Merge pull request #30 from NETWAYS/error_on_non_existing_pressure

Enhanced error checking and more elaborate error message  for psi and netdev

.golangci.yml

@@ -15,6 +15,12 @@ issues:
         - gocognit
         - maintidx
         - gocyclo
+    - path: 'cmd/psi.go'
+      linters:
+        - funlen
+        - gocognit
+        - gocyclo
+        - maintidx
     - path: 'internal/sensors/sensors.go'
       linters:
         - funlen

cmd/psi.go

@@ -1,7 +1,9 @@
 package cmd
 
 import (
+	"errors"
 	"fmt"
+	"os"
 
 	"github.com/NETWAYS/check_system_basics/internal/common/thresholds"
 	"github.com/NETWAYS/check_system_basics/internal/psi"
@@ -232,14 +234,22 @@ func init() {
 	psiFs.BoolVar(&config.IncludeIO, "include-io", false, "Include IO values explicitly (by default all are included)")
 }
 
-//nolint:funlen,gocognit,gocyclo
 func checkPsiCPUPressure(config *psiConfig) result.PartialResult {
+	var cpuCheck result.PartialResult
+	_ = cpuCheck.SetDefaultState(check.OK)
+
 	psiCPU, err := psi.ReadCPUPressure()
 	if err != nil {
+		if errors.Is(err, os.ErrNotExist) {
+			_ = cpuCheck.SetState(check.Unknown)
+			cpuCheck.Output = "CPU pressure file not found. Perhaps the PSI interface is not active on this system? It might be necessary to change the kernel config"
+
+			return cpuCheck
+		}
+
 		check.ExitError(err)
 	}
 
-	var cpuCheck result.PartialResult
 	cpuCheck.Perfdata = *psiCPU.Perfdata()
 	_ = cpuCheck.SetState(check.OK)
 
@@ -358,15 +368,22 @@ func checkPsiCPUPressure(config *psiConfig) result.PartialResult {
 	return cpuCheck
 }
 
-//nolint:funlen,gocognit
 func checkPsiIoPressure(config *psiConfig) result.PartialResult {
+	var ioCheck result.PartialResult
+	_ = ioCheck.SetDefaultState(check.OK)
+
 	psiIo, err := psi.ReadIoPressure()
 	if err != nil {
+		if errors.Is(err, os.ErrNotExist) {
+			_ = ioCheck.SetState(check.Unknown)
+			ioCheck.Output = "IO pressure file not found. Perhaps the PSI interface is not active on this system? It might be necessary to change the kernel config"
+
+			return ioCheck
+		}
+
 		check.ExitError(err)
 	}
 
-	var ioCheck result.PartialResult
-	_ = ioCheck.SetState(check.OK)
 	ioCheck.Perfdata = *psiIo.Perfdata()
 
 	//nolint:nestif
@@ -479,15 +496,22 @@ func checkPsiIoPressure(config *psiConfig) result.PartialResult {
 	return ioCheck
 }
 
-//nolint:funlen,gocognit
 func checkPsiMemoryPressure(config *psiConfig) result.PartialResult {
+	var memoryCheck result.PartialResult
+	_ = memoryCheck.SetDefaultState(check.OK)
+
 	psiMemory, err := psi.ReadMemoryPressure()
 	if err != nil {
+		if errors.Is(err, os.ErrNotExist) {
+			_ = memoryCheck.SetState(check.Unknown)
+			memoryCheck.Output = "IO pressure file not found. Perhaps the PSI interface is not active on this system? It might be necessary to change the kernel config"
+
+			return memoryCheck
+		}
+
 		check.ExitError(err)
 	}
 
-	var memoryCheck result.PartialResult
-	_ = memoryCheck.SetState(check.OK)
 	memoryCheck.Perfdata = *psiMemory.Perfdata()
 
 	//nolint:nestif

internal/netdev/netdev.go

@@ -2,6 +2,7 @@ package netdev
 
 import (
 	"os"
+	"path"
 	"strconv"
 	"strings"
 
@@ -18,6 +19,10 @@ const (
 	Unknown        = 4
 )
 
+const (
+	netDevicePath = "/sys/class/net"
+)
+
 func TranslateIfaceState(state uint) string {
 	switch state {
 	case Up:
@@ -140,7 +145,7 @@ func GetAllInterfaces() ([]IfaceData, error) {
 }
 
 func listInterfaces() ([]string, error) {
-	file, err := os.Open("/sys/class/net")
+	file, err := os.Open(netDevicePath)
 	if err != nil {
 		return []string{}, err
 	}
@@ -150,7 +155,24 @@ func listInterfaces() ([]string, error) {
 		return []string{}, err
 	}
 
-	return devices, nil
+	result := make([]string, 0, len(devices))
+
+	for idx := range devices {
+		fileInfo, err := os.Stat(path.Join(netDevicePath, devices[idx]))
+		if err != nil {
+			// Could not stat file there, not sure if this can be handled usefully. Just die for now.
+			return []string{}, err
+		}
+
+		// Filter here, to get only entries which end in a directory (the entries itself are mostly symlinks)
+		// Despite of what man 5 sysfs says, not all the files represent network interfaces, but might also
+		// be normal files (at least a "bonding_masters" file was found on a system with bonded interfaces)
+		if fileInfo.Mode().IsDir() {
+			result = append(result, devices[idx])
+		}
+	}
+
+	return result, nil
 }
 
 // getInterfaceState receives the name of an interfaces and returns
@@ -159,9 +181,9 @@ func listInterfaces() ([]string, error) {
 // @result = 2 => Interface is down
 // @result = 3 => Interface is unknown or state of the interface is unknown for some reason
 func getInterfaceState(data *IfaceData) error {
-	basePath := "/sys/class/net/" + data.Name
+	basePath := path.Join(netDevicePath, data.Name)
 
-	bytes, err := os.ReadFile(basePath + "/operstate")
+	bytes, err := os.ReadFile(path.Join(basePath, "operstate"))
 	if err != nil {
 		return err
 	}
@@ -188,12 +210,12 @@ func getInterfaceState(data *IfaceData) error {
 // Get interfaces statistics
 // @result: ifaceStats, err
 func getInfacesStatistics(data *IfaceData) error {
-	basePath := "/sys/class/net/" + data.Name + "/statistics"
+	basePath := path.Join(netDevicePath, data.Name, "statistics")
 
 	var val uint64
 
 	for idx, stat := range GetIfaceStatNames() {
-		numberBytes, err := os.ReadFile(basePath + "/" + stat)
+		numberBytes, err := os.ReadFile(path.Join(basePath, stat))
 
 		if err != nil {
 			return err