From ce189e2c504bbb97de6af0a9924e5d736a15b789 Mon Sep 17 00:00:00 2001
From: Nathan Paulin <nathan.paulin@inria.fr>
Date: Mon, 5 May 2025 10:09:32 +0200
Subject: [PATCH] [sophia][esterel11] add kwollect metrics
---
.../sophia/clusters/esterel11/esterel11.json | 27 +++++++++++++++++++
.../clusters/esterel11/esterel11_metrics.yaml | 23 ++++++++++++++++
2 files changed, 50 insertions(+)
diff --git a/data/grid5000/sites/sophia/clusters/esterel11/esterel11.json b/data/grid5000/sites/sophia/clusters/esterel11/esterel11.json
index 96ad96b5f22..9dac17a9078 100644
--- a/data/grid5000/sites/sophia/clusters/esterel11/esterel11.json
+++ b/data/grid5000/sites/sophia/clusters/esterel11/esterel11.json
@@ -49,6 +49,33 @@
"port": 9100,
"protocol": "prometheus"
}
+ },
+ {
+ "description": "Default subset of metrics from Prometheus Nvidia DCGM Exporter",
+ "name": "prom_nvgpu_default_metrics",
+ "optional_period": 15000,
+ "period": 0,
+ "source": {
+ "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"
+ ],
+ "port": 9400,
+ "protocol": "prometheus"
+ }
+ },
+ {
+ "description": "All metrics from Prometheus Nvidia DCGM Exporter",
+ "name": "prom_nvgpu_all_metrics",
+ "optional_period": 15000,
+ "period": 0,
+ "source": {
+ "port": 9400,
+ "protocol": "prometheus"
+ }
}
],
"model": "Dell PowerEdge T630",
diff --git a/input/grid5000/sites/sophia/clusters/esterel11/esterel11_metrics.yaml b/input/grid5000/sites/sophia/clusters/esterel11/esterel11_metrics.yaml
index 729e9beb582..a3b0c1ba4e7 100644
--- a/input/grid5000/sites/sophia/clusters/esterel11/esterel11_metrics.yaml
+++ b/input/grid5000/sites/sophia/clusters/esterel11/esterel11_metrics.yaml
@@ -38,3 +38,26 @@ 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
+
--
GitLab