logical_disk: Implement Perfdata collector (#1673)

This commit is contained in:
Jan-Otto Kröpke
2024-10-07 00:15:54 +02:00
committed by GitHub
parent efb20b1e31
commit 2ef1a5fdf1
19 changed files with 400 additions and 88 deletions

View File

@@ -0,0 +1,43 @@
package logical_disk
const (
avgDiskReadQueueLength = "Avg. Disk Read Queue Length"
avgDiskSecPerRead = "Avg. Disk sec/Read"
avgDiskSecPerTransfer = "Avg. Disk sec/Transfer"
avgDiskSecPerWrite = "Avg. Disk sec/Write"
avgDiskWriteQueueLength = "Avg. Disk Write Queue Length"
currentDiskQueueLength = "Current Disk Queue Length"
freeSpace = "Free Megabytes"
diskReadBytesPerSec = "Disk Read Bytes/sec"
diskReadsPerSec = "Disk Reads/sec"
diskWriteBytesPerSec = "Disk Write Bytes/sec"
diskWritesPerSec = "Disk Writes/sec"
percentDiskReadTime = "% Disk Read Time"
percentDiskWriteTime = "% Disk Write Time"
percentFreeSpace = "% Free Space"
percentIdleTime = "% Idle Time"
SplitIOPerSec = "Split IO/Sec"
)
// Win32_PerfRawData_PerfDisk_LogicalDisk docs:
// - https://msdn.microsoft.com/en-us/windows/hardware/aa394307(v=vs.71) - Win32_PerfRawData_PerfDisk_LogicalDisk class
// - https://msdn.microsoft.com/en-us/library/ms803973.aspx - LogicalDisk object reference.
type logicalDisk struct {
Name string
CurrentDiskQueueLength float64 `perflib:"Current Disk Queue Length"`
AvgDiskReadQueueLength float64 `perflib:"Avg. Disk Read Queue Length"`
AvgDiskWriteQueueLength float64 `perflib:"Avg. Disk Write Queue Length"`
DiskReadBytesPerSec float64 `perflib:"Disk Read Bytes/sec"`
DiskReadsPerSec float64 `perflib:"Disk Reads/sec"`
DiskWriteBytesPerSec float64 `perflib:"Disk Write Bytes/sec"`
DiskWritesPerSec float64 `perflib:"Disk Writes/sec"`
PercentDiskReadTime float64 `perflib:"% Disk Read Time"`
PercentDiskWriteTime float64 `perflib:"% Disk Write Time"`
PercentFreeSpace float64 `perflib:"% Free Space_Base"`
PercentFreeSpace_Base float64 `perflib:"Free Megabytes"`
PercentIdleTime float64 `perflib:"% Idle Time"`
SplitIOPerSec float64 `perflib:"Split IO/Sec"`
AvgDiskSecPerRead float64 `perflib:"Avg. Disk sec/Read"`
AvgDiskSecPerWrite float64 `perflib:"Avg. Disk sec/Write"`
AvgDiskSecPerTransfer float64 `perflib:"Avg. Disk sec/Transfer"`
}

View File

@@ -4,6 +4,7 @@ package logical_disk
import (
"encoding/binary"
"errors"
"fmt"
"log/slog"
"regexp"
@@ -12,9 +13,11 @@ import (
"strings"
"github.com/alecthomas/kingpin/v2"
"github.com/prometheus-community/windows_exporter/internal/perfdata"
"github.com/prometheus-community/windows_exporter/internal/perfdata/perftypes"
v1 "github.com/prometheus-community/windows_exporter/internal/perfdata/v1"
"github.com/prometheus-community/windows_exporter/internal/types"
"github.com/prometheus-community/windows_exporter/internal/utils"
"github.com/prometheus/client_golang/prometheus"
"github.com/yusufpapurcu/wmi"
"golang.org/x/sys/windows"
@@ -36,6 +39,8 @@ var ConfigDefaults = Config{
type Collector struct {
config Config
perfDataCollector perfdata.Collector
avgReadQueue *prometheus.Desc
avgWriteQueue *prometheus.Desc
freeSpace *prometheus.Desc
@@ -125,6 +130,10 @@ func (c *Collector) GetName() string {
}
func (c *Collector) GetPerfCounter(_ *slog.Logger) ([]string, error) {
if utils.PDHEnabled() {
return []string{}, nil
}
return []string{"LogicalDisk"}, nil
}
@@ -133,6 +142,34 @@ func (c *Collector) Close(_ *slog.Logger) error {
}
func (c *Collector) Build(_ *slog.Logger, _ *wmi.Client) error {
if utils.PDHEnabled() {
counters := []string{
currentDiskQueueLength,
avgDiskReadQueueLength,
avgDiskWriteQueueLength,
diskReadBytesPerSec,
diskReadsPerSec,
diskWriteBytesPerSec,
diskWritesPerSec,
percentDiskReadTime,
percentDiskWriteTime,
percentFreeSpace,
freeSpace,
percentIdleTime,
SplitIOPerSec,
avgDiskSecPerRead,
avgDiskSecPerWrite,
avgDiskSecPerTransfer,
}
var err error
c.perfDataCollector, err = perfdata.NewCollector(perfdata.V1, "LogicalDisk", perfdata.AllInstances, counters)
if err != nil {
return fmt.Errorf("failed to create LogicalDisk collector: %w", err)
}
}
c.information = prometheus.NewDesc(
prometheus.BuildFQName(types.Namespace, Name, "info"),
"A metric with a constant '1' value labeled with logical disk information",
@@ -264,6 +301,11 @@ func (c *Collector) Build(_ *slog.Logger, _ *wmi.Client) error {
// to the provided prometheus Metric channel.
func (c *Collector) Collect(ctx *types.ScrapeContext, logger *slog.Logger, ch chan<- prometheus.Metric) error {
logger = logger.With(slog.String("collector", Name))
if utils.PDHEnabled() {
return c.collectPDH(logger, ch)
}
if err := c.collect(ctx, logger, ch); err != nil {
logger.Error("failed collecting logical_disk metrics",
slog.Any("err", err),
@@ -275,32 +317,172 @@ func (c *Collector) Collect(ctx *types.ScrapeContext, logger *slog.Logger, ch ch
return nil
}
// Win32_PerfRawData_PerfDisk_LogicalDisk docs:
// - https://msdn.microsoft.com/en-us/windows/hardware/aa394307(v=vs.71) - Win32_PerfRawData_PerfDisk_LogicalDisk class
// - https://msdn.microsoft.com/en-us/library/ms803973.aspx - LogicalDisk object reference.
type logicalDisk struct {
Name string
CurrentDiskQueueLength float64 `perflib:"Current Disk Queue Length"`
AvgDiskReadQueueLength float64 `perflib:"Avg. Disk Read Queue Length"`
AvgDiskWriteQueueLength float64 `perflib:"Avg. Disk Write Queue Length"`
DiskReadBytesPerSec float64 `perflib:"Disk Read Bytes/sec"`
DiskReadsPerSec float64 `perflib:"Disk Reads/sec"`
DiskWriteBytesPerSec float64 `perflib:"Disk Write Bytes/sec"`
DiskWritesPerSec float64 `perflib:"Disk Writes/sec"`
PercentDiskReadTime float64 `perflib:"% Disk Read Time"`
PercentDiskWriteTime float64 `perflib:"% Disk Write Time"`
PercentFreeSpace float64 `perflib:"% Free Space_Base"`
PercentFreeSpace_Base float64 `perflib:"Free Megabytes"`
PercentIdleTime float64 `perflib:"% Idle Time"`
SplitIOPerSec float64 `perflib:"Split IO/Sec"`
AvgDiskSecPerRead float64 `perflib:"Avg. Disk sec/Read"`
AvgDiskSecPerWrite float64 `perflib:"Avg. Disk sec/Write"`
AvgDiskSecPerTransfer float64 `perflib:"Avg. Disk sec/Transfer"`
func (c *Collector) collectPDH(logger *slog.Logger, ch chan<- prometheus.Metric) error {
var (
err error
diskID string
info volumeInfo
)
perfData, err := c.perfDataCollector.Collect()
if err != nil {
return fmt.Errorf("failed to collect LogicalDisk metrics: %w", err)
}
if len(perfData) == 0 {
return errors.New("perflib query for LogicalDisk returned empty result set")
}
for name, volume := range perfData {
if name == "_Total" ||
c.config.VolumeExclude.MatchString(name) ||
!c.config.VolumeInclude.MatchString(name) {
continue
}
diskID, err = getDiskIDByVolume(name)
if err != nil {
logger.Warn("failed to get disk ID for "+name,
slog.Any("err", err),
)
}
info, err = getVolumeInfo(name)
if err != nil {
logger.Warn("failed to get volume information for "+name,
slog.Any("err", err),
)
}
ch <- prometheus.MustNewConstMetric(
c.information,
prometheus.GaugeValue,
1,
diskID,
info.volumeType,
name,
info.label,
info.filesystem,
info.serialNumber,
)
ch <- prometheus.MustNewConstMetric(
c.requestsQueued,
prometheus.GaugeValue,
volume[currentDiskQueueLength].FirstValue,
name,
)
ch <- prometheus.MustNewConstMetric(
c.avgReadQueue,
prometheus.GaugeValue,
volume[avgDiskReadQueueLength].FirstValue*perftypes.TicksToSecondScaleFactor,
name,
)
ch <- prometheus.MustNewConstMetric(
c.avgWriteQueue,
prometheus.GaugeValue,
volume[avgDiskWriteQueueLength].FirstValue*perftypes.TicksToSecondScaleFactor,
name,
)
ch <- prometheus.MustNewConstMetric(
c.readBytesTotal,
prometheus.CounterValue,
volume[diskReadBytesPerSec].FirstValue,
name,
)
ch <- prometheus.MustNewConstMetric(
c.readsTotal,
prometheus.CounterValue,
volume[diskReadsPerSec].FirstValue,
name,
)
ch <- prometheus.MustNewConstMetric(
c.writeBytesTotal,
prometheus.CounterValue,
volume[diskWriteBytesPerSec].FirstValue,
name,
)
ch <- prometheus.MustNewConstMetric(
c.writesTotal,
prometheus.CounterValue,
volume[diskWritesPerSec].FirstValue,
name,
)
ch <- prometheus.MustNewConstMetric(
c.readTime,
prometheus.CounterValue,
volume[percentDiskReadTime].FirstValue,
name,
)
ch <- prometheus.MustNewConstMetric(
c.writeTime,
prometheus.CounterValue,
volume[percentDiskWriteTime].FirstValue,
name,
)
ch <- prometheus.MustNewConstMetric(
c.freeSpace,
prometheus.GaugeValue,
volume[freeSpace].FirstValue*1024*1024,
name,
)
ch <- prometheus.MustNewConstMetric(
c.totalSpace,
prometheus.GaugeValue,
volume[percentFreeSpace].FirstValue*1024*1024,
name,
)
ch <- prometheus.MustNewConstMetric(
c.idleTime,
prometheus.CounterValue,
volume[percentIdleTime].FirstValue,
name,
)
ch <- prometheus.MustNewConstMetric(
c.splitIOs,
prometheus.CounterValue,
volume[SplitIOPerSec].FirstValue,
name,
)
ch <- prometheus.MustNewConstMetric(
c.readLatency,
prometheus.CounterValue,
volume[avgDiskSecPerRead].FirstValue*perftypes.TicksToSecondScaleFactor,
name,
)
ch <- prometheus.MustNewConstMetric(
c.writeLatency,
prometheus.CounterValue,
volume[avgDiskSecPerWrite].FirstValue*perftypes.TicksToSecondScaleFactor,
name,
)
ch <- prometheus.MustNewConstMetric(
c.readWriteLatency,
prometheus.CounterValue,
volume[avgDiskSecPerTransfer].FirstValue*perftypes.TicksToSecondScaleFactor,
name,
)
}
return nil
}
func (c *Collector) collect(ctx *types.ScrapeContext, logger *slog.Logger, ch chan<- prometheus.Metric) error {
logger = logger.With(slog.String("collector", Name))
var (
err error
diskID string

View File

@@ -6,6 +6,7 @@ import (
"github.com/alecthomas/kingpin/v2"
"github.com/prometheus-community/windows_exporter/internal/collector/logical_disk"
"github.com/prometheus-community/windows_exporter/internal/testutils"
"github.com/prometheus-community/windows_exporter/internal/types"
)
func BenchmarkCollector(b *testing.B) {
@@ -14,3 +15,9 @@ func BenchmarkCollector(b *testing.B) {
kingpin.CommandLine.GetArg("collector.logical_disk.volume-include").StringVar(&localVolumeInclude)
testutils.FuncBenchmarkCollector(b, "logical_disk", logical_disk.NewWithFlags)
}
func TestCollector(t *testing.T) {
testutils.TestCollector(t, logical_disk.New, &logical_disk.Config{
VolumeInclude: types.RegExpAny,
})
}