From e0b40e51f80eaea7ebe0e108ead88c2a78842978 Mon Sep 17 00:00:00 2001
From: Julien Lelaurain <julien.lelaurain@inria.fr>
Date: Tue, 1 Apr 2025 09:18:43 +0200
Subject: [PATCH] [sophia][esterel32] Add kwollect metrics
---
.../sophia/clusters/esterel32/esterel32.json | 27 +++++++++++++++++++
.../clusters/esterel32/esterel32_metrics.yaml | 20 ++++++++++++++
2 files changed, 47 insertions(+)
diff --git a/data/grid5000/sites/sophia/clusters/esterel32/esterel32.json b/data/grid5000/sites/sophia/clusters/esterel32/esterel32.json
index a30d4954ac5..04b3bec8be7 100644
--- a/data/grid5000/sites/sophia/clusters/esterel32/esterel32.json
+++ b/data/grid5000/sites/sophia/clusters/esterel32/esterel32.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 T640",
diff --git a/input/grid5000/sites/sophia/clusters/esterel32/esterel32_metrics.yaml b/input/grid5000/sites/sophia/clusters/esterel32/esterel32_metrics.yaml
index 729e9beb582..e2714b591b6 100644
--- a/input/grid5000/sites/sophia/clusters/esterel32/esterel32_metrics.yaml
+++ b/input/grid5000/sites/sophia/clusters/esterel32/esterel32_metrics.yaml
@@ -38,3 +38,23 @@ 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