feat(inputs.smart): Add a "device_type" tag to differentiate disks behind a RAID controller (#14613)

Co-authored-by: Thomas Delbende <thomas.delbende@bleemeo.com>

plugins/inputs/smart/README.md

@@ -111,6 +111,10 @@ See the [CONFIGURATION.md][CONFIGURATION.md] for more details.
     ## without a password.
     # use_sudo = false
 
+    ## Adds an extra tag "device_type", which can be used to differentiate
+    ## multiple disks behind the same controller (e.g., MegaRAID).
+    # tag_with_device_type = false
+
     ## Skip checking disks in this power mode. Defaults to
     ## "standby" to not wake up disks that have stopped rotating.
     ## See --nocheck in the man pages for smartctl.
@@ -182,6 +186,7 @@ execute this script.
   - tags:
     - capacity
     - device
+    - device_type (only emitted if `tag_with_device_type` is set to `true`)
     - enabled
     - model
     - serial_no
@@ -201,6 +206,7 @@ execute this script.
   - tags:
     - capacity
     - device
+    - device_type (only emitted if `tag_with_device_type` is set to `true`)
     - enabled
     - fail
     - flags

plugins/inputs/smart/sample.conf

@@ -17,6 +17,10 @@
     ## without a password.
     # use_sudo = false
 
+    ## Adds an extra tag "device_type", which can be used to differentiate
+    ## multiple disks behind the same controller (e.g., MegaRAID).
+    # tag_with_device_type = false
+
     ## Skip checking disks in this power mode. Defaults to
     ## "standby" to not wake up disks that have stopped rotating.
     ## See --nocheck in the man pages for smartctl.

plugins/inputs/smart/smart.go

@@ -353,18 +353,19 @@ var (
 
 // Smart plugin reads metrics from storage devices supporting S.M.A.R.T.
 type Smart struct {
-	Path             string          `toml:"path" deprecated:"1.16.0;use 'path_smartctl' instead"`
+	Path              string          `toml:"path" deprecated:"1.16.0;use 'path_smartctl' instead"`
-	PathSmartctl     string          `toml:"path_smartctl"`
+	PathSmartctl      string          `toml:"path_smartctl"`
-	PathNVMe         string          `toml:"path_nvme"`
+	PathNVMe          string          `toml:"path_nvme"`
-	Nocheck          string          `toml:"nocheck"`
+	Nocheck           string          `toml:"nocheck"`
-	EnableExtensions []string        `toml:"enable_extensions"`
+	EnableExtensions  []string        `toml:"enable_extensions"`
-	Attributes       bool            `toml:"attributes"`
+	Attributes        bool            `toml:"attributes"`
-	Excludes         []string        `toml:"excludes"`
+	Excludes          []string        `toml:"excludes"`
-	Devices          []string        `toml:"devices"`
+	Devices           []string        `toml:"devices"`
-	UseSudo          bool            `toml:"use_sudo"`
+	UseSudo           bool            `toml:"use_sudo"`
-	Timeout          config.Duration `toml:"timeout"`
+	TagWithDeviceType bool            `toml:"tag_with_device_type"`
-	ReadMethod       string          `toml:"read_method"`
+	Timeout           config.Duration `toml:"timeout"`
-	Log              telegraf.Logger `toml:"-"`
+	ReadMethod        string          `toml:"read_method"`
+	Log               telegraf.Logger `toml:"-"`
 }
 
 type nvmeDevice struct {
@@ -741,8 +742,16 @@ func (m *Smart) gatherDisk(acc telegraf.Accumulator, device string, wg *sync.Wai
 	}
 
 	deviceTags := map[string]string{}
-	deviceNode := strings.Split(device, " ")[0]
+	if m.TagWithDeviceType {
-	deviceTags["device"] = path.Base(deviceNode)
+		deviceNode := strings.SplitN(device, " ", 2)
+		deviceTags["device"] = path.Base(deviceNode[0])
+		if len(deviceNode) == 2 && deviceNode[1] != "" {
+			deviceTags["device_type"] = strings.TrimPrefix(deviceNode[1], "-d ")
+		}
+	} else {
+		deviceNode := strings.Split(device, " ")[0]
+		deviceTags["device"] = path.Base(deviceNode)
+	}
 	deviceFields := make(map[string]interface{})
 	deviceFields["exit_status"] = exitStatus
 
@@ -798,7 +807,7 @@ func (m *Smart) gatherDisk(acc telegraf.Accumulator, device string, wg *sync.Wai
 
 		if m.Attributes {
 			//add power mode
-			keys := [...]string{"device", "model", "serial_no", "wwn", "capacity", "enabled", "power"}
+			keys := [...]string{"device", "device_type", "model", "serial_no", "wwn", "capacity", "enabled", "power"}
 			for _, key := range keys {
 				if value, ok := deviceTags[key]; ok {
 					tags[key] = value

plugins/inputs/smart/smart_test.go

@@ -2,6 +2,7 @@ package smart
 
 import (
 	"errors"
+	"fmt"
 	"sync"
 	"testing"
 	"time"
@@ -313,6 +314,39 @@ func TestGatherSSDRaid(t *testing.T) {
 	require.Equal(t, uint64(15), acc.NMetrics(), "Wrong number of metrics gathered")
 }
 
+func TestGatherDeviceTypeTag(t *testing.T) {
+	runCmd = func(timeout config.Duration, sudo bool, command string, args ...string) ([]byte, error) {
+		switch args[0] {
+		case "--scan":
+			return nil, errors.New("scan command should not be run, since devices are provided in config")
+		case "--info":
+			switch args[len(args)-1] {
+			case "megaraid,0":
+				return []byte(smartctlMegaraidInfo1), nil
+			case "megaraid,1":
+				return []byte(smartctlMegaraidInfo2), nil
+			default:
+				return nil, fmt.Errorf("unexpected device type %q", args[len(args)-1])
+			}
+		default:
+			return nil, fmt.Errorf("unexpected command %q", args[0])
+		}
+	}
+
+	s := newSmart()
+	s.Devices = []string{"/dev/bus/0 -d megaraid,0", "/dev/bus/0 -d megaraid,1"}
+	s.TagWithDeviceType = true
+
+	acc := testutil.Accumulator{}
+
+	err := s.Gather(&acc)
+	require.NoError(t, err)
+	require.NoError(t, errors.Join(acc.Errors...))
+
+	result := acc.GetTelegrafMetrics()
+	testutil.RequireMetricsEqual(t, testSmartctlDeviceTypeTag, result, testutil.SortMetrics(), testutil.IgnoreTime())
+}
+
 func TestGatherNVMe(t *testing.T) {
 	runCmd = func(timeout config.Duration, sudo bool, command string, args ...string) ([]byte, error) {
 		return []byte(smartctlNVMeInfoData), nil
@@ -792,6 +826,45 @@ var (
 	mockModel  = "INTEL SSDPEDABCDEFG"
 	mockSerial = "CVFT5123456789ABCD"
 
+	testSmartctlDeviceTypeTag = []telegraf.Metric{
+		testutil.MustMetric(
+			"smart_device",
+			map[string]string{
+				"capacity":    "600000000000",
+				"device":      "0",
+				"device_type": "megaraid,0",
+				"enabled":     "Enabled",
+				"model":       "ST3450857SS",
+				"power":       "ACTIVE",
+				"serial_no":   "xxx",
+			},
+			map[string]any{
+				"exit_status": int64(0),
+				"health_ok":   true,
+				"temp_c":      int64(37),
+			},
+			time.Unix(0, 0),
+		),
+		testutil.MustMetric(
+			"smart_device",
+			map[string]string{
+				"capacity":    "600000000000",
+				"device":      "0",
+				"device_type": "megaraid,1",
+				"enabled":     "Enabled",
+				"model":       "ST3450857SS",
+				"power":       "ACTIVE",
+				"serial_no":   "xxx",
+			},
+			map[string]any{
+				"exit_status": int64(0),
+				"health_ok":   true,
+				"temp_c":      int64(47),
+			},
+			time.Unix(0, 0),
+		),
+	}
+
 	testSmartctlNVMeAttributes = []telegraf.Metric{
 		testutil.MustMetric("smart_device",
 			map[string]string{
@@ -2237,6 +2310,93 @@ Selective self-test flags (0x0):
 	After scanning selected spans, do NOT read-scan remainder of disk.
 If Selective self-test is pending on power-up, resume after 0 minute delay.
 `
+
+	smartctlMegaraidInfo1 = `smartctl 7.3 2022-02-28 r5338 [x86_64-linux-6.2.16-12-pve] (local build)
+Copyright (C) 2002-22, Bruce Allen, Christian Franke, www.smartmontools.org
+
+=== START OF INFORMATION SECTION ===
+Vendor:               SEAGATE
+Product:              ST3450857SS
+Revision:             ES12
+Compliance:           SPC-3
+User Capacity:        600,000,000,000 bytes [600 GB]
+Logical block size:   512 bytes
+Rotation Rate:        15000 rpm
+Form Factor:          3.5 inches
+Logical Unit id:      0x6000c60641d10397
+Serial number:        xxx
+Device type:          disk
+Transport protocol:   SAS (SPL-4)
+Local Time is:        Fri Jan 12 11:43:49 2024 CET
+SMART support is:     Available - device has SMART capability.
+SMART support is:     Enabled
+Temperature Warning:  Disabled or Not Supported
+Power mode is:        ACTIVE
+
+=== START OF READ SMART DATA SECTION ===
+SMART Health Status: OK
+
+Current Drive Temperature:     37 C
+Drive Trip Temperature:        63 C
+
+Accumulated power on time, hours:minutes 16003:18
+Elements in grown defect list: 0
+
+Vendor (Seagate Cache) information
+  Blocks sent to initiator = 3000000000
+  Blocks received from initiator = 3000000000
+  Blocks read from cache and sent to initiator = 3000000000
+  Number of read and write commands whose size <= segment size = 3000000000
+  Number of read and write commands whose size > segment size = 300
+
+Vendor (Seagate/Hitachi) factory information
+  number of hours powered up = 30000.30
+  number of minutes until next internal SMART test = 7
+`
+
+	smartctlMegaraidInfo2 = `smartctl 7.3 2022-02-28 r5338 [x86_64-linux-6.2.16-12-pve] (local build)
+Copyright (C) 2002-22, Bruce Allen, Christian Franke, www.smartmontools.org
+
+=== START OF INFORMATION SECTION ===
+Vendor:               SEAGATE
+Product:              ST3450857SS
+Revision:             ES12
+Compliance:           SPC-3
+User Capacity:        600,000,000,000 bytes [600 GB]
+Logical block size:   512 bytes
+Rotation Rate:        15000 rpm
+Form Factor:          3.5 inches
+Logical Unit id:      0x6000c60641d10497
+Serial number:        xxx
+Device type:          disk
+Transport protocol:   SAS (SPL-4)
+Local Time is:        Fri Jan 12 11:44:49 2024 CET
+SMART support is:     Available - device has SMART capability.
+SMART support is:     Enabled
+Temperature Warning:  Disabled or Not Supported
+Power mode is:        ACTIVE
+
+=== START OF READ SMART DATA SECTION ===
+SMART Health Status: OK
+
+Current Drive Temperature:     47 C
+Drive Trip Temperature:        64 C
+
+Accumulated power on time, hours:minutes 16004:18
+Elements in grown defect list: 0
+
+Vendor (Seagate Cache) information
+  Blocks sent to initiator = 4000000000
+  Blocks received from initiator = 4000000000
+  Blocks read from cache and sent to initiator = 4000000000
+  Number of read and write commands whose size <= segment size = 4000000000
+  Number of read and write commands whose size > segment size = 400
+
+Vendor (Seagate/Hitachi) factory information
+  number of hours powered up = 30000.30
+  number of minutes until next internal SMART test = 7
+`
+
 	smartctlNVMeInfoData = `smartctl 6.5 2016-05-07 r4318 [x86_64-linux-4.1.27-gvt-yocto-standard] (local build)
 Copyright (C) 2002-16, Bruce Allen, Christian Franke, www.smartmontools.org