rewriting event log metrics

README.md

@@ -62,7 +62,7 @@ metrics:
   system: true
   sensors: true
   power: true
-  sel: false
+  events: false
   storage: false
   memory: false
   network: false
@@ -125,11 +125,11 @@ idrac_power_control_avg_consumed_watts{id="0",name="System Power Control"} 166
 idrac_power_control_interval_in_minutes{id="0",name="System Power Control"} 1
 ```
 
-### System Event Log
-On iDRAC only, the system event log can also be exported. This is not exactly an ordinary metric, but it is often convenient to be informed about new entries in the event log. The value of this metric is the unix timestamp for when the entry was created (as reported by iDRAC).
+### Event Log
+This is not exactly an ordinary metric, but it is often convenient to be informed about new entries in the event log. The value of this metric is the unix timestamp for when the entry was created.
 
 ```text
-idrac_sel_entry{id="1",message="The process of installing an operating system or hypervisor is successfully completed",component="BaseOSBoot/InstallationStatus",severity="OK"} 1631175352
+idrac_log_entry{id="1",message="The process of installing an operating system or hypervisor is successfully completed",severity="OK"} 1631175352
 ```
 
 ### Storage

internal/collector/client.go

@@ -38,6 +38,7 @@ type Client struct {
 	storagePath string
 	memoryPath  string
 	networkPath string
+	eventPath   string
 }
 
 func newHttpClient() *http.Client {
@@ -137,6 +138,16 @@ func (client *Client) findAllEndpoints() error {
 		}
 	}
 
+	// Path for event log
+	switch client.vendor {
+	case DELL:
+		client.eventPath = "/redfish/v1/Managers/iDRAC.Embedded.1/Logs/Sel"
+	case LENOVO:
+		client.eventPath = "/redfish/v1/Systems/1/LogServices/PlatformLog/Entries"
+	case HPE:
+		client.eventPath = "/redfish/v1/Systems/1/LogServices/IML/Entries"
+	}
+
 	return nil
 }
 
@@ -287,23 +298,37 @@ func (client *Client) RefreshPower(mc *Collector, ch chan<- prometheus.Metric) e
 	return nil
 }
 
-func (client *Client) RefreshIdracSel(mc *Collector, ch chan<- prometheus.Metric) error {
-	if client.vendor != DELL {
+func (client *Client) RefreshEventLog(mc *Collector, ch chan<- prometheus.Metric) error {
+	if client.eventPath == "" {
 		return nil
 	}
 
-	resp := IdracSelResponse{}
-	err := client.redfishGet(redfishRootPath+"/Managers/iDRAC.Embedded.1/Logs/Sel", &resp)
+	resp := EventLogResponse{}
+	err := client.redfishGet(client.eventPath, &resp)
 	if err != nil {
 		return err
 	}
 
+	level := config.Config.Event.SeverityLevel
+	maxage := config.Config.Event.MaxAgeSeconds
+
 	for _, e := range resp.Members {
-		st := string(e.SensorType)
-		if st == "" {
-			st = "Unknown"
+		t, err := time.Parse(time.RFC3339, e.Created)
+		if err != nil {
+			continue
+		}
+
+		d := time.Since(t)
+		if d.Seconds() > maxage {
+			continue
+		}
+
+		severity := health2value(e.Severity)
+		if severity < level {
+			continue
 		}
-		ch <- mc.NewSelEntry(e.Id, e.Message, st, e.Severity, e.Created)
+
+		ch <- mc.NewEventLogEntry(e.Id, e.Message, e.Severity, t)
 	}
 
 	return nil

internal/collector/collector.go

@@ -61,7 +61,7 @@ type Collector struct {
 	PowerControlInterval         *prometheus.Desc
 
 	// System event log
-	SelEntry *prometheus.Desc
+	EventLogEntry *prometheus.Desc
 
 	// Disk drives
 	DriveInfo     *prometheus.Desc
@@ -210,10 +210,10 @@ func NewCollector() *Collector {
 			"Interval for measurements of power control system",
 			[]string{"id", "name"}, nil,
 		),
-		SelEntry: prometheus.NewDesc(
-			prometheus.BuildFQName(prefix, "sel", "entry"),
+		EventLogEntry: prometheus.NewDesc(
+			prometheus.BuildFQName(prefix, "log", "entry"),
 			"Entry from the system event log",
-			[]string{"id", "message", "component", "severity"}, nil,
+			[]string{"id", "message", "severity"}, nil,
 		),
 		DriveInfo: prometheus.NewDesc(
 			prometheus.BuildFQName(prefix, "drive", "info"),
@@ -310,7 +310,7 @@ func (collector *Collector) Describe(ch chan<- *prometheus.Desc) {
 	ch <- collector.PowerControlMaxConsumedWatts
 	ch <- collector.PowerControlAvgConsumedWatts
 	ch <- collector.PowerControlInterval
-	ch <- collector.SelEntry
+	ch <- collector.EventLogEntry
 	ch <- collector.DriveInfo
 	ch <- collector.DriveHealth
 	ch <- collector.DriveCapacity
@@ -372,10 +372,10 @@ func (collector *Collector) Collect(ch chan<- prometheus.Metric) {
 		}()
 	}
 
-	if config.Config.Collect.SEL {
+	if config.Config.Collect.Events {
 		wg.Add(1)
 		go func() {
-			err := collector.client.RefreshIdracSel(collector, ch)
+			err := collector.client.RefreshEventLog(collector, ch)
 			if err != nil {
 				collector.errors.Add(1)
 			}

internal/collector/metrics.go

@@ -8,7 +8,7 @@ import (
 	"github.com/prometheus/client_golang/prometheus"
 )
 
-func health2value(health string) float64 {
+func health2value(health string) int {
 	switch health {
 	case "OK":
 		return 0
@@ -20,7 +20,7 @@ func health2value(health string) float64 {
 	return 10
 }
 
-func linkstatus2value(status string) float64 {
+func linkstatus2value(status string) int {
 	switch status {
 	case "Up", "LinkUp":
 		return 1
@@ -45,7 +45,7 @@ func (mc *Collector) NewSystemHealth(health string) prometheus.Metric {
 	return prometheus.MustNewConstMetric(
 		mc.SystemHealth,
 		prometheus.GaugeValue,
-		value,
+		float64(value),
 		health,
 	)
 }
@@ -117,7 +117,7 @@ func (mc *Collector) NewSensorsFanHealth(id, name, health string) prometheus.Met
 	return prometheus.MustNewConstMetric(
 		mc.SensorsFanHealth,
 		prometheus.GaugeValue,
-		value,
+		float64(value),
 		id,
 		name,
 		health,
@@ -140,7 +140,7 @@ func (mc *Collector) NewPowerSupplyHealth(health, id string) prometheus.Metric {
 	return prometheus.MustNewConstMetric(
 		mc.PowerSupplyHealth,
 		prometheus.GaugeValue,
-		value,
+		float64(value),
 		id,
 		health,
 	)
@@ -251,14 +251,13 @@ func (mc *Collector) NewPowerControlInterval(interval int, id, name string) prom
 	)
 }
 
-func (mc *Collector) NewSelEntry(id string, message string, component string, severity string, created time.Time) prometheus.Metric {
+func (mc *Collector) NewEventLogEntry(id string, message string, severity string, created time.Time) prometheus.Metric {
 	return prometheus.MustNewConstMetric(
-		mc.SelEntry,
+		mc.EventLogEntry,
 		prometheus.CounterValue,
 		float64(created.Unix()),
 		id,
 		message,
-		component,
 		severity,
 	)
 }
@@ -292,7 +291,7 @@ func (mc *Collector) NewDriveHealth(id, health string) prometheus.Metric {
 	return prometheus.MustNewConstMetric(
 		mc.DriveHealth,
 		prometheus.GaugeValue,
-		value,
+		float64(value),
 		id,
 		health,
 	)
@@ -336,7 +335,7 @@ func (mc *Collector) NewMemoryModuleHealth(id, health string) prometheus.Metric
 	return prometheus.MustNewConstMetric(
 		mc.MemoryModuleHealth,
 		prometheus.GaugeValue,
-		value,
+		float64(value),
 		id,
 		health,
 	)
@@ -365,7 +364,7 @@ func (mc *Collector) NewNetworkInterfaceHealth(id, health string) prometheus.Met
 	return prometheus.MustNewConstMetric(
 		mc.NetworkInterfaceHealth,
 		prometheus.GaugeValue,
-		value,
+		float64(value),
 		id,
 		health,
 	)
@@ -376,7 +375,7 @@ func (mc *Collector) NewNetworkPortHealth(iface, id, health string) prometheus.M
 	return prometheus.MustNewConstMetric(
 		mc.NetworkPortHealth,
 		prometheus.GaugeValue,
-		value,
+		float64(value),
 		iface,
 		id,
 		health,
@@ -398,7 +397,7 @@ func (mc *Collector) NewNetworkPortLinkUp(iface, id, status string) prometheus.M
 	return prometheus.MustNewConstMetric(
 		mc.NetworkPortLinkUp,
 		prometheus.GaugeValue,
-		value,
+		float64(value),
 		iface,
 		id,
 		status,

internal/collector/model.go

@@ -2,7 +2,6 @@ package collector
 
 import (
 	"strconv"
-	"time"
 )
 
 const (
@@ -472,21 +471,21 @@ func (psu *PowerSupplyUnit) GetOutputPower() float64 {
 	return psu.LastPowerOutputWatts
 }
 
-type IdracSelResponse struct {
+type EventLogResponse struct {
 	Name        string `json:"Name"`
 	Description string `json:"Description"`
 	Members     []struct {
-		Id           string    `json:"Id"`
-		Name         string    `json:"Name"`
-		Created      time.Time `json:"Created"`
-		Description  string    `json:"Description"`
-		EntryCode    xstring   `json:"EntryCode"`
-		EntryType    string    `json:"EntryType"`
-		Message      string    `json:"Message"`
-		MessageArgs  []any     `json:"MessageArgs"`
-		MessageId    string    `json:"MessageId"`
-		SensorNumber int       `json:"SensorNumber"`
-		SensorType   xstring   `json:"SensorType"`
-		Severity     string    `json:"Severity"`
+		Id           string  `json:"Id"`
+		Name         string  `json:"Name"`
+		Created      string  `json:"Created"`
+		Description  string  `json:"Description"`
+		EntryCode    xstring `json:"EntryCode"`
+		EntryType    string  `json:"EntryType"`
+		Message      string  `json:"Message"`
+		MessageArgs  []any   `json:"MessageArgs"`
+		MessageId    string  `json:"MessageId"`
+		SensorNumber int     `json:"SensorNumber"`
+		SensorType   xstring `json:"SensorType"`
+		Severity     string  `json:"Severity"`
 	} `json:"Members"`
 }

internal/config/config.go

@@ -3,6 +3,8 @@ package config
 import (
 	"math"
 	"os"
+	"strings"
+	"time"
 
 	"github.com/mrlhansen/idrac_exporter/internal/log"
 	"gopkg.in/yaml.v2"
@@ -54,6 +56,7 @@ func ReadConfig(filename string) {
 
 	readConfigEnv()
 
+	// main section
 	if Config.Address == "" {
 		Config.Address = "0.0.0.0"
 	}
@@ -74,6 +77,7 @@ func ReadConfig(filename string) {
 		Config.MetricsPrefix = "idrac"
 	}
 
+	// hosts section
 	if len(Config.Hosts) == 0 {
 		log.Fatal("Invalid configuration: empty section: hosts")
 	}
@@ -87,4 +91,26 @@ func ReadConfig(filename string) {
 		}
 		v.Hostname = k
 	}
+
+	// events section
+	switch strings.ToLower(Config.Event.Severity) {
+	case "ok":
+		Config.Event.SeverityLevel = 0
+	case "warning", "":
+		Config.Event.SeverityLevel = 1
+	case "critical":
+		Config.Event.SeverityLevel = 2
+	default:
+		log.Fatal("Invalid configuration: invalid value: %s", Config.Event.Severity)
+	}
+
+	if Config.Event.MaxAge == "" {
+		Config.Event.MaxAge = "168h"
+	}
+
+	t, err := time.ParseDuration(Config.Event.MaxAge)
+	if err != nil {
+		log.Fatal("Invalid configuration: unable to parse duration: %v", err)
+	}
+	Config.Event.MaxAgeSeconds = t.Seconds()
 }

internal/config/env.go

@@ -49,14 +49,16 @@ func readConfigEnv() {
 	getEnvString("CONFIG_METRICS_PREFIX", &Config.MetricsPrefix)
 	getEnvString("CONFIG_DEFAULT_USERNAME", &username)
 	getEnvString("CONFIG_DEFAULT_PASSWORD", &password)
+	getEnvString("CONFIG_EVENT_SEVERITY", &Config.Event.Severity)
+	getEnvString("CONFIG_EVENT_MAXAGE", &Config.Event.MaxAge)
 
 	getEnvUint("CONFIG_PORT", &Config.Port)
 	getEnvUint("CONFIG_TIMEOUT", &Config.Timeout)
 	getEnvUint("CONFIG_RETRIES", &Config.Retries)
 
 	getEnvBool("CONFIG_METRICS_SYSTEM", &Config.Collect.System)
 	getEnvBool("CONFIG_METRICS_SENSORS", &Config.Collect.Sensors)
-	getEnvBool("CONFIG_METRICS_SEL", &Config.Collect.SEL)
+	getEnvBool("CONFIG_METRICS_EVENTS", &Config.Collect.Events)
 	getEnvBool("CONFIG_METRICS_POWER", &Config.Collect.Power)
 	getEnvBool("CONFIG_METRICS_STORAGE", &Config.Collect.Storage)
 	getEnvBool("CONFIG_METRICS_MEMORY", &Config.Collect.Memory)

internal/config/model.go

@@ -11,19 +11,27 @@ type HostConfig struct {
 type CollectConfig struct {
 	System  bool `yaml:"system"`
 	Sensors bool `yaml:"sensors"`
-	SEL     bool `yaml:"sel"`
+	Events  bool `yaml:"events"`
 	Power   bool `yaml:"power"`
 	Storage bool `yaml:"storage"`
 	Memory  bool `yaml:"memory"`
 	Network bool `yaml:"network"`
 }
 
+type EventConfig struct {
+	Severity      string `yaml:"severity"`
+	MaxAge        string `yaml:"maxage"`
+	SeverityLevel int
+	MaxAgeSeconds float64
+}
+
 type RootConfig struct {
 	mutex         sync.Mutex
 	Address       string                 `yaml:"address"`
 	Port          uint                   `yaml:"port"`
 	MetricsPrefix string                 `yaml:"metrics_prefix"`
 	Collect       CollectConfig          `yaml:"metrics"`
+	Event         EventConfig            `yaml:"events"`
 	Timeout       uint                   `yaml:"timeout"`
 	Retries       uint                   `yaml:"retries"`
 	Hosts         map[string]*HostConfig `yaml:"hosts"`