metricbeat: suppress error when RAID metrics are enabled on non-RAID …

…system (#41825) (#41856)

* metricbeat: return partial metrics errror when RAID metrics are enabled on non-RAID system

When the Linux Metrics integration (beta) is installed with the RAID
metrics option enabled it causes an error if the host does not have a
RAID configuration. This error causes the Agent to go into a degraded
state.

This happens because we report not having `/sys/block/md*` devices as an
error, this only means that no RAID configuration is in place.

Instead, return a partial metrics error for this case. This still shows up in the
agent status but does not cause a degraded state.

* use partial metrics error

* fix linter issue with iface assertion

(cherry picked from commit 444b8e4)

Co-authored-by: Mauri de Souza Meneguzzo <[email protected]>

metricbeat/module/system/raid/blockinfo/getdev.go

@@ -19,14 +19,15 @@ package blockinfo
 
 import (
 	"fmt"
-	"io/ioutil"
 	"os"
 	"path/filepath"
+
+	"github.com/elastic/beats/v7/metricbeat/mb"
 )
 
 // ListAll lists all the multi-disk devices in a RAID array
 func ListAll(path string) ([]MDDevice, error) {
-	dir, err := ioutil.ReadDir(path)
+	dir, err := os.ReadDir(path)
 	if err != nil {
 		return nil, fmt.Errorf("could not read directory: %w", err)
 	}
@@ -44,7 +45,7 @@ func ListAll(path string) ([]MDDevice, error) {
 	}
 
 	if len(mds) == 0 {
-		return nil, fmt.Errorf("no matches from path %s", path)
+		return nil, mb.PartialMetricsError{Err: fmt.Errorf("no RAID devices found. You have probably enabled the RAID metrics on a non-RAID system.")}
 	}
 
 	return mds, nil
@@ -69,8 +70,5 @@ func getMDDevice(path string) (MDDevice, error) {
 // Right now, we're doing this by looking for an `md` directory in the device dir.
 func isMD(path string) bool {
 	_, err := os.Stat(filepath.Join(path, "md"))
-	if err != nil {
-		return false
-	}
-	return true
+	return err == nil
 }

metricbeat/module/system/raid/raid.go

@@ -41,8 +41,11 @@ type MetricSet struct {
 
 // New creates a new instance of the raid metricset.
 func New(base mb.BaseMetricSet) (mb.MetricSet, error) {
+	sys, ok := base.Module().(resolve.Resolver)
+	if !ok {
+		return nil, fmt.Errorf("unexpected module type: %T", base.Module())
+	}
 
-	sys := base.Module().(resolve.Resolver)
 	return &MetricSet{
 		BaseMetricSet: base,
 
@@ -62,7 +65,7 @@ func blockto1024(b int64) int64 {
 func (m *MetricSet) Fetch(r mb.ReporterV2) error {
 	devices, err := blockinfo.ListAll(m.mod.ResolveHostFS("/sys/block"))
 	if err != nil {
-		return fmt.Errorf("failed to parse sysfs: %w", err)
+		return fmt.Errorf("failed to list RAID devices: %w", err)
 	}
 
 	for _, blockDev := range devices {

metricbeat/module/system/raid/raid_test.go

@@ -18,10 +18,14 @@
 package raid
 
 import (
+	"errors"
+	"os"
+	"path/filepath"
 	"testing"
 
 	"github.com/stretchr/testify/assert"
 
+	"github.com/elastic/beats/v7/metricbeat/mb"
 	mbtest "github.com/elastic/beats/v7/metricbeat/mb/testing"
 	_ "github.com/elastic/beats/v7/metricbeat/module/system"
 )
@@ -46,6 +50,22 @@ func TestFetch(t *testing.T) {
 		events[0].BeatEvent("system", "raid").Fields.StringToPrint())
 }
 
+func TestFetchNoRAID(t *testing.T) {
+	// Ensure that we return partial metrics when no RAID devices are present.
+	tmpDir := t.TempDir()
+	assert.NoError(t, os.MkdirAll(filepath.Join(tmpDir, "sys/block"), 0755))
+	c := getConfig()
+	c["hostfs"] = tmpDir
+
+	f := mbtest.NewReportingMetricSetV2Error(t, c)
+	events, errs := mbtest.ReportingFetchV2Error(f)
+
+	assert.Len(t, errs, 1)
+	assert.ErrorAs(t, errors.Join(errs...), &mb.PartialMetricsError{})
+	assert.Contains(t, errors.Join(errs...).Error(), "failed to list RAID devices: no RAID devices found. You have probably enabled the RAID metrics on a non-RAID system.")
+	assert.Empty(t, events)
+}
+
 func getConfig() map[string]interface{} {
 	return map[string]interface{}{
 		"module":     "system",