feat(inputs.nvidia_smi): Add power-limit field for v12 scheme (#15282)

influxdata · May 3, 2024 · ab6133b · ab6133b
1 parent 59a7eab
commit ab6133b
Show file tree

Hide file tree

Showing 4 changed files with 818 additions and 1 deletion.
diff --git a/plugins/inputs/nvidia_smi/nvidia_smi_test.go b/plugins/inputs/nvidia_smi/nvidia_smi_test.go
@@ -257,7 +257,8 @@ func TestGatherValidXML(t *testing.T) {
 						"memory_used":                   1,
 						"pcie_link_gen_current":         1,
 						"pcie_link_width_current":       8,
-						"power_draw":                    float64(4.61),
+						"power_draw":                    4.61,
+						"power_limit":                   75.0,
 						"serial":                        "0322218049033",
 						"temperature_gpu":               34,
 						"utilization_gpu":               0,
@@ -493,6 +494,54 @@ func TestGatherValidXML(t *testing.T) {
 					time.Unix(1689872450, 0)),
 			},
 		},
+		{
+			name:     "RTC 3090 schema v12",
+			filename: "rtx-3090-v12.xml",
+			expected: []telegraf.Metric{
+				testutil.MustMetric(
+					"nvidia_smi",
+					map[string]string{
+						"compute_mode": "Default",
+						"index":        "0",
+						"name":         "NVIDIA GeForce RTX 3090",
+						"arch":         "Ampere",
+						"pstate":       "P8",
+						"uuid":         "GPU-12345678-aaaa-bbbb-cccc-0123456789ab",
+					},
+					map[string]interface{}{
+						"clocks_current_graphics":       0,
+						"clocks_current_memory":         405,
+						"clocks_current_sm":             0,
+						"clocks_current_video":          555,
+						"cuda_version":                  "12.0",
+						"display_active":                "Disabled",
+						"display_mode":                  "Disabled",
+						"driver_version":                "525.147.05",
+						"encoder_stats_average_fps":     0,
+						"encoder_stats_average_latency": 0,
+						"encoder_stats_session_count":   0,
+						"fbc_stats_average_fps":         0,
+						"fbc_stats_average_latency":     0,
+						"fbc_stats_session_count":       0,
+						"fan_speed":                     0,
+						"power_draw":                    27.23,
+						"power_limit":                   200.0,
+						"memory_free":                   24258,
+						"memory_total":                  24576,
+						"memory_used":                   1,
+						"memory_reserved":               316,
+						"pcie_link_gen_current":         1,
+						"pcie_link_width_current":       16,
+						"temperature_gpu":               37,
+						"utilization_gpu":               0,
+						"utilization_memory":            0,
+						"utilization_encoder":           0,
+						"utilization_decoder":           0,
+						"vbios_version":                 "94.02.71.40.72",
+					},
+					time.Unix(1689872450, 0)),
+			},
+		},
 		{
 			name:     "A100-SXM4 schema v12",
 			filename: "a100-sxm4-v12.xml",

diff --git a/plugins/inputs/nvidia_smi/schema_v12/parser.go b/plugins/inputs/nvidia_smi/schema_v12/parser.go
@@ -76,7 +76,9 @@ func Parse(acc telegraf.Accumulator, buf []byte) error {
 		common.SetIfUsed("int", fields, "clocks_current_memory", gpu.Clocks.MemClock)
 		common.SetIfUsed("int", fields, "clocks_current_video", gpu.Clocks.VideoClock)
 		common.SetIfUsed("float", fields, "power_draw", gpu.PowerReadings.PowerDraw)
+		common.SetIfUsed("float", fields, "power_limit", gpu.PowerReadings.PowerLimit)
 		common.SetIfUsed("float", fields, "power_draw", gpu.GpuPowerReadings.PowerDraw)
+		common.SetIfUsed("float", fields, "power_limit", gpu.GpuPowerReadings.PowerLimit)
 		common.SetIfUsed("float", fields, "module_power_draw", gpu.ModulePowerReadings.PowerDraw)
 		acc.AddFields("nvidia_smi", fields, tags, timestamp)
 

diff --git a/plugins/inputs/nvidia_smi/schema_v12/types.go b/plugins/inputs/nvidia_smi/schema_v12/types.go
@@ -114,6 +114,7 @@ type smi struct {
 			MaxPowerLimit       string `xml:"max_power_limit"`
 			MinPowerLimit       string `xml:"min_power_limit"`
 			PowerDraw           string `xml:"power_draw"`
+			PowerLimit          string `xml:"power_limit"`
 			PowerState          string `xml:"power_state"`
 			RequestedPowerLimit string `xml:"requested_power_limit"`
 		} `xml:"gpu_power_readings"`