From 9bf28cbb32bc0afdd0a305d4192110cc83abdc92 Mon Sep 17 00:00:00 2001
From: Nathan Paulin <nathan.paulin@inria.fr>
Date: Mon, 24 Mar 2025 10:43:30 +0100
Subject: [PATCH] [sophia][esterel4] kwollect add gpu metrics
---
.../clusters/esterel4/esterel4_metrics.yaml | 22 +++++++++++++++++++
1 file changed, 22 insertions(+)
diff --git a/input/grid5000/sites/sophia/clusters/esterel4/esterel4_metrics.yaml b/input/grid5000/sites/sophia/clusters/esterel4/esterel4_metrics.yaml
index 3c11d0073c1..bbcc255a649 100644
--- a/input/grid5000/sites/sophia/clusters/esterel4/esterel4_metrics.yaml
+++ b/input/grid5000/sites/sophia/clusters/esterel4/esterel4_metrics.yaml
@@ -40,3 +40,25 @@ metrics:
source:
protocol: prometheus
port: 9100
+
+ - name: prom_nvgpu_default_metrics
+ description: Default subset of metrics from Prometheus Nvidia DCGM Exporter
+ period: 0
+ optional_period: 15000
+ source:
+ protocol: prometheus
+ port: 9400
+ id:
+ - DCGM_FI_DEV_SM_CLOCK
+ - DCGM_FI_DEV_MEM_CLOCK
+ - DCGM_FI_DEV_GPU_TEMP
+ - DCGM_FI_DEV_POWER_USAGE
+ - DCGM_FI_DEV_MEM_COPY_UTIL
+
+ - name: prom_nvgpu_all_metrics
+ description: All metrics from Prometheus Nvidia DCGM Exporter
+ period: 0
+ optional_period: 15000
+ source:
+ protocol: prometheus
+ port: 9400
\ No newline at end of file
--
GitLab