fix(inputs.nvidia_smi): Add process info metrics (#15023)

influxdata · Mar 20, 2024 · 9080580 · 9080580
1 parent fcb713a
commit 9080580
Show file tree

Hide file tree

Showing 4 changed files with 133 additions and 4 deletions.
diff --git a/plugins/inputs/nvidia_smi/nvidia_smi_test.go b/plugins/inputs/nvidia_smi/nvidia_smi_test.go
@@ -385,6 +385,76 @@ func TestGatherValidXML(t *testing.T) {
 						"vbios_version":                 "94.02.71.40.72",
 					},
 					time.Unix(1689872450, 0)),
+				testutil.MustMetric(
+					"nvidia_smi_process",
+					map[string]string{
+						"name": "/usr/lib/Xorg",
+						"type": "G",
+					},
+					map[string]interface{}{
+						"pid":         int64(835),
+						"used_memory": int64(550),
+					},
+					time.Unix(1689872450, 0)),
+				testutil.MustMetric(
+					"nvidia_smi_process",
+					map[string]string{
+						"name": "/usr/bin/gnome-shell",
+						"type": "G",
+					},
+					map[string]interface{}{
+						"pid":         int64(1481),
+						"used_memory": int64(18),
+					},
+					time.Unix(1689872450, 0)),
+				testutil.MustMetric(
+					"nvidia_smi_process",
+					map[string]string{
+						"name": "/opt/microsoft/msedge/msedge --type=gpu-process " +
+							"--crashpad-handler-pid=2176 --enable-crash-reporter=, " +
+							"--change-stack-guard-on-fork=enable --gpu-preferences=" +
+							"WAAAAAAAAAAgAAAEAAAAAAAAAAAAAAAAAABgAAAAAAA4AAAAAAAAAAA" +
+							"AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAABAAAAGAAAAAAAAAAYAA" +
+							"AAAAAAAAgAAAAAAAAACAAAAAAAAAAIAAAAAAAAAA== --shared-files " +
+							"--field-trial-handle=0,i,3110290512380155730," +
+							"7457693378709978105,262144 --variations-seed-version",
+						"type": "G",
+					},
+					map[string]interface{}{
+						"pid":         int64(2214),
+						"used_memory": int64(79),
+					},
+					time.Unix(1689872450, 0)),
+				testutil.MustMetric(
+					"nvidia_smi_process",
+					map[string]string{
+						"name": "/usr/lib/firefox/firefox",
+						"type": "G",
+					},
+					map[string]interface{}{
+						"pid":         int64(4044),
+						"used_memory": int64(541),
+					},
+					time.Unix(1689872450, 0)),
+				testutil.MustMetric(
+					"nvidia_smi_process",
+					map[string]string{
+						"name": "/opt/visual-studio-code/code --type=gpu-process " +
+							"--enable-crash-reporter=6f39585a-ecc4-42e2-b899-9456cbe56b44" +
+							",no_channel --user-data-dir=/home/powersj/.config/Code " +
+							"--gpu-preferences=WAAAAAAAAAAgAAAEAAAAAAAAAAAAAAAAAABg" +
+							"AAAAAAA4AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA" +
+							"ABAAAAGAAAAAAAAAAYAAAAAAAAAAgAAAAAAAAACAAAAAAAAAAIAAAAAAAAAA== " +
+							"--shared-files --field-trial-handle=0,i,685715063932313394," +
+							"4769839452661094675,262144 --disable-features=" +
+							"CalculateNativeWinOcclusion,SpareRendererForSitePerProcess",
+						"type": "G",
+					},
+					map[string]interface{}{
+						"pid":         int64(42416),
+						"used_memory": int64(159),
+					},
+					time.Unix(1689872450, 0)),
 			},
 		},
 		{

diff --git a/plugins/inputs/nvidia_smi/schema_v12/parser.go b/plugins/inputs/nvidia_smi/schema_v12/parser.go
@@ -103,6 +103,18 @@ func Parse(acc telegraf.Accumulator, buf []byte) error {
 
 			acc.AddFields("nvidia_smi_mig", fields, tags, timestamp)
 		}
+
+		for _, process := range gpu.Processes.ProcessInfo {
+			tags := map[string]string{}
+			common.SetTagIfUsed(tags, "name", process.ProcessName)
+			common.SetTagIfUsed(tags, "type", process.Type)
+
+			fields := map[string]interface{}{}
+			common.SetIfUsed("int", fields, "pid", process.Pid)
+			common.SetIfUsed("int", fields, "used_memory", process.UsedMemory)
+
+			acc.AddFields("nvidia_smi_process", fields, tags, timestamp)
+		}
 	}
 
 	return nil

diff --git a/plugins/inputs/nvidia_smi/schema_v12/types.go b/plugins/inputs/nvidia_smi/schema_v12/types.go
@@ -226,10 +226,17 @@ type smi struct {
 			MinPowerLimit      string `xml:"min_power_limit"`
 			MaxPowerLimit      string `xml:"max_power_limit"`
 		} `xml:"power_readings"`
-		Processes           struct{} `xml:"processes"`
-		ProductArchitecture string   `xml:"product_architecture"`
-		ProductBrand        string   `xml:"product_brand"`
-		ProductName         string   `xml:"product_name"`
+		Processes struct {
+			ProcessInfo []struct {
+				Pid         string `xml:"pid"`
+				Type        string `xml:"type"`
+				ProcessName string `xml:"process_name"`
+				UsedMemory  string `xml:"used_memory"`
+			} `xml:"process_info"`
+		} `xml:"processes"`
+		ProductArchitecture string `xml:"product_architecture"`
+		ProductBrand        string `xml:"product_brand"`
+		ProductName         string `xml:"product_name"`
 		RemappedRows        struct {
 			// Manually added
 			Correctable   string `xml:"remapped_row_corr"`

diff --git a/plugins/inputs/nvidia_smi/testdata/rtx-3080-v12.xml b/plugins/inputs/nvidia_smi/testdata/rtx-3080-v12.xml
@@ -778,6 +778,46 @@
 			</supported_mem_clock>
 		</supported_clocks>
 		<processes>
+			<process_info>
+				<gpu_instance_id>N/A</gpu_instance_id>
+				<compute_instance_id>N/A</compute_instance_id>
+				<pid>835</pid>
+				<type>G</type>
+				<process_name>/usr/lib/Xorg</process_name>
+				<used_memory>550 MiB</used_memory>
+			</process_info>
+			<process_info>
+				<gpu_instance_id>N/A</gpu_instance_id>
+				<compute_instance_id>N/A</compute_instance_id>
+				<pid>1481</pid>
+				<type>G</type>
+				<process_name>/usr/bin/gnome-shell</process_name>
+				<used_memory>18 MiB</used_memory>
+			</process_info>
+			<process_info>
+				<gpu_instance_id>N/A</gpu_instance_id>
+				<compute_instance_id>N/A</compute_instance_id>
+				<pid>2214</pid>
+				<type>G</type>
+				<process_name>/opt/microsoft/msedge/msedge --type=gpu-process --crashpad-handler-pid=2176 --enable-crash-reporter=, --change-stack-guard-on-fork=enable --gpu-preferences=WAAAAAAAAAAgAAAEAAAAAAAAAAAAAAAAAABgAAAAAAA4AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAABAAAAGAAAAAAAAAAYAAAAAAAAAAgAAAAAAAAACAAAAAAAAAAIAAAAAAAAAA== --shared-files --field-trial-handle=0,i,3110290512380155730,7457693378709978105,262144 --variations-seed-version</process_name>
+				<used_memory>79 MiB</used_memory>
+			</process_info>
+			<process_info>
+				<gpu_instance_id>N/A</gpu_instance_id>
+				<compute_instance_id>N/A</compute_instance_id>
+				<pid>4044</pid>
+				<type>G</type>
+				<process_name>/usr/lib/firefox/firefox</process_name>
+				<used_memory>541 MiB</used_memory>
+			</process_info>
+			<process_info>
+				<gpu_instance_id>N/A</gpu_instance_id>
+				<compute_instance_id>N/A</compute_instance_id>
+				<pid>42416</pid>
+				<type>G</type>
+				<process_name>/opt/visual-studio-code/code --type=gpu-process --enable-crash-reporter=6f39585a-ecc4-42e2-b899-9456cbe56b44,no_channel --user-data-dir=/home/powersj/.config/Code --gpu-preferences=WAAAAAAAAAAgAAAEAAAAAAAAAAAAAAAAAABgAAAAAAA4AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAABAAAAGAAAAAAAAAAYAAAAAAAAAAgAAAAAAAAACAAAAAAAAAAIAAAAAAAAAA== --shared-files --field-trial-handle=0,i,685715063932313394,4769839452661094675,262144 --disable-features=CalculateNativeWinOcclusion,SpareRendererForSitePerProcess</process_name>
+				<used_memory>159 MiB</used_memory>
+			</process_info>
 		</processes>
 		<accounted_processes>
 		</accounted_processes>