Fix wrong memory measurements of containers and vms (#8290)

plugins/inputs/proxmox/README.md

@@ -11,6 +11,8 @@ Telegraf minimum version: Telegraf 1.16.0
   ## API connection configuration. The API token was introduced in Proxmox v6.2. Required permissions for user and token: PVEAuditor role on /.
   base_url = "https://localhost:8006/api2/json"
   api_token = "USER@REALM!TOKENID=UUID"
+  ## Optional node name config
+  # node_name = "localhost"
 
   ## Optional TLS Config
   # tls_ca = "/etc/telegraf/ca.pem"

plugins/inputs/proxmox/proxmox.go

@@ -2,6 +2,7 @@ package proxmox
 
 import (
 	"encoding/json"
+	"errors"
 	"github.com/influxdata/telegraf"
 	"github.com/influxdata/telegraf/plugins/inputs"
 	"io/ioutil"
@@ -48,11 +49,10 @@ func (px *Proxmox) Gather(acc telegraf.Accumulator) error {
 }
 
 func (px *Proxmox) Init() error {
-	hostname, err := os.Hostname()
+
-	if err != nil {
+	if px.NodeName == "" {
-		return err
+		return errors.New("node_name must be configured")
 	}
-	px.hostname = hostname
 
 	tlsCfg, err := px.ClientConfig.TLSConfig()
 	if err != nil {
@@ -73,11 +73,15 @@ func init() {
 		requestFunction: performRequest,
 	}
 
+	// Set hostname as default node name for backwards compatibility
+	hostname, _ := os.Hostname()
+	px.NodeName = hostname
+
 	inputs.Add("proxmox", func() telegraf.Input { return &px })
 }
 
 func getNodeSearchDomain(px *Proxmox) error {
-	apiUrl := "/nodes/" + px.hostname + "/dns"
+	apiUrl := "/nodes/" + px.NodeName + "/dns"
 	jsonData, err := px.requestFunction(px, apiUrl, http.MethodGet, nil)
 	if err != nil {
 		return err
@@ -88,6 +92,10 @@ func getNodeSearchDomain(px *Proxmox) error {
 	if err != nil {
 		return err
 	}
+
+	if nodeDns.Data.Searchdomain == "" {
+		return errors.New("node_name not found")
+	}
 	px.nodeSearchDomain = nodeDns.Data.Searchdomain
 
 	return nil
@@ -137,7 +145,12 @@ func gatherVmData(px *Proxmox, acc telegraf.Accumulator, rt ResourceType) {
 			return
 		}
 		tags := getTags(px, vmStat.Name, vmConfig, rt)
-		fields, err := getFields(vmStat)
+		currentVMStatus, err := getCurrentVMStatus(px, rt, vmStat.ID)
+		if err != nil {
+			px.Log.Error("Error getting VM curent VM status: %v", err)
+			return
+		}
+		fields, err := getFields(currentVMStatus)
 		if err != nil {
 			px.Log.Error("Error getting VM measurements: %v", err)
 			return
@@ -146,8 +159,25 @@ func gatherVmData(px *Proxmox, acc telegraf.Accumulator, rt ResourceType) {
 	}
 }
 
+func getCurrentVMStatus(px *Proxmox, rt ResourceType, id string) (VmStat, error) {
+	apiUrl := "/nodes/" + px.NodeName + "/" + string(rt) + "/" + id + "/status/current"
+
+	jsonData, err := px.requestFunction(px, apiUrl, http.MethodGet, nil)
+	if err != nil {
+		return VmStat{}, err
+	}
+
+	var currentVmStatus VmCurrentStats
+	err = json.Unmarshal(jsonData, &currentVmStatus)
+	if err != nil {
+		return VmStat{}, err
+	}
+
+	return currentVmStatus.Data, nil
+}
+
 func getVmStats(px *Proxmox, rt ResourceType) (VmStats, error) {
-	apiUrl := "/nodes/" + px.hostname + "/" + string(rt)
+	apiUrl := "/nodes/" + px.NodeName + "/" + string(rt)
 	jsonData, err := px.requestFunction(px, apiUrl, http.MethodGet, nil)
 	if err != nil {
 		return VmStats{}, err
@@ -163,7 +193,7 @@ func getVmStats(px *Proxmox, rt ResourceType) (VmStats, error) {
 }
 
 func getVmConfig(px *Proxmox, vmId string, rt ResourceType) (VmConfig, error) {
-	apiUrl := "/nodes/" + px.hostname + "/" + string(rt) + "/" + vmId + "/config"
+	apiUrl := "/nodes/" + px.NodeName + "/" + string(rt) + "/" + vmId + "/config"
 	jsonData, err := px.requestFunction(px, apiUrl, http.MethodGet, nil)
 	if err != nil {
 		return VmConfig{}, err
@@ -245,7 +275,7 @@ func getTags(px *Proxmox, name string, vmConfig VmConfig, rt ResourceType) map[s
 	fqdn := hostname + "." + domain
 
 	return map[string]string{
-		"node_fqdn": px.hostname + "." + px.nodeSearchDomain,
+		"node_fqdn": px.NodeName + "." + px.nodeSearchDomain,
 		"vm_name":   name,
 		"vm_fqdn":   fqdn,
 		"vm_type":   string(rt),

plugins/inputs/proxmox/proxmox_test.go

@@ -14,6 +14,8 @@ var qemuTestData = `{"data":[{"name":"qemu1","status":"running","maxdisk":107374
 var qemuConfigTestData = `{"data":{"hostname":"qemu1","searchdomain":"test.example.com"}}`
 var lxcTestData = `{"data":[{"vmid":"111","type":"lxc","uptime":2078164,"swap":9412608,"disk":"744189952","maxmem":536870912,"mem":98500608,"maxswap":536870912,"cpu":0.00371567669193613,"status":"running","maxdisk":"5217320960","name":"container1"}]}`
 var lxcConfigTestData = `{"data":{"hostname":"container1","searchdomain":"test.example.com"}}`
+var lxcCurrentStatusTestData = `{"data":{"vmid":"111","type":"lxc","uptime":2078164,"swap":9412608,"disk":"744189952","maxmem":536870912,"mem":98500608,"maxswap":536870912,"cpu":0.00371567669193613,"status":"running","maxdisk":"5217320960","name":"container1"}}`
+var qemuCurrentStatusTestData = `{"data":{"name":"qemu1","status":"running","maxdisk":10737418240,"cpu":0.029336643550795,"vmid":"113","uptime":2159739,"disk":0,"maxmem":2147483648,"mem":1722451796}}`
 
 func performTestRequest(px *Proxmox, apiUrl string, method string, data url.Values) ([]byte, error) {
 	var bytedata = []byte("")
@@ -28,6 +30,10 @@ func performTestRequest(px *Proxmox, apiUrl string, method string, data url.Valu
 		bytedata = []byte(lxcTestData)
 	} else if strings.HasSuffix(apiUrl, "111/config") {
 		bytedata = []byte(lxcConfigTestData)
+	} else if strings.HasSuffix(apiUrl, "111/status/current") {
+		bytedata = []byte(lxcCurrentStatusTestData)
+	} else if strings.HasSuffix(apiUrl, "113/status/current") {
+		bytedata = []byte(qemuCurrentStatusTestData)
 	}
 
 	return bytedata, nil
@@ -36,12 +42,12 @@ func performTestRequest(px *Proxmox, apiUrl string, method string, data url.Valu
 func setUp(t *testing.T) *Proxmox {
 	px := &Proxmox{
 		requestFunction: performTestRequest,
+		NodeName:        "testnode",
 	}
 
 	require.NoError(t, px.Init())
 
-	// Override hostname and logger for test
+	// Override logger for test
-	px.hostname = "testnode"
 	px.Log = testutil.Logger{}
 	return px
 }

plugins/inputs/proxmox/structs.go

@@ -13,9 +13,10 @@ type Proxmox struct {
 	BaseURL         string            `toml:"base_url"`
 	APIToken        string            `toml:"api_token"`
 	ResponseTimeout internal.Duration `toml:"response_timeout"`
+	NodeName        string            `toml:"node_name"`
+
 	tls.ClientConfig
 
-	hostname         string
 	httpClient       *http.Client
 	nodeSearchDomain string
 
@@ -34,6 +35,10 @@ type VmStats struct {
 	Data []VmStat `json:"data"`
 }
 
+type VmCurrentStats struct {
+	Data VmStat `json:"data"`
+}
+
 type VmStat struct {
 	ID        string      `json:"vmid"`
 	Name      string      `json:"name"`