go.d storcli add initial support for mpt3sas controllers (netdata#17938)

src/go/collectors/go.d.plugin/modules/storcli/charts.go

@@ -11,7 +11,8 @@ import (
 )
 
 const (
-	prioControllerStatus = module.Priority + iota
+	prioControllerHealthStatus = module.Priority + iota
+	prioControllerStatus
 	prioControllerBBUStatus
 
 	prioPhysDriveErrors
@@ -22,12 +23,30 @@ const (
 	prioBBUTemperature
 )
 
-var controllerChartsTmpl = module.Charts{
+var controllerMegaraidChartsTmpl = module.Charts{
+	controllerHealthStatusChartTmpl.Copy(),
 	controllerStatusChartTmpl.Copy(),
 	controllerBBUStatusChartTmpl.Copy(),
 }
 
+var controllerMpt3sasChartsTmpl = module.Charts{
+	controllerHealthStatusChartTmpl.Copy(),
+}
+
 var (
+	controllerHealthStatusChartTmpl = module.Chart{
+		ID:       "controller_%s_health_status",
+		Title:    "Controller health status",
+		Units:    "status",
+		Fam:      "cntrl status",
+		Ctx:      "storcli.controller_health_status",
+		Type:     module.Line,
+		Priority: prioControllerHealthStatus,
+		Dims: module.Dims{
+			{ID: "cntrl_%s_health_status_healthy", Name: "healthy"},
+			{ID: "cntrl_%s_health_status_unhealthy", Name: "unhealthy"},
+		},
+	}
 	controllerStatusChartTmpl = module.Chart{
 		ID:       "controller_%s_status",
 		Title:    "Controller status",
@@ -139,15 +158,25 @@ var (
 )
 
 func (s *StorCli) addControllerCharts(cntrl controllerInfo) {
-	charts := controllerChartsTmpl.Copy()
+	var charts *module.Charts
+
+	switch cntrl.Version.DriverName {
+	case driverNameMegaraid:
+		charts = controllerMegaraidChartsTmpl.Copy()
+	case driverNameSas:
+		charts = controllerMpt3sasChartsTmpl.Copy()
+	default:
+		return
+	}
 
 	num := strconv.Itoa(cntrl.Basics.Controller)
 
 	for _, chart := range *charts {
 		chart.ID = fmt.Sprintf(chart.ID, num)
 		chart.Labels = []module.Label{
 			{Key: "controller_number", Value: num},
-			{Key: "model", Value: cntrl.Basics.Model},
+			{Key: "model", Value: strings.TrimSpace(cntrl.Basics.Model)},
+			{Key: "driver_name", Value: cntrl.Version.DriverName},
 		}
 		for _, dim := range chart.Dims {
 			dim.ID = fmt.Sprintf(dim.ID, num)

src/go/collectors/go.d.plugin/modules/storcli/collect.go

@@ -4,6 +4,11 @@ package storcli
 
 import "fmt"
 
+const (
+	driverNameMegaraid = "megaraid_sas"
+	driverNameSas      = "mpt3sas"
+)
+
 func (s *StorCli) collect() (map[string]int64, error) {
 	cntrlResp, err := s.queryControllersInfo()
 	if err != nil {
@@ -12,20 +17,28 @@ func (s *StorCli) collect() (map[string]int64, error) {
 
 	mx := make(map[string]int64)
 
-	if err := s.collectControllersInfo(mx, cntrlResp); err != nil {
-		return nil, fmt.Errorf("error collecting controller info: %s", err)
-	}
-
-	drives := cntrlResp.Controllers[0].ResponseData.PDList
 	driver := cntrlResp.Controllers[0].ResponseData.Version.DriverName
-	if driver == "megaraid_sas" && len(drives) > 0 {
-		drivesResp, err := s.queryDrivesInfo()
-		if err != nil {
-			return nil, fmt.Errorf("error collecting drives info: %s", err)
+
+	switch driver {
+	case driverNameMegaraid:
+		if err := s.collectMegaraidControllersInfo(mx, cntrlResp); err != nil {
+			return nil, fmt.Errorf("failed to collect megaraid controller info: %s", err)
+		}
+		if len(cntrlResp.Controllers[0].ResponseData.PDList) > 0 {
+			drivesResp, err := s.queryDrivesInfo()
+			if err != nil {
+				return nil, fmt.Errorf("failed to collect megaraid drive info: %s", err)
+			}
+			if err := s.collectMegaRaidDrives(mx, drivesResp); err != nil {
+				return nil, err
+			}
 		}
-		if err := s.collectMegaRaidDrives(mx, drivesResp); err != nil {
-			return nil, err
+	case driverNameSas:
+		if err := s.collectMpt3sasControllersInfo(mx, cntrlResp); err != nil {
+			return nil, fmt.Errorf("failed to collect mpt3sas controller info: %s", err)
 		}
+	default:
+		return nil, fmt.Errorf("unknown driver: %s", driver)
 	}
 
 	return mx, nil

src/go/collectors/go.d.plugin/modules/storcli/collect_controllers.go

@@ -30,8 +30,8 @@ type (
 			DriverName string `json:"Driver Name"`
 		} `json:"Version"`
 		Status struct {
-			ControllerStatus string     `json:"Controller Status"`
-			BBUStatus        storNumber `json:"BBU Status"`
+			ControllerStatus string      `json:"Controller Status"`
+			BBUStatus        *storNumber `json:"BBU Status"`
 		} `json:"Status"`
 		BBUInfo []struct {
 			Model string `json:"Model"`
@@ -43,7 +43,7 @@ type (
 	}
 )
 
-func (s *StorCli) collectControllersInfo(mx map[string]int64, resp *controllersInfoResponse) error {
+func (s *StorCli) collectMegaraidControllersInfo(mx map[string]int64, resp *controllersInfoResponse) error {
 	for _, v := range resp.Controllers {
 		cntrl := v.ResponseData
 
@@ -56,22 +56,33 @@ func (s *StorCli) collectControllersInfo(mx map[string]int64, resp *controllersI
 
 		px := fmt.Sprintf("cntrl_%s_", cntrlNum)
 
+		for _, st := range []string{"healthy", "unhealthy"} {
+			mx[px+"health_status_"+st] = 0
+		}
+		if strings.ToLower(cntrl.Status.ControllerStatus) == "optimal" {
+			mx[px+"health_status_healthy"] = 1
+		} else {
+			mx[px+"health_status_unhealthy"] = 1
+		}
+
 		for _, st := range []string{"optimal", "degraded", "partially_degraded", "failed"} {
 			mx[px+"status_"+st] = 0
 		}
 		mx[px+"status_"+strings.ToLower(cntrl.Status.ControllerStatus)] = 1
 
-		for _, st := range []string{"healthy", "unhealthy", "na"} {
-			mx[px+"bbu_status_"+st] = 0
-		}
-		// https://github.com/prometheus-community/node-exporter-textfile-collector-scripts/issues/27
-		switch cntrl.Status.BBUStatus {
-		case "0", "8", "4096": // 0 good, 8 charging
-			mx[px+"bbu_status_healthy"] = 1
-		case "NA", "N/A":
-			mx[px+"bbu_status_na"] = 1
-		default:
-			mx[px+"bbu_status_unhealthy"] = 1
+		if cntrl.Status.BBUStatus != nil {
+			for _, st := range []string{"healthy", "unhealthy", "na"} {
+				mx[px+"bbu_status_"+st] = 0
+			}
+			// https://github.com/prometheus-community/node-exporter-textfile-collector-scripts/issues/27
+			switch *cntrl.Status.BBUStatus {
+			case "0", "8", "4096": // 0 good, 8 charging
+				mx[px+"bbu_status_healthy"] = 1
+			case "NA", "N/A":
+				mx[px+"bbu_status_na"] = 1
+			default:
+				mx[px+"bbu_status_unhealthy"] = 1
+			}
 		}
 
 		for i, bbu := range cntrl.BBUInfo {
@@ -92,6 +103,32 @@ func (s *StorCli) collectControllersInfo(mx map[string]int64, resp *controllersI
 	return nil
 }
 
+func (s *StorCli) collectMpt3sasControllersInfo(mx map[string]int64, resp *controllersInfoResponse) error {
+	for _, v := range resp.Controllers {
+		cntrl := v.ResponseData
+
+		cntrlNum := strconv.Itoa(cntrl.Basics.Controller)
+
+		if !s.controllers[cntrlNum] {
+			s.controllers[cntrlNum] = true
+			s.addControllerCharts(cntrl)
+		}
+
+		px := fmt.Sprintf("cntrl_%s_", cntrlNum)
+
+		for _, st := range []string{"healthy", "unhealthy"} {
+			mx[px+"health_status_"+st] = 0
+		}
+		if strings.ToLower(cntrl.Status.ControllerStatus) == "ok" {
+			mx[px+"health_status_healthy"] = 1
+		} else {
+			mx[px+"health_status_unhealthy"] = 1
+		}
+	}
+
+	return nil
+}
+
 func (s *StorCli) queryControllersInfo() (*controllersInfoResponse, error) {
 	bs, err := s.exec.controllersInfo()
 	if err != nil {

src/go/collectors/go.d.plugin/modules/storcli/collect_drives.go

@@ -55,6 +55,10 @@ type storNumber string // some int values can be 'N/A'
 func (n *storNumber) UnmarshalJSON(b []byte) error { *n = storNumber(b); return nil }
 
 func (s *StorCli) collectMegaRaidDrives(mx map[string]int64, resp *drivesInfoResponse) error {
+	if resp == nil {
+		return nil
+	}
+
 	for _, cntrl := range resp.Controllers {
 		var ids []string
 		for k := range cntrl.ResponseData {

src/go/collectors/go.d.plugin/modules/storcli/metadata.yaml

@@ -81,9 +81,9 @@ modules:
       problems:
         list: []
     alerts:
-      - name: storcli_controller_status
-        metric: storcli.controller_status
-        info: RAID controller ${label:controller_number} health status is not optimal
+      - name: storcli_controller_health_status
+        metric: storcli.controller_health_status
+        info: RAID controller ${label:controller_number} is unhealthy
         link: https://github.com/netdata/netdata/blob/master/src/health/health.d/storcli.conf
       - name: storcli_controller_bbu_status
         metric: storcli.controller_bbu_status
@@ -111,7 +111,16 @@ modules:
               description: Controller number (index)
             - name: model
               description: Controller model
+            - name: driver_name
+              description: Controller driver (megaraid_sas or mpt3sas)
           metrics:
+            - name: storcli.controller_health_status
+              description: Controller health status
+              unit: status
+              chart_type: line
+              dimensions:
+                - name: healthy
+                - name: unhealthy
             - name: storcli.controller_status
               description: Controller status
               unit: status

src/go/collectors/go.d.plugin/modules/storcli/storcli_test.go

@@ -19,15 +19,17 @@ var (
 
 	dataMegaControllerInfo, _ = os.ReadFile("testdata/megaraid-controllers-info.json")
 	dataMegaDrivesInfo, _     = os.ReadFile("testdata/megaraid-drives-info.json")
+
+	dataSasControllerInfo, _ = os.ReadFile("testdata/mpt3sas-controllers-info.json")
 )
 
 func Test_testDataIsValid(t *testing.T) {
 	for name, data := range map[string][]byte{
-		"dataConfigJSON": dataConfigJSON,
-		"dataConfigYAML": dataConfigYAML,
-
+		"dataConfigJSON":         dataConfigJSON,
+		"dataConfigYAML":         dataConfigYAML,
 		"dataMegaControllerInfo": dataMegaControllerInfo,
 		"dataMegaDrivesInfo":     dataMegaDrivesInfo,
+		"dataSasControllerInfo":  dataSasControllerInfo,
 	} {
 		require.NotNil(t, data, name)
 	}
@@ -147,12 +149,14 @@ func TestStorCli_Collect(t *testing.T) {
 	}{
 		"success MegaRAID controller": {
 			prepareMock: prepareMockMegaRaidOK,
-			wantCharts:  len(controllerChartsTmpl)*1 + len(physDriveChartsTmpl)*6 + len(bbuChartsTmpl)*1,
+			wantCharts:  len(controllerMegaraidChartsTmpl)*1 + len(physDriveChartsTmpl)*6 + len(bbuChartsTmpl)*1,
 			wantMetrics: map[string]int64{
 				"bbu_0_cntrl_0_temperature":                                       34,
 				"cntrl_0_bbu_status_healthy":                                      1,
 				"cntrl_0_bbu_status_na":                                           0,
 				"cntrl_0_bbu_status_unhealthy":                                    0,
+				"cntrl_0_health_status_healthy":                                   1,
+				"cntrl_0_health_status_unhealthy":                                 0,
 				"cntrl_0_status_degraded":                                         0,
 				"cntrl_0_status_failed":                                           0,
 				"cntrl_0_status_optimal":                                          1,
@@ -195,6 +199,14 @@ func TestStorCli_Collect(t *testing.T) {
 				"phys_drive_5000C500E5659BA7_cntrl_0_temperature":                 27,
 			},
 		},
+		"success SAS controller": {
+			prepareMock: prepareMockSasOK,
+			wantCharts:  len(controllerMpt3sasChartsTmpl) * 1,
+			wantMetrics: map[string]int64{
+				"cntrl_0_health_status_healthy":   1,
+				"cntrl_0_health_status_unhealthy": 0,
+			},
+		},
 		"err on exec": {
 			prepareMock: prepareMockErr,
 			wantMetrics: nil,
@@ -231,6 +243,13 @@ func prepareMockMegaRaidOK() *mockStorCliExec {
 	}
 }
 
+func prepareMockSasOK() *mockStorCliExec {
+	return &mockStorCliExec{
+		controllersInfoData: dataSasControllerInfo,
+		drivesInfoData:      nil,
+	}
+}
+
 func prepareMockErr() *mockStorCliExec {
 	return &mockStorCliExec{
 		errOnInfo: true,