Implementation:Triton inference server Server L0 Metrics Test

Knowledge Sources	Triton Inference Server
Domains	Testing, Metrics
Last Updated	2026-02-13 17:00 GMT

Overview

QA test script for validating Prometheus metrics collection and reporting in the Triton Inference Server.

Description

This test validates that the Triton Inference Server correctly exposes Prometheus-compatible metrics through its metrics endpoint. It verifies that inference count, inference duration, queue time, GPU utilization, GPU memory usage, and other key metrics are accurately reported and updated after inference requests. The test also checks metric formatting, label correctness, and that per-model and per-GPU metrics are properly disaggregated.

Usage

Run as part of the Triton QA test suite. Requires a GPU-enabled Docker environment with pre-generated test models.

Code Reference

Source Location

• Repository: Triton Inference Server
• File: qa/L0_metrics/test.sh
• Lines: 1-489

Signature

#!/bin/bash
source ../common/util.sh
# Test orchestration for Prometheus metrics collection and reporting

Import

source ../common/util.sh

I/O Contract

Inputs

Name	Type	Required	Description
DATADIR	env var	No	Test data directory
MODEL_REPO	directory	Yes	Test model repository

Outputs

Name	Type	Description
exit code	int	0 on success, 1 on failure
test logs	files	Server and test output logs

Usage Examples

Running the Test

cd qa/L0_metrics/
bash test.sh

Related Pages

• Environment:Triton_inference_server_Server_GPU_CUDA_Runtime