From 14d5e6716f890115b394922e90d3a600a155e582 Mon Sep 17 00:00:00 2001
From: Aleksandr Snisarenko <a.snisarenko@petshop.ru>
Date: Thu, 12 Jan 2023 17:18:28 +0300
Subject: [PATCH] added chassis cooling fault and drive fault metrics

---
 collector_chassis.go | 24 ++++++++++++++++++++++++
 freeipmi/freeipmi.go | 30 ++++++++++++++++++++++++++++++
 2 files changed, 54 insertions(+)

diff --git a/collector_chassis.go b/collector_chassis.go
index d18aeaa..55bc5b4 100644
--- a/collector_chassis.go
+++ b/collector_chassis.go
@@ -31,6 +31,18 @@ var (
 		[]string{},
 		nil,
 	)
+	chassisDriveFaultDesc = prometheus.NewDesc(
+		prometheus.BuildFQName(namespace, "chassis", "drive_fault_state"),
+		"Current drive fault state (1=false, 0=true).",
+		[]string{},
+		nil,
+	)
+	chassisCoolingFaultDesc = prometheus.NewDesc(
+		prometheus.BuildFQName(namespace, "chassis", "cooling_fault_state"),
+		"Current Cooling/fan fault state (1=false, 0=true).",
+		[]string{},
+		nil,
+	)
 )
 
 type ChassisCollector struct{}
@@ -49,6 +61,8 @@ func (c ChassisCollector) Args() []string {
 
 func (c ChassisCollector) Collect(result freeipmi.Result, ch chan<- prometheus.Metric, target ipmiTarget) (int, error) {
 	currentChassisPowerState, err := freeipmi.GetChassisPowerState(result)
+	currentChassisDriveFault, err := freeipmi.GetChassisDriveFault(result)
+	currentChassisCoolingFault, err := freeipmi.GetChassisCoolingFault(result)
 	if err != nil {
 		level.Error(logger).Log("msg", "Failed to collect chassis data", "target", targetName(target.host), "error", err)
 		return 0, err
@@ -58,5 +72,15 @@ func (c ChassisCollector) Collect(result freeipmi.Result, ch chan<- prometheus.M
 		prometheus.GaugeValue,
 		currentChassisPowerState,
 	)
+	ch <- prometheus.MustNewConstMetric(
+		chassisDriveFaultDesc,
+		prometheus.GaugeValue,
+		currentChassisDriveFault,
+	)
+	ch <- prometheus.MustNewConstMetric(
+		chassisCoolingFaultDesc,
+		prometheus.GaugeValue,
+		currentChassisCoolingFault,
+	)
 	return 1, nil
 }
diff --git a/freeipmi/freeipmi.go b/freeipmi/freeipmi.go
index 7210141..0231aee 100644
--- a/freeipmi/freeipmi.go
+++ b/freeipmi/freeipmi.go
@@ -35,6 +35,8 @@ import (
 var (
 	ipmiDCMICurrentPowerRegex         = regexp.MustCompile(`^Current Power\s*:\s*(?P<value>[0-9.]*)\s*Watts.*`)
 	ipmiChassisPowerRegex             = regexp.MustCompile(`^System Power\s*:\s(?P<value>.*)`)
+	ipmiChassisDriveFaultRegex        = regexp.MustCompile(`^Drive Fault\s*:\s(?P<value>.*)`)
+	ipmiChassisCoolingFaultRegex      = regexp.MustCompile(`^Cooling/fan fault\s*:\s(?P<value>.*)`)
 	ipmiSELEntriesRegex               = regexp.MustCompile(`^Number of log entries\s*:\s(?P<value>[0-9.]*)`)
 	ipmiSELFreeSpaceRegex             = regexp.MustCompile(`^Free space remaining\s*:\s(?P<value>[0-9.]*)\s*bytes.*`)
 	bmcInfoFirmwareRevisionRegex      = regexp.MustCompile(`^Firmware Revision\s*:\s*(?P<value>[0-9.]*).*`)
@@ -219,6 +221,34 @@ func GetChassisPowerState(ipmiOutput Result) (float64, error) {
 	return 0, err
 }
 
+func GetChassisDriveFault(ipmiOutput Result) (float64, error) {
+	if ipmiOutput.err != nil {
+		return -1, fmt.Errorf("%s: %s", ipmiOutput.err, ipmiOutput.output)
+	}
+	value, err := getValue(ipmiOutput.output, ipmiChassisDriveFaultRegex)
+	if err != nil {
+		return -1, err
+	}
+	if value == "false" {
+		return 1, err
+	}
+	return 0, err
+}
+
+func GetChassisCoolingFault(ipmiOutput Result) (float64, error) {
+	if ipmiOutput.err != nil {
+		return -1, fmt.Errorf("%s: %s", ipmiOutput.err, ipmiOutput.output)
+	}
+	value, err := getValue(ipmiOutput.output, ipmiChassisCoolingFaultRegex)
+	if err != nil {
+		return -1, err
+	}
+	if value == "false" {
+		return 1, err
+	}
+	return 0, err
+}
+
 func GetBMCInfoFirmwareRevision(ipmiOutput Result) (string, error) {
 	// Workaround for an issue described here: https://github.com/prometheus-community/ipmi_exporter/issues/57
 	// The command may fail, but produce usable output (minus the system firmware revision).