telegraf raid plugin

build/docker/Dockerfile.telegraf-raid

@@ -0,0 +1,3 @@
+FROM registry.cn-beijing.aliyuncs.com/yunionio/yunion-rootfs:v1.0
+
+ADD ./_output/bin/telegraf-raid-plugin /opt/yunion/bin/telegraf-raid-plugin

cmd/raidcli/main.go

@@ -33,7 +33,8 @@ type BaseOptions struct {
 	Host       string `help:"SSH Host IP" default:"$RAID_HOST" metavar:"RAID_HOST"`
 	Username   string `help:"Username, usually root" default:"$RAID_USERNAME" metavar:"RAID_USERNAME"`
 	Password   string `help:"Password" default:"$RAID_PASSWORD" metavar:"RAID_PASSWORD"`
-	Driver     string `help:"Password" default:"$RAID_DRIVER" metavar:"RAID_DRIVER" choices:"MegaRaid|HPSARaid|Mpt2SAS|MarvelRaid"`
+	Driver     string `help:"Raid dirver" default:"$RAID_DRIVER" metavar:"RAID_DRIVER" choices:"MegaRaid|HPSARaid|Mpt2SAS|MarvelRaid"`
+	LocalHost  bool   `help:"Run raidcli in localhost"`
 	SUBCOMMAND string `help:"s3cli subcommand" subcommand:"true"`
 }
 
@@ -83,43 +84,48 @@ func showErrorAndExit(e error) {
 }
 
 func newClient() (raid.IRaidDriver, error) {
-	if len(options.Host) == 0 {
-		return nil, fmt.Errorf("Missing host")
-	}
-
-	if len(options.Username) == 0 {
-		return nil, fmt.Errorf("Missing username")
-	}
-
-	if len(options.Password) == 0 {
-		return nil, fmt.Errorf("Missing password")
+	if options.Debug {
+		raid.Debug = true
 	}
 
 	if len(options.Driver) == 0 {
 		return nil, fmt.Errorf("Missing driver")
 	}
 
-	if options.Debug {
-		raid.Debug = true
+	var drv raid.IRaidDriver
+	if !options.LocalHost {
+		if len(options.Host) == 0 {
+			return nil, fmt.Errorf("Missing host")
+		}
+
+		if len(options.Username) == 0 {
+			return nil, fmt.Errorf("Missing username")
+		}
+
+		if len(options.Password) == 0 {
+			return nil, fmt.Errorf("Missing password")
+		}
+
+		sshClient, err := ssh.NewClient(
+			options.Host,
+			22,
+			options.Username,
+			options.Password,
+			"",
+		)
+		if err != nil {
+			return nil, fmt.Errorf("ssh client init fail: %s", err)
+		}
+
+		drv = drivers.GetDriver(options.Driver, sshClient)
+		if drv == nil {
+			return nil, fmt.Errorf("not supported driver %s", options.Driver)
+		}
+	} else {
+		drv = drivers.GetLocalDriver(options.Driver)
 	}
 
-	sshClient, err := ssh.NewClient(
-		options.Host,
-		22,
-		options.Username,
-		options.Password,
-		"",
-	)
-	if err != nil {
-		return nil, fmt.Errorf("ssh client init fail: %s", err)
-	}
-
-	drv := drivers.GetDriver(options.Driver, sshClient)
-	if drv == nil {
-		return nil, fmt.Errorf("not supported driver %s", options.Driver)
-	}
-
-	err = drv.ParsePhyDevs()
+	err := drv.ParsePhyDevs()
 	if err != nil {
 		return nil, fmt.Errorf("parse phyical devices error %s", err)
 	}

cmd/telegraf-raid-plugin/main.go

@@ -0,0 +1,192 @@
+package main
+
+import (
+	"context"
+	"fmt"
+	"os"
+	"strings"
+	"time"
+
+	v1 "k8s.io/api/core/v1"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	"k8s.io/client-go/kubernetes"
+	"k8s.io/client-go/rest"
+
+	"yunion.io/x/jsonutils"
+	"yunion.io/x/log"
+
+	"yunion.io/x/onecloud/pkg/baremetal/utils/detect_storages"
+	"yunion.io/x/onecloud/pkg/baremetal/utils/raid/drivers"
+	"yunion.io/x/onecloud/pkg/compute/baremetal"
+	"yunion.io/x/onecloud/pkg/util/httputils"
+)
+
+const (
+	INTERVAL_SECOND = 300
+	TelegrafServer  = "http://localhost:8087/write"
+)
+
+// Failed, Offline, Degraded, Rebuilding, Out of Sync (OSY)
+
+func main() {
+	// creates the in-cluster config
+	config, err := rest.InClusterConfig()
+	if err != nil {
+		log.Fatalln(err)
+	}
+	// creates the clientset
+	clientset, err := kubernetes.NewForConfig(config)
+	if err != nil {
+		log.Fatalln(err)
+	}
+
+	nodeName := os.Getenv("NODENAME")
+	if len(nodeName) == 0 {
+		log.Fatalf("Missing env nodename")
+	}
+
+	node, err := clientset.CoreV1().Nodes().Get(nodeName, metav1.GetOptions{})
+	if err != nil {
+		log.Fatalln(err)
+	}
+	var masterAddress string
+	if length := len(node.Status.Conditions); length > 0 {
+		if node.Status.Conditions[length-1].Type == v1.NodeReady &&
+			node.Status.Conditions[length-1].Status == v1.ConditionTrue {
+			for _, addr := range node.Status.Addresses {
+				if addr.Type == v1.NodeInternalIP {
+					masterAddress = addr.Address
+					break
+				}
+			}
+		}
+	}
+
+	log.Infof("Start Colloct Raid Info And Send To Telegraf ...")
+	c := NewRaidInfoCollector(nodeName, masterAddress, INTERVAL_SECOND)
+	c.Start()
+}
+
+type RaidInfoCollector struct {
+	waitingReportData []string
+	LastCollectTime   time.Time
+	ReportInterval    int // seconds
+	Hostname          string
+	HostIp            string
+}
+
+func NewRaidInfoCollector(hostname, hostIp string, interval int) *RaidInfoCollector {
+	return &RaidInfoCollector{
+		waitingReportData: make([]string, 0),
+		ReportInterval:    interval,
+		Hostname:          hostname,
+		HostIp:            hostIp,
+	}
+}
+
+func (c *RaidInfoCollector) runMain() {
+	timeBegin := time.Now()
+	elapse := timeBegin.Sub(c.LastCollectTime)
+	if elapse < time.Second*time.Duration(c.ReportInterval) {
+		return
+	} else {
+		c.LastCollectTime = timeBegin
+	}
+	c.runMonitor()
+}
+
+func (c *RaidInfoCollector) runMonitor() {
+	reportData := c.collectReportData()
+	if len(reportData) > 0 {
+		c.reportRaidInfoToTelegraf(reportData)
+	}
+}
+
+func (c *RaidInfoCollector) collectReportData() string {
+	if len(c.waitingReportData) > 60 {
+		c.waitingReportData = c.waitingReportData[1:]
+	}
+	return c.CollectReportData()
+}
+
+func (c *RaidInfoCollector) CollectReportData() string {
+	raidDiskInfo := make([]*baremetal.BaremetalStorage, 0)
+	// raidDrivers := []string{}
+	for _, drv := range drivers.GetDrivers(drivers.NewExecutor()) {
+		if err := drv.ParsePhyDevs(); err != nil {
+			log.Warningf("Raid driver %s ParsePhyDevs failed: %s", drv.GetName(), err)
+			continue
+		}
+		raidDiskInfo = append(raidDiskInfo, detect_storages.GetRaidDevices(drv)...)
+		// raidDrivers = append(raidDrivers, drv.GetName())
+	}
+	if len(raidDiskInfo) > 0 {
+		ret := c.toTelegrafReportData(raidDiskInfo)
+		return ret
+	}
+	return ""
+}
+
+func (c *RaidInfoCollector) Start() {
+	for {
+		c.runMain()
+		time.Sleep(time.Second * 1)
+	}
+}
+
+const MEASUREMENT = "host_raid"
+
+func (c *RaidInfoCollector) toTelegrafReportData(raidDiskInfo []*baremetal.BaremetalStorage) string {
+	tag := fmt.Sprintf("%s=%s,%s=%s", "hostname", c.Hostname, "host_ip", c.HostIp)
+	ret := []string{}
+	for i := 0; i < len(raidDiskInfo); i++ {
+		statArr := []string{}
+		jStat := jsonutils.Marshal(raidDiskInfo[i])
+		jMap, _ := jStat.GetMap()
+		for k, v := range jMap {
+			statArr = append(statArr, fmt.Sprintf("%s=%s", k, v.String()))
+		}
+		stat := strings.Join(statArr, ",")
+		diskTag := fmt.Sprintf(
+			"%s,%s=%s,%s=%d,%s=%d", tag, "driver", raidDiskInfo[i].Driver,
+			"adapter", raidDiskInfo[i].Adapter, "slot", raidDiskInfo[i].Slot,
+		)
+		line := fmt.Sprintf("%s,%s %s", MEASUREMENT, diskTag, stat)
+		ret = append(ret, line)
+	}
+	return strings.Join(ret, "\n")
+}
+
+func (c *RaidInfoCollector) reportRaidInfoToTelegraf(data string) {
+	body := strings.NewReader(data)
+	res, err := httputils.Request(
+		httputils.GetDefaultClient(), context.Background(), "POST", TelegrafServer, nil, body, false)
+	if err != nil {
+		log.Errorf("Upload guest metric failed: %s", err)
+		return
+	}
+	defer res.Body.Close()
+	if res.StatusCode != 204 {
+		log.Errorf("upload guest metric failed %d", res.StatusCode)
+		timestamp := time.Now().UnixNano()
+		for _, line := range strings.Split(data, "\n") {
+			c.waitingReportData = append(c.waitingReportData,
+				fmt.Sprintf("%s %d", line, timestamp))
+		}
+	} else {
+		if len(c.waitingReportData) > 0 {
+			oldDatas := strings.Join(c.waitingReportData, "\n")
+			body = strings.NewReader(oldDatas)
+			res, err = httputils.Request(
+				httputils.GetDefaultClient(), context.Background(), "POST", TelegrafServer, nil, body, false)
+			if err == nil {
+				defer res.Body.Close()
+			}
+			if res.StatusCode == 204 {
+				c.waitingReportData = c.waitingReportData[len(c.waitingReportData):]
+			} else {
+				log.Errorf("upload guest metric failed code: %d", res.StatusCode)
+			}
+		}
+	}
+}

pkg/baremetal/utils/detect_storages/detect_storages.go

@@ -24,7 +24,6 @@ import (
 	"yunion.io/x/onecloud/pkg/baremetal/utils/raid/drivers"
 	"yunion.io/x/onecloud/pkg/cloudcommon/types"
 	"yunion.io/x/onecloud/pkg/compute/baremetal"
-	"yunion.io/x/onecloud/pkg/util/ssh"
 	"yunion.io/x/onecloud/pkg/util/sysutils"
 )
 
@@ -48,7 +47,7 @@ func GetRaidLogicVolumes(drv raid.IRaidDriver) ([]*raid.RaidLogicalVolume, error
 	return lvs, nil
 }
 
-func DetectStorageInfo(term *ssh.Client, wait bool) ([]*baremetal.BaremetalStorage, []*baremetal.BaremetalStorage, []*baremetal.BaremetalStorage, error) {
+func DetectStorageInfo(term raid.IExecTerm, wait bool) ([]*baremetal.BaremetalStorage, []*baremetal.BaremetalStorage, []*baremetal.BaremetalStorage, error) {
 	raidDiskInfo := make([]*baremetal.BaremetalStorage, 0)
 	lvDiskInfo := make([]*raid.RaidLogicalVolume, 0)
 

pkg/baremetal/utils/raid/drivers/drivers.go

@@ -29,10 +29,9 @@ import (
 	_ "yunion.io/x/onecloud/pkg/baremetal/utils/raid/mvcli"
 	_ "yunion.io/x/onecloud/pkg/baremetal/utils/raid/sas2iru"
 	"yunion.io/x/onecloud/pkg/compute/baremetal"
-	"yunion.io/x/onecloud/pkg/util/ssh"
 )
 
-func GetDriver(name string, term *ssh.Client) raid.IRaidDriver {
+func GetDriver(name string, term raid.IExecTerm) raid.IRaidDriver {
 	factory := raid.RaidDrivers[name]
 	if factory == nil {
 		return nil
@@ -40,7 +39,15 @@ func GetDriver(name string, term *ssh.Client) raid.IRaidDriver {
 	return factory(term)
 }
 
-func GetDriverWithInit(name string, term *ssh.Client) (raid.IRaidDriver, error) {
+func GetLocalDriver(name string) raid.IRaidDriver {
+	factory := raid.RaidDrivers[name]
+	if factory == nil {
+		return nil
+	}
+	return factory(NewExecutor())
+}
+
+func GetDriverWithInit(name string, term raid.IExecTerm) (raid.IRaidDriver, error) {
 	drv := GetDriver(name, term)
 	if drv == nil {
 		return nil, errors.Errorf("Not found raid driver %q", name)
@@ -48,7 +55,7 @@ func GetDriverWithInit(name string, term *ssh.Client) (raid.IRaidDriver, error)
 	return drv, drv.ParsePhyDevs()
 }
 
-func GetDriverByKernelModule(module string, term *ssh.Client) (raid.IRaidDriver, error) {
+func GetDriverByKernelModule(module string, term raid.IExecTerm) (raid.IRaidDriver, error) {
 	name := ""
 	switch module {
 	case raid.MODULE_MEGARAID:
@@ -64,7 +71,7 @@ func GetDriverByKernelModule(module string, term *ssh.Client) (raid.IRaidDriver,
 	return GetDriverWithInit(name, term)
 }
 
-func GetDrivers(term *ssh.Client) []raid.IRaidDriver {
+func GetDrivers(term raid.IExecTerm) []raid.IRaidDriver {
 	ret := []raid.IRaidDriver{}
 	for _, factory := range raid.RaidDrivers {
 		ret = append(ret, factory(term))

pkg/baremetal/utils/raid/drivers/executor.go

@@ -0,0 +1,56 @@
+package drivers
+
+import (
+	"bytes"
+	"fmt"
+	"io"
+	"os/exec"
+
+	"yunion.io/x/log"
+	"yunion.io/x/pkg/errors"
+
+	"yunion.io/x/onecloud/pkg/util/ssh"
+)
+
+type Executor struct{}
+
+func (e *Executor) Run(cmds ...string) ([]string, error) {
+	return e.run(true, cmds, nil)
+}
+
+func (e *Executor) RunWithInput(input io.Reader, cmds ...string) ([]string, error) {
+	return e.run(true, cmds, input)
+}
+
+func (e *Executor) run(parseOutput bool, cmds []string, input io.Reader) ([]string, error) {
+	ret := []string{}
+	for _, cmd := range cmds {
+		log.Debugf("Run command: %s", cmd)
+		proc := exec.Command("sh", "-c", cmd)
+		var stdOut bytes.Buffer
+		var stdErr bytes.Buffer
+		proc.Stdout = &stdOut
+		proc.Stderr = &stdErr
+		proc.Stdin = input
+		if err := proc.Run(); err != nil {
+			var outputErr error
+			errMsg := stdErr.String()
+			if len(stdOut.String()) != 0 {
+				errMsg = fmt.Sprintf("%s %s", errMsg, stdOut.String())
+			}
+			outputErr = errors.Error(errMsg)
+			err = errors.Wrapf(outputErr, "%q error: %v, cmd error", cmd, err)
+			return nil, err
+		}
+		if parseOutput {
+			ret = append(ret, ssh.ParseOutput(stdOut.Bytes())...)
+		} else {
+			ret = append(ret, stdOut.String())
+		}
+	}
+	return ret, nil
+}
+
+func NewExecutor() *Executor {
+	return new(Executor)
+}

pkg/baremetal/utils/raid/hpssactl/hpssactl.go

@@ -31,7 +31,6 @@ import (
 	"yunion.io/x/onecloud/pkg/baremetal/utils/raid"
 	"yunion.io/x/onecloud/pkg/compute/baremetal"
 	"yunion.io/x/onecloud/pkg/util/regutils2"
-	"yunion.io/x/onecloud/pkg/util/ssh"
 )
 
 type HPSARaidPhyDev struct {
@@ -347,11 +346,11 @@ func (adapter *HPSARaidAdaptor) RemoveLogicVolumes() error {
 }
 
 type HPSARaid struct {
-	term     *ssh.Client
+	term     raid.IExecTerm
 	adapters []*HPSARaidAdaptor
 }
 
-func NewHPSARaid(term *ssh.Client) raid.IRaidDriver {
+func NewHPSARaid(term raid.IExecTerm) raid.IRaidDriver {
 	return &HPSARaid{
 		term:     term,
 		adapters: make([]*HPSARaidAdaptor, 0),

pkg/baremetal/utils/raid/interface.go

@@ -15,6 +15,8 @@
 package raid
 
 import (
+	"io"
+
 	api "yunion.io/x/onecloud/pkg/apis/compute"
 	"yunion.io/x/onecloud/pkg/compute/baremetal"
 )
@@ -41,3 +43,8 @@ type IRaidAdapter interface {
 	BuildRaid10(devs []*baremetal.BaremetalStorage, conf *api.BaremetalDiskConfig) error
 	BuildNoneRaid(devs []*baremetal.BaremetalStorage) error
 }
+
+type IExecTerm interface {
+	Run(cmds ...string) ([]string, error)
+	RunWithInput(input io.Reader, cmds ...string) ([]string, error)
+}

pkg/baremetal/utils/raid/megactl/megactl.go

@@ -27,10 +27,10 @@ import (
 	"yunion.io/x/pkg/utils"
 
 	api "yunion.io/x/onecloud/pkg/apis/compute"
+	"yunion.io/x/onecloud/pkg/baremetal/utils/raid"
 	raiddrivers "yunion.io/x/onecloud/pkg/baremetal/utils/raid"
 	"yunion.io/x/onecloud/pkg/compute/baremetal"
 	"yunion.io/x/onecloud/pkg/util/regutils2"
-	"yunion.io/x/onecloud/pkg/util/ssh"
 )
 
 var (
@@ -320,7 +320,7 @@ func (adapter *MegaRaidAdaptor) GetIndex() int {
 	return adapter.index
 }
 
-func (adapter *MegaRaidAdaptor) getTerm() *ssh.Client {
+func (adapter *MegaRaidAdaptor) getTerm() raid.IExecTerm {
 	return adapter.raid.term
 }
 
@@ -376,7 +376,7 @@ func (adapter *MegaRaidAdaptor) parseLogicVolumes(lines []string) ([]*raiddriver
 	return lvs, nil
 }
 
-func getLogicVolumeDeviceById(hostNum, scsiId int, term *ssh.Client) (string, error) {
+func getLogicVolumeDeviceById(hostNum, scsiId int, term raid.IExecTerm) (string, error) {
 	items, err := raiddrivers.SGMap(term)
 	if err != nil {
 		return "", err
@@ -892,13 +892,13 @@ func (adapter *MegaRaidAdaptor) clearJBODDisks() {
 }
 
 type MegaRaid struct {
-	term       *ssh.Client
+	term       raid.IExecTerm
 	adapters   []*MegaRaidAdaptor
 	PhyDevsCnt int
 	Capacity   int64
 }
 
-func NewMegaRaid(term *ssh.Client) raiddrivers.IRaidDriver {
+func NewMegaRaid(term raid.IExecTerm) raiddrivers.IRaidDriver {
 	return &MegaRaid{
 		term:     term,
 		adapters: make([]*MegaRaidAdaptor, 0),

pkg/baremetal/utils/raid/mvcli/mvcli.go

@@ -25,7 +25,6 @@ import (
 	api "yunion.io/x/onecloud/pkg/apis/compute"
 	"yunion.io/x/onecloud/pkg/baremetal/utils/raid"
 	"yunion.io/x/onecloud/pkg/compute/baremetal"
-	"yunion.io/x/onecloud/pkg/util/ssh"
 )
 
 type MarvelRaidPhyDev struct {
@@ -243,11 +242,11 @@ func (adapter *MarvelRaidAdaptor) BuildNoneRaid(devs []*baremetal.BaremetalStora
 }
 
 type MarvelRaid struct {
-	term     *ssh.Client
+	term     raid.IExecTerm
 	adapters []*MarvelRaidAdaptor
 }
 
-func NewMarvelRaid(term *ssh.Client) raid.IRaidDriver {
+func NewMarvelRaid(term raid.IExecTerm) raid.IRaidDriver {
 	return &MarvelRaid{
 		term:     term,
 		adapters: make([]*MarvelRaidAdaptor, 0),

pkg/baremetal/utils/raid/raid.go

@@ -24,7 +24,6 @@ import (
 
 	"yunion.io/x/onecloud/pkg/apis/compute"
 	"yunion.io/x/onecloud/pkg/compute/baremetal"
-	"yunion.io/x/onecloud/pkg/util/ssh"
 	"yunion.io/x/onecloud/pkg/util/sysutils"
 )
 
@@ -45,7 +44,7 @@ const (
 	UnknownLogicalVolumeIndex = MaxInt
 )
 
-type RaidDriverFactory func(term *ssh.Client) IRaidDriver
+type RaidDriverFactory func(term IExecTerm) IRaidDriver
 
 type sRaidDrivers map[string]RaidDriverFactory
 
@@ -106,7 +105,7 @@ func (dev *RaidBasePhyDev) ToBaremetalStorage(index int) *baremetal.BaremetalSto
 	}
 }
 
-func GetModules(term *ssh.Client) []string {
+func GetModules(term IExecTerm) []string {
 	ret := []string{}
 	lines, err := term.Run("/sbin/lsmod")
 	if err != nil {
@@ -142,7 +141,7 @@ type RaidLogicalVolume struct {
 	BlockDev string
 }
 
-func SGMap(term *ssh.Client) ([]compute.SGMapItem, error) {
+func SGMap(term IExecTerm) ([]compute.SGMapItem, error) {
 	lines, err := term.Run("/usr/bin/sg_map -x")
 	if err != nil {
 		return nil, errors.Wrap(err, "run sg_map")

pkg/baremetal/utils/raid/sas2iru/sas2iru.go

@@ -26,7 +26,6 @@ import (
 	api "yunion.io/x/onecloud/pkg/apis/compute"
 	"yunion.io/x/onecloud/pkg/baremetal/utils/raid"
 	"yunion.io/x/onecloud/pkg/compute/baremetal"
-	"yunion.io/x/onecloud/pkg/util/ssh"
 )
 
 type Mpt2SASRaidPhyDev struct {
@@ -307,12 +306,12 @@ func (adapter *Mpt2SASRaidAdaptor) RemoveLogicVolumes() error {
 }
 
 type Mpt2SASRaid struct {
-	term     *ssh.Client
+	term     raid.IExecTerm
 	utility  string
 	adapters []*Mpt2SASRaidAdaptor
 }
 
-func NewMpt2SASRaid(term *ssh.Client) raid.IRaidDriver {
+func NewMpt2SASRaid(term raid.IExecTerm) raid.IRaidDriver {
 	return &Mpt2SASRaid{
 		term:     term,
 		adapters: make([]*Mpt2SASRaidAdaptor, 0),

pkg/hostman/hostinfo/hostconsts/doc.go

@@ -0,0 +1 @@
+package hostconsts // import "yunion.io/x/onecloud/pkg/hostman/hostinfo/hostconsts"

pkg/hostman/hostinfo/hostconsts/hostconsts.go

@@ -0,0 +1,12 @@
+package hostconsts
+
+const (
+	TELEGRAF_TAG_KEY_BRAND     = "brand"
+	TELEGRAF_TAG_KEY_RES_TYPE  = "res_type"
+	TELEGRAF_TAG_KEY_HOST_TYPE = "host_type"
+
+	TELEGRAF_TAG_ONECLOUD_BRAND                = "OneCloud"
+	TELEGRAF_TAG_ONECLOUD_RES_TYPE             = "host"
+	TELEGRAF_TAG_ONECLOUD_HOST_TYPE_HOST       = "host"
+	TELEGRAF_TAG_ONECLOUD_HOST_TYPE_CONTROLLER = "controller"
+)

pkg/hostman/hostinfo/hostinfo.go

@@ -40,6 +40,7 @@ import (
 	"yunion.io/x/onecloud/pkg/hostman/host_health"
 	deployapi "yunion.io/x/onecloud/pkg/hostman/hostdeployer/apis"
 	"yunion.io/x/onecloud/pkg/hostman/hostinfo/hostbridge"
+	"yunion.io/x/onecloud/pkg/hostman/hostinfo/hostconsts"
 	"yunion.io/x/onecloud/pkg/hostman/hostutils"
 	"yunion.io/x/onecloud/pkg/hostman/isolated_device"
 	"yunion.io/x/onecloud/pkg/hostman/options"
@@ -1610,18 +1611,18 @@ func (h *SHostInfo) OnCatalogChanged(catalog mcclient.KeystoneServiceCatalogV3)
 	conf := map[string]interface{}{}
 	conf["hostname"] = h.getHostname()
 	conf["tags"] = map[string]string{
-		"host_id":        h.HostId,
-		"zone_id":        h.ZoneId,
-		"zone":           h.Zone,
-		"cloudregion_id": h.CloudregionId,
-		"cloudregion":    h.Cloudregion,
-		"domain_id":      h.Domain_id,
-		"project_domain": h.Project_domain,
-		"region":         options.HostOptions.Region,
-		"host_ip":        h.GetMasterIp(),
-		//"platform":       "kvm",
-		"brand":    "OneCloud",
-		"res_type": "host",
+		"host_id":                             h.HostId,
+		"zone_id":                             h.ZoneId,
+		"zone":                                h.Zone,
+		"cloudregion_id":                      h.CloudregionId,
+		"cloudregion":                         h.Cloudregion,
+		"domain_id":                           h.Domain_id,
+		"project_domain":                      h.Project_domain,
+		"region":                              options.HostOptions.Region,
+		"host_ip":                             h.GetMasterIp(),
+		hostconsts.TELEGRAF_TAG_KEY_BRAND:     hostconsts.TELEGRAF_TAG_ONECLOUD_BRAND,
+		hostconsts.TELEGRAF_TAG_KEY_RES_TYPE:  hostconsts.TELEGRAF_TAG_ONECLOUD_RES_TYPE,
+		hostconsts.TELEGRAF_TAG_KEY_HOST_TYPE: hostconsts.TELEGRAF_TAG_ONECLOUD_HOST_TYPE_HOST,
 	}
 	conf["nics"] = h.getNicsTelegrafConf()
 	urls, _ := catalog.GetServiceURLs("kafka", options.HostOptions.Region, "", defaultEndpointType)

pkg/hostman/system_service/system_service.go

@@ -16,6 +16,7 @@ package system_service
 
 import (
 	"fmt"
+	"strings"
 
 	"yunion.io/x/log"
 
@@ -97,7 +98,7 @@ func (s *SBaseSystemService) reload(conf, conFile string) error {
 func (s *SBaseSystemService) reloadConf(conf, conFile string) (bool, error) {
 	output, _ := procutils.NewRemoteCommandAsFarAsPossible("cat", conFile).Output()
 	oldConf := string(output)
-	if conf != oldConf {
+	if strings.TrimSpace(conf) != strings.TrimSpace(oldConf) {
 		log.Debugf("Reload service %s ...", s.name)
 		err := procutils.NewRemoteCommandAsFarAsPossible("rm", "-f", conFile).Run()
 		if err != nil {

pkg/hostman/system_service/telegraf.go

@@ -15,8 +15,14 @@
 package system_service
 
 import (
+	"context"
 	"fmt"
+	"sort"
 	"strings"
+
+	"yunion.io/x/log"
+
+	"yunion.io/x/onecloud/pkg/util/httputils"
 )
 
 type STelegraf struct {
@@ -32,8 +38,13 @@ func (s *STelegraf) GetConfig(kwargs map[string]interface{}) string {
 	conf += "[global_tags]\n"
 	if tags, ok := kwargs["tags"]; ok {
 		tgs, _ := tags.(map[string]string)
-		for k, v := range tgs {
-			conf += fmt.Sprintf("  %s = \"%s\"\n", k, v)
+		keys := []string{}
+		for k := range tgs {
+			keys = append(keys, k)
+		}
+		sort.Strings(keys)
+		for _, k := range keys {
+			conf += fmt.Sprintf("  %s = \"%s\"\n", k, tgs[k])
 		}
 	}
 	conf += "\n"
@@ -119,6 +130,9 @@ func (s *STelegraf) GetConfig(kwargs map[string]interface{}) string {
 	conf += "\n"
 	conf += "[[inputs.system]]\n"
 	conf += "\n"
+	conf += "[[inputs.smart]]\n"
+	conf += "  path=\"/usr/sbin/smartctl\"\n"
+	conf += "\n"
 	conf += "[[inputs.net]]\n"
 	if nics, ok := kwargs["nics"]; ok {
 		ns, _ := nics.([]map[string]interface{})
@@ -180,5 +194,26 @@ func (s *STelegraf) BgReload(kwargs map[string]interface{}) {
 }
 
 func (s *STelegraf) BgReloadConf(kwargs map[string]interface{}) {
-	go s.reloadConf(s.GetConfig(kwargs), s.GetConfigFile())
+	go func() {
+		reload, err := s.reloadConf(s.GetConfig(kwargs), s.GetConfigFile())
+		if err != nil {
+			log.Errorf("Failed reload conf: %s", err)
+		}
+		if reload {
+			err := s.ReloadTelegraf()
+			if err != nil {
+				log.Errorf("failed reload telegraf: %s", err)
+			}
+		}
+	}()
+}
+
+func (s *STelegraf) ReloadTelegraf() error {
+	log.Infof("Start reolad telegraf...")
+	telegrafReoladUrl := "http://localhost:8087/reload"
+	_, _, err := httputils.JSONRequest(
+		httputils.GetDefaultClient(), context.Background(),
+		"POST", telegrafReoladUrl, nil, nil, false,
+	)
+	return err
 }