Add drive I/O metrics for Prometheus (#955)

Below metrics are exported: * directpv_stats_drive_ready * directpv_stats_drive_total_read_bytes * directpv_stats_drive_total_write_bytes * directpv_stats_drive_read_latency_seconds * directpv_stats_drive_write_latency_seconds * directpv_stats_drive_wait_time_seconds Fixes #839 Signed-off-by: Bala.FA <[email protected]>
minio · Oct 15, 2024 · 35aaf55 · 35aaf55
1 parent 557e925
commit 35aaf55
Show file tree

Hide file tree

Showing 3 changed files with 165 additions and 4 deletions.
diff --git a/docs/monitoring.md b/docs/monitoring.md
@@ -3,7 +3,13 @@
 DirectPV nodes export Prometheus compatible metrics data via port `10443`. The metrics data includes
 * directpv_stats_bytes_used
 * directpv_stats_bytes_total
-and categorized by labels `tenant`, `volumeID` and `node`.
+* directpv_stats_drive_ready
+* directpv_stats_drive_total_read_bytes
+* directpv_stats_drive_total_write_bytes
+* directpv_stats_drive_read_latency_seconds
+* directpv_stats_drive_write_latency_seconds
+* directpv_stats_drive_wait_time_seconds
+and categorized by labels `drive`, `tenant`, `volumeID` and `node`.
 
 To scrape data in Prometheus, each node must be accessible by port `10443`. A simple example is below
 

diff --git a/pkg/device/sysfs_linux.go b/pkg/device/sysfs_linux.go
@@ -106,3 +106,21 @@ func getHolders(name string) ([]string, error) {
 func getDMName(name string) (string, error) {
 	return readFirstLine("/sys/class/block/" + name + "/dm/name")
 }
+
+// GetStat returns statistics for a given device name.
+func GetStat(name string) (stats []uint64, err error) {
+	line, err := readFirstLine("/sys/class/block/" + name + "/stat")
+	if err != nil {
+		return nil, err
+	}
+
+	for _, token := range strings.Fields(line) {
+		ui64, err := strconv.ParseUint(token, 10, 64)
+		if err != nil {
+			return nil, err
+		}
+		stats = append(stats, ui64)
+	}
+
+	return stats, nil
+}
diff --git a/pkg/metrics/collector.go b/pkg/metrics/collector.go
@@ -18,17 +18,51 @@ package metrics
 
 import (
 	"context"
+	"fmt"
 
 	directpvtypes "github.com/minio/directpv/pkg/apis/directpv.min.io/types"
 	"github.com/minio/directpv/pkg/client"
 	"github.com/minio/directpv/pkg/consts"
+	"github.com/minio/directpv/pkg/device"
 	"github.com/minio/directpv/pkg/sys"
 	"github.com/minio/directpv/pkg/types"
+	"github.com/minio/directpv/pkg/utils"
 	"github.com/minio/directpv/pkg/xfs"
 	"github.com/prometheus/client_golang/prometheus"
 	"k8s.io/klog/v2"
 )
 
+const defaultSectorSize = 512
+
+type driveStats struct {
+	readBytes   float64
+	readTicks   float64
+	writeBytes  float64
+	writeTicks  float64
+	timeInQueue float64
+}
+
+func getDriveStats(driveName string) (*driveStats, error) {
+	stat, err := device.GetStat(driveName)
+	switch {
+	case err != nil:
+		return nil, err
+	case len(stat) == 0:
+		return nil, fmt.Errorf("unable to read stat from drive %v", driveName)
+	case len(stat) < 10:
+		return nil, fmt.Errorf("invalid stat format from drive %v", driveName)
+	}
+
+	// Refer https://www.kernel.org/doc/Documentation/block/stat.txt for meaning of each field.
+	return &driveStats{
+		readBytes:   float64(stat[2] * defaultSectorSize),
+		readTicks:   float64(stat[3]),
+		writeBytes:  float64(stat[6] * defaultSectorSize),
+		writeTicks:  float64(stat[7]),
+		timeInQueue: float64(stat[10]),
+	}, nil
+}
+
 type metricsCollector struct {
 	nodeID            directpvtypes.NodeID
 	desc              *prometheus.Desc
@@ -95,21 +129,124 @@ func (c *metricsCollector) publishVolumeStats(ctx context.Context, volume *types
 	)
 }
 
+func (c *metricsCollector) publishDriveStats(drive *types.Drive, ch chan<- prometheus.Metric) {
+	deviceID, err := c.getDeviceByFSUUID(drive.Status.FSUUID)
+	if err != nil {
+		klog.ErrorS(
+			err,
+			"unable to find device by FSUUID; "+
+				"either device is removed or run command "+
+				"`sudo udevadm control --reload-rules && sudo udevadm trigger`"+
+				" on the host to reload",
+			"FSUUID", drive.Status.FSUUID)
+		client.Eventf(
+			drive, client.EventTypeWarning, client.EventReasonMetrics,
+			"unable to find device by FSUUID %v; "+
+				"either device is removed or run command "+
+				"`sudo udevadm control --reload-rules && sudo udevadm trigger`"+
+				" on the host to reload", drive.Status.FSUUID)
+
+		return
+	}
+	deviceName := utils.TrimDevPrefix(deviceID)
+
+	status := float64(1) // Online
+	driveStat, err := getDriveStats(deviceName)
+	if err != nil {
+		klog.ErrorS(err, "unable to read drive statistics")
+		status = float64(0) // Offline
+	}
+
+	// Metrics
+	ch <- prometheus.MustNewConstMetric(
+		prometheus.NewDesc(
+			prometheus.BuildFQName(consts.AppName, "stats", "drive_ready"),
+			"Drive Online/Offline Status",
+			[]string{"drive"}, nil),
+		prometheus.GaugeValue,
+		status, drive.Name,
+	)
+
+	if driveStat == nil {
+		return
+	}
+
+	ch <- prometheus.MustNewConstMetric(
+		prometheus.NewDesc(
+			prometheus.BuildFQName(consts.AppName, "stats", "drive_total_read_bytes"),
+			"Total number of bytes read from the drive",
+			[]string{"drive"}, nil),
+		prometheus.CounterValue,
+		driveStat.readBytes, drive.Name,
+	)
+
+	ch <- prometheus.MustNewConstMetric(
+		prometheus.NewDesc(
+			prometheus.BuildFQName(consts.AppName, "stats", "drive_total_write_bytes"),
+			"Total number of bytes written to the drive",
+			[]string{"drive"}, nil),
+		prometheus.CounterValue,
+		driveStat.writeBytes, drive.Name,
+	)
+
+	// Drive Read/Write Latency
+	ch <- prometheus.MustNewConstMetric(
+		prometheus.NewDesc(
+			prometheus.BuildFQName(consts.AppName, "stats", "drive_read_latency_seconds"),
+			"Drive Read Latency",
+			[]string{"drive"}, nil),
+		prometheus.GaugeValue,
+		driveStat.readTicks/1000, drive.Name,
+	)
+
+	ch <- prometheus.MustNewConstMetric(
+		prometheus.NewDesc(
+			prometheus.BuildFQName(consts.AppName, "stats", "drive_write_latency_seconds"),
+			"Drive Write Latency",
+			[]string{"drive"}, nil),
+		prometheus.GaugeValue,
+		driveStat.writeTicks/1000, drive.Name,
+	)
+
+	// Wait Time
+	ch <- prometheus.MustNewConstMetric(
+		prometheus.NewDesc(
+			prometheus.BuildFQName(consts.AppName, "stats", "drive_wait_time_seconds"),
+			"Drive Wait Time",
+			[]string{"drive"}, nil),
+		prometheus.GaugeValue,
+		driveStat.timeInQueue/1000, drive.Name,
+	)
+}
+
 // Collect is called by Prometheus registry when collecting metrics.
 func (c *metricsCollector) Collect(ch chan<- prometheus.Metric) {
 	ctx, cancelFunc := context.WithCancel(context.Background())
 	defer cancelFunc()
 
-	resultCh := client.NewVolumeLister().
+	// Collecting volume statistics
+	volumeResultCh := client.NewVolumeLister().
 		NodeSelector([]directpvtypes.LabelValue{directpvtypes.ToLabelValue(string(c.nodeID))}).
 		List(ctx)
-	for result := range resultCh {
+	for result := range volumeResultCh {
 		if result.Err != nil {
-			return
+			break
 		}
 
 		if result.Volume.Status.TargetPath != "" {
 			c.publishVolumeStats(ctx, &result.Volume, ch)
 		}
 	}
+
+	// Collecting drive statistics
+	driveResultCh := client.NewDriveLister().
+		NodeSelector([]directpvtypes.LabelValue{directpvtypes.ToLabelValue(string(c.nodeID))}).
+		List(ctx)
+	for result := range driveResultCh {
+		if result.Err != nil {
+			break
+		}
+
+		c.publishDriveStats(&result.Drive, ch)
+	}
 }