diff --git a/data/grid5000/sites/toulouse/clusters/montcalm/montcalm.json b/data/grid5000/sites/toulouse/clusters/montcalm/montcalm.json
index ff54a74aaa5e8e1628e72861be30bdbfb92cc120..276547dc98189bf9fe8791a98594055a9fd85977 100644
--- a/data/grid5000/sites/toulouse/clusters/montcalm/montcalm.json
+++ b/data/grid5000/sites/toulouse/clusters/montcalm/montcalm.json
@@ -4,11 +4,84 @@
"exotic": false,
"kavlan": false,
"metrics": [
-
+ {
+ "description": "Default subset of metrics from Prometheus Node Exporter",
+ "name": "prom_default_metrics",
+ "optional_period": 15000,
+ "period": 0,
+ "source": {
+ "id": [
+ "node_boot_time_seconds",
+ "node_cpu_scaling_frequency_hertz",
+ "node_cpu_seconds_total",
+ "node_filesystem_free_bytes",
+ "node_filesystem_size_bytes",
+ "node_load1",
+ "node_load15",
+ "node_load5",
+ "node_memory_Buffers_bytes",
+ "node_memory_Cached_bytes",
+ "node_memory_MemAvailable_bytes",
+ "node_memory_MemFree_bytes",
+ "node_memory_MemTotal_bytes",
+ "node_memory_Shmem_bytes",
+ "node_memory_SwapFree_bytes",
+ "node_memory_SwapTotal_bytes",
+ "node_network_receive_bytes_total",
+ "node_network_receive_packets_total",
+ "node_network_transmit_bytes_total",
+ "node_network_transmit_packets_total",
+ "node_procs_blocked",
+ "node_procs_running",
+ "kwollect_custom"
+ ],
+ "port": 9100,
+ "protocol": "prometheus"
+ }
+ },
+ {
+ "description": "All metrics from Prometheus Node Exporter",
+ "name": "prom_all_metrics",
+ "optional_period": 15000,
+ "period": 0,
+ "source": {
+ "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_GPU_UTIL",
+ "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": "HPE Proliant DL360 Gen10+",
"queues": [
- "admin"
+ "admin",
+ "testing"
],
"type": "cluster",
"uid": "montcalm"
diff --git a/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-1.json b/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-1.json
index 1a07a3a40cf5b570a054cdd1b3f8b10f831cbdab..43895f08f43eea0c503851254f130a6d4c7c09a8 100644
--- a/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-1.json
+++ b/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-1.json
@@ -174,7 +174,8 @@
"deploy": true,
"max_walltime": 0,
"queues": [
- "admin"
+ "admin",
+ "testing"
],
"virtual": "ivt"
},
diff --git a/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-2.json b/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-2.json
index 392996647f0a49773728ebcdab39582c60b2114d..955973fbb3271585be2e712ba1c0d1d7d1e9e9d9 100644
--- a/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-2.json
+++ b/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-2.json
@@ -174,7 +174,8 @@
"deploy": true,
"max_walltime": 0,
"queues": [
- "admin"
+ "admin",
+ "testing"
],
"virtual": "ivt"
},
diff --git a/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-3.json b/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-3.json
index 6746cd26f8cfc187390023d275a19662fec81ff8..fa4f15d39b81fe45c7981ff7ca5cf48802ffbcc9 100644
--- a/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-3.json
+++ b/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-3.json
@@ -174,7 +174,8 @@
"deploy": true,
"max_walltime": 0,
"queues": [
- "admin"
+ "admin",
+ "testing"
],
"virtual": "ivt"
},
diff --git a/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-4.json b/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-4.json
index dcf0ee836a2f34ad9195adda84b7b89b447a7d81..f5e946568348d82ea7f0a83bbbeeca975f32cdd1 100644
--- a/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-4.json
+++ b/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-4.json
@@ -174,7 +174,8 @@
"deploy": true,
"max_walltime": 0,
"queues": [
- "admin"
+ "admin",
+ "testing"
],
"virtual": "ivt"
},
diff --git a/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-5.json b/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-5.json
index 13a47413886c38aefa403cd1c7f8b594c92f1c50..dd4675d06442ebe3a2b22f27fd8534941e8591b7 100644
--- a/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-5.json
+++ b/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-5.json
@@ -174,7 +174,8 @@
"deploy": true,
"max_walltime": 0,
"queues": [
- "admin"
+ "admin",
+ "testing"
],
"virtual": "ivt"
},
diff --git a/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-6.json b/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-6.json
index 59abbf52b5ba98c832ea5d417c4e941180d6ea47..fb7d90117313ab805d6b7febe72926511715a29b 100644
--- a/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-6.json
+++ b/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-6.json
@@ -174,7 +174,8 @@
"deploy": true,
"max_walltime": 0,
"queues": [
- "admin"
+ "admin",
+ "testing"
],
"virtual": "ivt"
},
diff --git a/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-7.json b/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-7.json
index b03fb35fb72f4c7e0f5fc30b80273ff7d03e54e6..43fcb67732391bdd688bd566319df0765811e027 100644
--- a/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-7.json
+++ b/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-7.json
@@ -174,7 +174,8 @@
"deploy": true,
"max_walltime": 0,
"queues": [
- "admin"
+ "admin",
+ "testing"
],
"virtual": "ivt"
},
diff --git a/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-8.json b/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-8.json
index ce3518cbc0879dc766b04dfbe5c24dd811e94091..21ce0e3e3e3c831500788676c373df221f5ff523 100644
--- a/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-8.json
+++ b/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-8.json
@@ -174,7 +174,8 @@
"deploy": true,
"max_walltime": 0,
"queues": [
- "admin"
+ "admin",
+ "testing"
],
"virtual": "ivt"
},
diff --git a/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-9.json b/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-9.json
index 423d1b4feba466a07d401f1183cdb58f07de3a6f..29c0bca59cca1942d808420b057e6f8c3de51f12 100644
--- a/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-9.json
+++ b/data/grid5000/sites/toulouse/clusters/montcalm/nodes/montcalm-9.json
@@ -174,7 +174,8 @@
"deploy": true,
"max_walltime": 0,
"queues": [
- "admin"
+ "admin",
+ "testing"
],
"virtual": "ivt"
},
diff --git a/data/grid5000/sites/toulouse/servers/srv-data.json b/data/grid5000/sites/toulouse/servers/srv-data.json
index d43566891959e99809b3007cb0a678d94d5397e9..12d4a5587c81e57870b021b4e24146a84430c2a9 100644
--- a/data/grid5000/sites/toulouse/servers/srv-data.json
+++ b/data/grid5000/sites/toulouse/servers/srv-data.json
@@ -2,7 +2,8 @@
"alias": [
"nfs",
"home-g5k",
- "modules"
+ "modules",
+ "public"
],
"kind": "physical",
"network_adapters": {
diff --git a/input/grid5000/sites/toulouse/clusters/montcalm/montcalm.yaml b/input/grid5000/sites/toulouse/clusters/montcalm/montcalm.yaml
index be5de43a6a3102aab1e867d9e8e7feb7825b42de..d715dc3deeeac75c4163c59cc38da9b3a19f4559 100644
--- a/input/grid5000/sites/toulouse/clusters/montcalm/montcalm.yaml
+++ b/input/grid5000/sites/toulouse/clusters/montcalm/montcalm.yaml
@@ -5,6 +5,7 @@ boot_type: uefi
exotic: false
queues:
- admin
+ - testing
nodes:
montcalm-[1-9]:
supported_job_types:
diff --git a/input/grid5000/sites/toulouse/clusters/montcalm/montcalm_metrics.yaml b/input/grid5000/sites/toulouse/clusters/montcalm/montcalm_metrics.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..ddd15665712b9f95932408c84bcffe1d1e242272
--- /dev/null
+++ b/input/grid5000/sites/toulouse/clusters/montcalm/montcalm_metrics.yaml
@@ -0,0 +1,65 @@
+---
+metrics:
+ - name: prom_default_metrics
+ description: Default subset of metrics from Prometheus Node Exporter
+ period: 0
+ optional_period: 15000
+ source:
+ protocol: prometheus
+ port: 9100
+ id:
+ - node_boot_time_seconds
+ - node_cpu_scaling_frequency_hertz
+ - node_cpu_seconds_total
+ - node_filesystem_free_bytes
+ - node_filesystem_size_bytes
+ - node_load1
+ - node_load15
+ - node_load5
+ - node_memory_Buffers_bytes
+ - node_memory_Cached_bytes
+ - node_memory_MemAvailable_bytes
+ - node_memory_MemFree_bytes
+ - node_memory_MemTotal_bytes
+ - node_memory_Shmem_bytes
+ - node_memory_SwapFree_bytes
+ - node_memory_SwapTotal_bytes
+ - node_network_receive_bytes_total
+ - node_network_receive_packets_total
+ - node_network_transmit_bytes_total
+ - node_network_transmit_packets_total
+ - node_procs_blocked
+ - node_procs_running
+ - kwollect_custom
+
+ - name: prom_all_metrics
+ description: All metrics from Prometheus Node Exporter
+ period: 0
+ optional_period: 15000
+ 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_GPU_UTIL
+ - 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
+
diff --git a/input/grid5000/sites/toulouse/servers/physical.yaml b/input/grid5000/sites/toulouse/servers/physical.yaml
index 20867a2c7b5a35552965ca1813053adf31d9bec2..df9b54cba40baefd909829e17e973ef4e6693ed7 100644
--- a/input/grid5000/sites/toulouse/servers/physical.yaml
+++ b/input/grid5000/sites/toulouse/servers/physical.yaml
@@ -29,7 +29,7 @@ pve-2:
srv-data:
kind: physical
serial: CZ22230FL8
- alias: [nfs, home-g5k, modules]
+ alias: [nfs, home-g5k, modules, public]
network_adapters:
bmc:
ip: 172.17.126.3