go.d intelgpu add an option to select specific GPU (#17884)

go.d intel_gpu_top add an option to select specific GPU

7 files changed, 30 insertions, 10 deletions

diff --git a/src/go/collectors/go.d.plugin/modules/intelgpu/config_schema.json b/src/go/collectors/go.d.plugin/modules/intelgpu/config_schema.json
index ddee75e7af..ac81834218 100644
--- a/src/go/collectors/go.d.plugin/modules/intelgpu/config_schema.json
+++ b/src/go/collectors/go.d.plugin/modules/intelgpu/config_schema.json
@@ -10,6 +10,11 @@
         "type": "integer",
         "minimum": 1,
         "default": 1
+      },
+      "device": {
+        "title": "Device selector",
+        "description": "Select Intel GPU ([supported filters](https://manpages.debian.org/testing/intel-gpu-tools/intel_gpu_top.1.en.html#DEVICE_SELECTION)). Use `intel_gpu_top -L` for listing devices. If not set, defaults to first.",
+        "type": "string"
       }
     },
     "additionalProperties": false,
@@ -20,6 +25,9 @@
   "uiSchema": {
     "uiOptions": {
       "fullPage": true
+    },
+    "device": {
+      "ui:placeholder": "For systems with multiple GPUs, create separate data collection jobs and specify the device for each job."
     }
   }
 }
diff --git a/src/go/collectors/go.d.plugin/modules/intelgpu/exec.go b/src/go/collectors/go.d.plugin/modules/intelgpu/exec.go
index d4a62b2a79..8a57c798ba 100644
--- a/src/go/collectors/go.d.plugin/modules/intelgpu/exec.go
+++ b/src/go/collectors/go.d.plugin/modules/intelgpu/exec.go
@@ -14,11 +14,12 @@ import (
 	"github.com/netdata/netdata/go/go.d.plugin/logger"
 )
 
-func newIntelGpuTopExec(ndsudoPath string, updateEvery int, log *logger.Logger) (*intelGpuTopExec, error) {
+func newIntelGpuTopExec(log *logger.Logger, ndsudoPath string, updateEvery int, device string) (*intelGpuTopExec, error) {
 	topExec := &intelGpuTopExec{
 		Logger:             log,
 		ndsudoPath:         ndsudoPath,
 		updateEvery:        updateEvery,
+		device:             device,
 		firstSampleTimeout: time.Second * 3,
 	}
 
@@ -34,6 +35,7 @@ type intelGpuTopExec struct {
 
 	ndsudoPath         string
 	updateEvery        int
+	device             string
 	firstSampleTimeout time.Duration
 
 	cmd  *exec.Cmd
@@ -44,7 +46,13 @@ type intelGpuTopExec struct {
 }
 
 func (e *intelGpuTopExec) run() error {
-	cmd := exec.Command(e.ndsudoPath, "igt-json", "--interval", e.calcIntervalArg())
+	var cmd *exec.Cmd
+
+	if e.device != "" {
+		cmd = exec.Command(e.ndsudoPath, "igt-device-json", "--interval", e.calcIntervalArg(), "--device", e.device)
+	} else {
+		cmd = exec.Command(e.ndsudoPath, "igt-json", "--interval", e.calcIntervalArg())
+	}
 
 	e.Debugf("executing '%s'", cmd)
 
@@ -146,11 +154,8 @@ func (e *intelGpuTopExec) stop() error {
 func (e *intelGpuTopExec) calcIntervalArg() string {
 	// intel_gpu_top appends the end marker ("},\n") of the previous sample to the beginning of the next sample.
 	// interval must be < than 'firstSampleTimeout'
-	var interval int
-	m := min(e.updateEvery, int(e.firstSampleTimeout.Seconds()))
-	if m <= 1 {
-		interval = 900
-	} else {
+	interval := 900
+	if m := min(e.updateEvery, int(e.firstSampleTimeout.Seconds())); m > 1 {
 		interval = m*1000 - 500 // milliseconds
 	}
 	return strconv.Itoa(interval)
diff --git a/src/go/collectors/go.d.plugin/modules/intelgpu/init.go b/src/go/collectors/go.d.plugin/modules/intelgpu/init.go
index e36db7c163..5b02d02b24 100644
--- a/src/go/collectors/go.d.plugin/modules/intelgpu/init.go
+++ b/src/go/collectors/go.d.plugin/modules/intelgpu/init.go
@@ -17,5 +17,5 @@ func (ig *IntelGPU) initIntelGPUTopExec() (intelGpuTop, error) {
 
 	}
 
-	return newIntelGpuTopExec(ndsudoPath, ig.UpdateEvery, ig.Logger)
+	return newIntelGpuTopExec(ig.Logger, ndsudoPath, ig.UpdateEvery, ig.Device)
 }
diff --git a/src/go/collectors/go.d.plugin/modules/intelgpu/intelgpu.go b/src/go/collectors/go.d.plugin/modules/intelgpu/intelgpu.go
index a0f34c2e84..d36295c2be 100644
--- a/src/go/collectors/go.d.plugin/modules/intelgpu/intelgpu.go
+++ b/src/go/collectors/go.d.plugin/modules/intelgpu/intelgpu.go
@@ -29,7 +29,8 @@ func New() *IntelGPU {
 }
 
 type Config struct {
-	UpdateEvery int `yaml:"update_every,omitempty" json:"update_every"`
+	UpdateEvery int    `yaml:"update_every,omitempty" json:"update_every"`
+	Device      string `yaml:"device,omitempty" json:"device"`
 }
 
 type (
diff --git a/src/go/collectors/go.d.plugin/modules/intelgpu/metadata.yaml b/src/go/collectors/go.d.plugin/modules/intelgpu/metadata.yaml
index 1ff16ed785..3b5b39f25b 100644
--- a/src/go/collectors/go.d.plugin/modules/intelgpu/metadata.yaml
+++ b/src/go/collectors/go.d.plugin/modules/intelgpu/metadata.yaml
@@ -60,6 +60,10 @@ modules:
               description: Data collection frequency.
               default_value: 1
               required: false
+            - name: device
+              description: 'Select a specific GPU using [supported filter](https://manpages.debian.org/testing/intel-gpu-tools/intel_gpu_top.1.en.html#DESCRIPTION).'
+              default_value: ""
+              required: false
         examples:
           folding:
             title: Config
diff --git a/src/go/collectors/go.d.plugin/modules/intelgpu/testdata/config.json b/src/go/collectors/go.d.plugin/modules/intelgpu/testdata/config.json
index 0e3f7c4039..167bd15fec 100644
--- a/src/go/collectors/go.d.plugin/modules/intelgpu/testdata/config.json
+++ b/src/go/collectors/go.d.plugin/modules/intelgpu/testdata/config.json
@@ -1,3 +1,4 @@
 {
-  "update_every": 123
+  "update_every": 123,
+  "device": "ok"
 }
diff --git a/src/go/collectors/go.d.plugin/modules/intelgpu/testdata/config.yaml b/src/go/collectors/go.d.plugin/modules/intelgpu/testdata/config.yaml
index f21a3a7a06..f27729e3cb 100644
--- a/src/go/collectors/go.d.plugin/modules/intelgpu/testdata/config.yaml
+++ b/src/go/collectors/go.d.plugin/modules/intelgpu/testdata/config.yaml
@@ -1 +1,2 @@
 update_every: 123
+device: "ok"