count | Number of session turns created for each Gemini Enterprise agent. |
Distribution metrics report the mean. For gcp.discoveryengine.agent_total_latencies, New Relic also reports the 50th, 95th, and 99th percentiles as separate metrics with the .p50, .p95, and .p99 suffixes.
Agent tool invocation metrics
The integration reports the following metrics for agent tool invocations.
Metric name | Unit | Description |
|---|---|---|
| Milliseconds | Distribution of total latency of Gemini Enterprise Tool invocations over a 1-minute window. |
Distribution metrics report the mean. For gcp.discoveryengine.tool_total_latencies, New Relic also reports the 50th, 95th, and 99th percentiles as separate metrics with the .p50, .p95, and .p99 suffixes.
Data connector metrics
The integration reports the following metrics for data connectors.
Metric name | Unit | Description |
|---|---|---|
| Count | Total number of requests made to the datastore. The metric breaks requests down by request status, gRPC canonical response code, connector tool, engine, and whether the tool reads or writes. |
| Milliseconds | Distribution of wall-clock latency (ms) of DataConnector tool invocations that Google Cloud measures at the ADK tool callback boundary. |
Distribution metrics report the mean. For gcp.discoveryengine.dataconnector.request_latencies, New Relic also reports the 50th, 95th, and 99th percentiles as separate metrics with the .p50, .p95, and .p99 suffixes.
Location metrics
Location metrics are per-location aggregates for your project. They include resource counts, such as the number of engines, data stores, documents, and user events, and search request volume. They also include usage, limits, and limit-exceeded counts for more than 140 Gemini Enterprise quotas, covering areas such as agents, image and video generation, web grounding, and search requests.
Most Location metrics are quota metrics. Each quota is reported as up to three metrics, named by adding a usage, limit, or exceeded suffix, for example gcp.discoveryengine.quota.search_requests_regional.usage. Quota metrics carry the metric.limit_name and metric.regional_location labels.
Location also defines non-quota metrics, such as gcp.discoveryengine.engines_regional. Google Cloud doesn't report data for these in every project, and when it doesn't, they don't appear in New Relic. The same counts are also reported as quota metrics, for example gcp.discoveryengine.quota.engines_regional.usage.
For the full list of Location metrics, see the discoveryengine section of Google's Cloud Monitoring metrics reference. New Relic collects the GA and beta metrics listed there.
Localized quota metrics
The integration reports per-user quotas for each user (gcp.ge_user) and location.
Metric name | Unit | Description |
|---|---|---|
| Count | Number of attempts to exceed the weekly limit for agentic services WTUs per user in the enterprise plus tier. |
| Count | Current weekly limit for agentic services WTUs per user in the enterprise plus tier. |
| Count | Number of attempts to exceed the weekly limit for agentic services WTUs per user in the enterprise standard tier. |
| Count | Current weekly limit for agentic services WTUs per user in the enterprise standard tier. |
For more on Gemini Enterprise metrics, see Access metrics in Google's Gemini Enterprise documentation, and the discoveryengine section of Google's Cloud Monitoring metrics reference for the full list of metrics.