improving the nvme exporter

Signed-off-by: GitHub <[email protected]>
Signed-off-by: Your Name <[email protected]>

Author: ShashwatHiregoudar

collector/fixtures/e2e-64k-page-output.txt

@@ -2825,7 +2825,7 @@ node_nfsd_server_rpcs_total 18628
 node_nfsd_server_threads 8
 # HELP node_nvme_info Non-numeric data from /sys/class/nvme/<device>, value is always 1.
 # TYPE node_nvme_info gauge
-node_nvme_info{device="nvme0",firmware_revision="1B2QEXP7",model="Samsung SSD 970 PRO 512GB",serial="S680HF8N190894I",state="live"} 1
+node_nvme_info{cntlid="1997",device="nvme0",firmware_revision="1B2QEXP7",model="Samsung SSD 970 PRO 512GB",serial="S680HF8N190894I",state="live"} 1
 # HELP node_os_info A metric with a constant '1' value labeled by build_id, id, id_like, image_id, image_version, name, pretty_name, variant, variant_id, version, version_codename, version_id.
 # TYPE node_os_info gauge
 node_os_info{build_id="",id="ubuntu",id_like="debian",image_id="",image_version="",name="Ubuntu",pretty_name="Ubuntu 20.04.2 LTS",variant="",variant_id="",version="20.04.2 LTS (Focal Fossa)",version_codename="focal",version_id="20.04"} 1

collector/fixtures/e2e-output.txt

@@ -2847,7 +2847,7 @@ node_nfsd_server_rpcs_total 18628
 node_nfsd_server_threads 8
 # HELP node_nvme_info Non-numeric data from /sys/class/nvme/<device>, value is always 1.
 # TYPE node_nvme_info gauge
-node_nvme_info{device="nvme0",firmware_revision="1B2QEXP7",model="Samsung SSD 970 PRO 512GB",serial="S680HF8N190894I",state="live"} 1
+node_nvme_info{cntlid="1997",device="nvme0",firmware_revision="1B2QEXP7",model="Samsung SSD 970 PRO 512GB",serial="S680HF8N190894I",state="live"} 1
 # HELP node_os_info A metric with a constant '1' value labeled by build_id, id, id_like, image_id, image_version, name, pretty_name, variant, variant_id, version, version_codename, version_id.
 # TYPE node_os_info gauge
 node_os_info{build_id="",id="ubuntu",id_like="debian",image_id="",image_version="",name="Ubuntu",pretty_name="Ubuntu 20.04.2 LTS",variant="",variant_id="",version="20.04.2 LTS (Focal Fossa)",version_codename="focal",version_id="20.04"} 1

collector/fixtures/sys.ttar

@@ -2179,6 +2179,11 @@ Lines: 1
 1B2QEXP7
 Mode: 444
 # ttar - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+Path: sys/class/nvme/nvme0/cntlid
+Lines: 1
+1997
+Mode: 666
+# ttar - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
 Path: sys/class/nvme/nvme0/model
 Lines: 1
 Samsung SSD 970 PRO 512GB

collector/nvme_linux.go

@@ -19,16 +19,26 @@ package collector
 import (
 	"errors"
 	"fmt"
-	"log/slog"
 	"os"
+	"path/filepath"
+	"regexp"
+	"strconv"
+	"strings"
 
 	"github.com/prometheus/client_golang/prometheus"
 	"github.com/prometheus/procfs/sysfs"
+	"log/slog"
 )
 
 type nvmeCollector struct {
-	fs     sysfs.FS
-	logger *slog.Logger
+	fs                             sysfs.FS
+	logger                         *slog.Logger
+	namespaceInfo                  *prometheus.Desc
+	namespaceCapacityBytes         *prometheus.Desc
+	namespaceSizeBytes             *prometheus.Desc
+	namespaceUsedBytes             *prometheus.Desc
+	namespaceLogicalBlockSizeBytes *prometheus.Desc
+	info                           *prometheus.Desc
 }
 
 func init() {
@@ -42,9 +52,51 @@ func NewNVMeCollector(logger *slog.Logger) (Collector, error) {
 		return nil, fmt.Errorf("failed to open sysfs: %w", err)
 	}
 
+	info := prometheus.NewDesc(
+		prometheus.BuildFQName(namespace, "nvme", "info"),
+		"Non-numeric data from /sys/class/nvme/<device>, value is always 1.",
+		[]string{"device", "firmware_revision", "model", "serial", "state", "cntlid"},
+		nil,
+	)
+	namespaceInfo := prometheus.NewDesc(
+		prometheus.BuildFQName(namespace, "nvme", "namespace_info"),
+		"Information about NVMe namespaces. Value is always 1",
+		[]string{"device", "nsid", "ana_state"}, nil,
+	)
+
+	namespaceCapacityBytes := prometheus.NewDesc(
+		prometheus.BuildFQName(namespace, "nvme", "namespace_capacity_bytes"),
+		"Capacity of the NVMe namespace in bytes. Computed as namespace_size * namespace_logical_block_size",
+		[]string{"device", "nsid"}, nil,
+	)
+
+	namespaceSizeBytes := prometheus.NewDesc(
+		prometheus.BuildFQName(namespace, "nvme", "namespace_size_bytes"),
+		"Size of the NVMe namespace in bytes. Available in /sys/class/nvme/<device>/<namespace>/size",
+		[]string{"device", "nsid"}, nil,
+	)
+
+	namespaceUsedBytes := prometheus.NewDesc(
+		prometheus.BuildFQName(namespace, "nvme", "namespace_used_bytes"),
+		"Used space of the NVMe namespace in bytes. Available in /sys/class/nvme/<device>/<namespace>/nuse",
+		[]string{"device", "nsid"}, nil,
+	)
+
+	namespaceLogicalBlockSizeBytes := prometheus.NewDesc(
+		prometheus.BuildFQName(namespace, "nvme", "namespace_logical_block_size_bytes"),
+		"Logical block size of the NVMe namespace in bytes. Usually 4Kb. Available in /sys/class/nvme/<device>/<namespace>/queue/logical_block_size",
+		[]string{"device", "nsid"}, nil,
+	)
+
 	return &nvmeCollector{
-		fs:     fs,
-		logger: logger,
+		fs:                             fs,
+		logger:                         logger,
+		namespaceInfo:                  namespaceInfo,
+		namespaceCapacityBytes:         namespaceCapacityBytes,
+		namespaceSizeBytes:             namespaceSizeBytes,
+		namespaceUsedBytes:             namespaceUsedBytes,
+		namespaceLogicalBlockSizeBytes: namespaceLogicalBlockSizeBytes,
+		info:                           info,
 	}, nil
 }
 
@@ -59,14 +111,92 @@ func (c *nvmeCollector) Update(ch chan<- prometheus.Metric) error {
 	}
 
 	for _, device := range devices {
-		infoDesc := prometheus.NewDesc(
-			prometheus.BuildFQName(namespace, "nvme", "info"),
-			"Non-numeric data from /sys/class/nvme/<device>, value is always 1.",
-			[]string{"device", "firmware_revision", "model", "serial", "state"},
-			nil,
-		)
 		infoValue := 1.0
-		ch <- prometheus.MustNewConstMetric(infoDesc, prometheus.GaugeValue, infoValue, device.Name, device.FirmwareRevision, device.Model, device.Serial, device.State)
+
+		devicePath := filepath.Join(*sysPath, "class/nvme", device.Name)
+		cntlid, err := readUintFromFile(filepath.Join(devicePath, "cntlid"))
+		if err != nil {
+			c.logger.Debug("failed to read cntlid", "device", device.Name, "err", err)
+		}
+		ch <- prometheus.MustNewConstMetric(c.info, prometheus.GaugeValue, infoValue, device.Name, device.FirmwareRevision, device.Model, device.Serial, device.State, strconv.FormatUint(cntlid, 10))
+		// Find namespace directories.
+		namespacePaths, err := filepath.Glob(filepath.Join(devicePath, "nvme[0-9]*c[0-9]*n[0-9]*"))
+		if err != nil {
+			c.logger.Error("failed to list NVMe namespaces", "device", device.Name, "err", err)
+			continue
+		}
+		re := regexp.MustCompile(`nvme[0-9]+c[0-9]+n([0-9]+)`)
+
+		for _, namespacePath := range namespacePaths {
+
+			// Read namespace data.
+			match := re.FindStringSubmatch(filepath.Base(namespacePath))
+			if len(match) == 0 {
+				continue
+			}
+			nsid := match[1]
+			nuse, err := readUintFromFile(filepath.Join(namespacePath, "nuse"))
+			if err != nil {
+				c.logger.Debug("failed to read nuse", "device", device.Name, "namespace", match[0], "err", err)
+			}
+			nsze, err := readUintFromFile(filepath.Join(namespacePath, "size"))
+			if err != nil {
+				c.logger.Debug("failed to read size", "device", device.Name, "namespace", match[0], "err", err)
+			}
+			lbaSize, err := readUintFromFile(filepath.Join(namespacePath, "queue", "logical_block_size"))
+			if err != nil {
+				c.logger.Debug("failed to read queue/logical_block_size", "device", device.Name, "namespace", match[0], "err", err)
+			}
+			ncap := nsze * lbaSize
+			anaState := "unknown"
+			anaStateSysfs, err := os.ReadFile(filepath.Join(namespacePath, "ana_state"))
+			if err == nil {
+				anaState = strings.TrimSpace(string(anaStateSysfs))
+			} else {
+				c.logger.Debug("failed to read ana_state", "device", device.Name, "namespace", match[0], "err", err)
+			}
+
+			ch <- prometheus.MustNewConstMetric(
+				c.namespaceInfo,
+				prometheus.GaugeValue,
+				1.0,
+				device.Name,
+				nsid,
+				anaState,
+			)
+
+			ch <- prometheus.MustNewConstMetric(
+				c.namespaceCapacityBytes,
+				prometheus.GaugeValue,
+				float64(ncap),
+				device.Name,
+				nsid,
+			)
+
+			ch <- prometheus.MustNewConstMetric(
+				c.namespaceSizeBytes,
+				prometheus.GaugeValue,
+				float64(nsze),
+				device.Name,
+				nsid,
+			)
+
+			ch <- prometheus.MustNewConstMetric(
+				c.namespaceUsedBytes,
+				prometheus.GaugeValue,
+				float64(nuse*lbaSize),
+				device.Name,
+				nsid,
+			)
+
+			ch <- prometheus.MustNewConstMetric(
+				c.namespaceLogicalBlockSizeBytes,
+				prometheus.GaugeValue,
+				float64(lbaSize),
+				device.Name,
+				nsid,
+			)
+		}
 	}
 
 	return nil