linux-scripts/windows-cluster-health-exporter.ps1

<#
.SYNOPSIS
    Windows Failover Cluster Health Prometheus Exporter
.DESCRIPTION
    Prometheus exporter for Windows Failover Clustering - cluster state, node
    status, resource group health, cluster shared volumes, network health,
    quorum status, and recent failover events via PowerShell.
.PARAMETER Mode
    Output mode: 'stdout' (default), 'textfile', or 'http'
.PARAMETER Port
    HTTP port for http mode (default: 9540)
.PARAMETER TextfileDir
    Directory for textfile collector output (default: C:\ProgramData\node_exporter)
.PARAMETER InstallScheduledTask
    Switch to create a scheduled task for automatic execution
.PARAMETER TaskIntervalMinutes
    Interval in minutes for the scheduled task (default: 2)
.NOTES
    Author:  Phil Connor
    Contact: contact@mylinux.work
    Website: https://mylinux.work
    License: MIT
    Version: 1.0

    Metrics Exported:
      Cluster:
        - windows_cluster_up
        - windows_cluster_exporter_info{version}
        - windows_cluster_node_count
        - windows_cluster_node_up{node,state}
        - windows_cluster_quorum_state{type}

      Resource Groups:
        - windows_cluster_group_state{group,node,state}
        - windows_cluster_group_count
        - windows_cluster_group_online_count

      Resources:
        - windows_cluster_resource_state{resource,group,type,node,state}

      CSV (Cluster Shared Volumes):
        - windows_cluster_csv_total_bytes{name,path}
        - windows_cluster_csv_used_bytes{name,path}
        - windows_cluster_csv_free_bytes{name,path}
        - windows_cluster_csv_utilization{name,path}

      Networks:
        - windows_cluster_network_state{network,state,role}

      Events:
        - windows_cluster_failover_events_total
        - windows_cluster_last_failover_timestamp

      Exporter:
        - windows_cluster_exporter_duration_seconds
        - windows_cluster_exporter_last_run_timestamp
#>

param(
    [ValidateSet('stdout','textfile','http')]
    [string]$Mode = 'stdout',

    [int]$Port = 9540,

    [string]$TextfileDir = 'C:\ProgramData\node_exporter',

    [switch]$InstallScheduledTask,

    [int]$TaskIntervalMinutes = 2
)

$ErrorActionPreference = 'SilentlyContinue'
$Version = '1.0'

# ============================================================================
# SCHEDULED TASK INSTALLER
# ============================================================================

if ($InstallScheduledTask) {
    $scriptPath = $MyInvocation.MyCommand.Path
    $action = New-ScheduledTaskAction -Execute 'powershell.exe' `
        -Argument "-NoProfile -ExecutionPolicy Bypass -File `"$scriptPath`" -Mode textfile"
    $trigger = New-ScheduledTaskTrigger -RepetitionInterval (New-TimeSpan -Minutes $TaskIntervalMinutes) `
        -Once -At (Get-Date)
    $settings = New-ScheduledTaskSettingsSet -AllowStartIfOnBatteries -DontStopIfGoingOnBatteries `
        -StartWhenAvailable -ExecutionTimeLimit (New-TimeSpan -Minutes 5)
    Register-ScheduledTask -TaskName 'WindowsClusterHealthExporter' -Action $action -Trigger $trigger `
        -Settings $settings -RunLevel Highest -User 'SYSTEM' -Force
    Write-Host 'Scheduled task "WindowsClusterHealthExporter" installed successfully.'
    exit 0
}

# ============================================================================
# HELPER FUNCTIONS
# ============================================================================

function Get-PrometheusEscape {
    param([string]$Value)
    $Value -replace '\\', '\\\\' -replace '"', '\"' -replace "`n", '\n'
}

function Write-MetricHeader {
    param([string]$Name, [string]$Type, [string]$Help)
    "# HELP $Name $Help"
    "# TYPE $Name $Type"
}

# ============================================================================
# METRIC COLLECTION
# ============================================================================

function Get-ClusterHealthMetrics {
    $startTime = Get-Date
    $metrics = [System.Collections.Generic.List[string]]::new()

    # --- Exporter status ---
    $metrics.AddRange([string[]](Write-MetricHeader 'windows_cluster_up' 'gauge' 'Cluster exporter status (1=up, 0=down)'))

    try {
        Import-Module FailoverClusters -ErrorAction Stop
        $cluster = Get-Cluster -ErrorAction Stop
        $metrics.Add('windows_cluster_up 1')
    } catch {
        $metrics.Add('windows_cluster_up 0')
        return ($metrics -join "`n")
    }

    $metrics.AddRange([string[]](Write-MetricHeader 'windows_cluster_exporter_info' 'gauge' 'Exporter version information'))
    $metrics.Add("windows_cluster_exporter_info{version=`"$Version`"} 1")

    # --- Node status ---
    try {
        $nodes = Get-ClusterNode -ErrorAction Stop
        $nodeCount = ($nodes | Measure-Object).Count

        $metrics.AddRange([string[]](Write-MetricHeader 'windows_cluster_node_count' 'gauge' 'Total number of cluster nodes'))
        $metrics.Add("windows_cluster_node_count $nodeCount")

        $metrics.AddRange([string[]](Write-MetricHeader 'windows_cluster_node_up' 'gauge' 'Node state (1=Up, 0=other)'))

        foreach ($node in $nodes) {
            $nodeName = Get-PrometheusEscape $node.Name
            $stateStr = $node.State.ToString()
            $stateVal = if ($stateStr -eq 'Up') { 1 } else { 0 }
            $metrics.Add("windows_cluster_node_up{node=`"$nodeName`",state=`"$stateStr`"} $stateVal")
        }
    } catch {
        # Node enumeration failed
    }

    # --- Quorum ---
    try {
        $quorum = Get-ClusterQuorum -ErrorAction Stop
        $qType = Get-PrometheusEscape $quorum.QuorumType.ToString()
        $metrics.AddRange([string[]](Write-MetricHeader 'windows_cluster_quorum_state' 'gauge' 'Quorum status (1=healthy)'))
        $metrics.Add("windows_cluster_quorum_state{type=`"$qType`"} 1")
    } catch {
        # Quorum info unavailable
    }

    # --- Resource groups ---
    try {
        $groups = Get-ClusterGroup -ErrorAction Stop
        $groupCount = ($groups | Measure-Object).Count
        $onlineCount = ($groups | Where-Object { $_.State -eq 'Online' } | Measure-Object).Count

        $metrics.AddRange([string[]](Write-MetricHeader 'windows_cluster_group_count' 'gauge' 'Total resource groups'))
        $metrics.Add("windows_cluster_group_count $groupCount")

        $metrics.AddRange([string[]](Write-MetricHeader 'windows_cluster_group_online_count' 'gauge' 'Online resource groups'))
        $metrics.Add("windows_cluster_group_online_count $onlineCount")

        $metrics.AddRange([string[]](Write-MetricHeader 'windows_cluster_group_state' 'gauge' 'Resource group state (1=Online, 0=other)'))

        foreach ($group in $groups) {
            $gName = Get-PrometheusEscape $group.Name
            $gNode = Get-PrometheusEscape $group.OwnerNode.ToString()
            $gState = $group.State.ToString()
            $gVal = if ($gState -eq 'Online') { 1 } else { 0 }
            $metrics.Add("windows_cluster_group_state{group=`"$gName`",node=`"$gNode`",state=`"$gState`"} $gVal")
        }
    } catch {
        # Group enumeration failed
    }

    # --- Resources ---
    try {
        $resources = Get-ClusterResource -ErrorAction Stop
        $metrics.AddRange([string[]](Write-MetricHeader 'windows_cluster_resource_state' 'gauge' 'Resource state (1=Online, 0=other)'))

        foreach ($res in $resources) {
            $rName = Get-PrometheusEscape $res.Name
            $rGroup = Get-PrometheusEscape $res.OwnerGroup.ToString()
            $rType = Get-PrometheusEscape $res.ResourceType.ToString()
            $rNode = Get-PrometheusEscape $res.OwnerNode.ToString()
            $rState = $res.State.ToString()
            $rVal = if ($rState -eq 'Online') { 1 } else { 0 }
            $metrics.Add("windows_cluster_resource_state{resource=`"$rName`",group=`"$rGroup`",type=`"$rType`",node=`"$rNode`",state=`"$rState`"} $rVal")
        }
    } catch {
        # Resource enumeration failed
    }

    # --- Cluster Shared Volumes ---
    try {
        $csvs = Get-ClusterSharedVolume -ErrorAction Stop
        $metrics.AddRange([string[]](Write-MetricHeader 'windows_cluster_csv_total_bytes' 'gauge' 'CSV total size in bytes'))
        $metrics.AddRange([string[]](Write-MetricHeader 'windows_cluster_csv_used_bytes' 'gauge' 'CSV used space in bytes'))
        $metrics.AddRange([string[]](Write-MetricHeader 'windows_cluster_csv_free_bytes' 'gauge' 'CSV free space in bytes'))
        $metrics.AddRange([string[]](Write-MetricHeader 'windows_cluster_csv_utilization' 'gauge' 'CSV utilization percentage'))

        foreach ($csv in $csvs) {
            $csvName = Get-PrometheusEscape $csv.Name
            foreach ($vol in $csv.SharedVolumeInfo) {
                $csvPath = Get-PrometheusEscape $vol.FriendlyVolumeName
                $partition = $vol.Partition
                $total = $partition.Size
                $free = $partition.FreeSpace
                $used = $total - $free
                $util = if ($total -gt 0) { [math]::Round(($used / $total) * 100, 2) } else { 0 }

                $metrics.Add("windows_cluster_csv_total_bytes{name=`"$csvName`",path=`"$csvPath`"} $total")
                $metrics.Add("windows_cluster_csv_used_bytes{name=`"$csvName`",path=`"$csvPath`"} $used")
                $metrics.Add("windows_cluster_csv_free_bytes{name=`"$csvName`",path=`"$csvPath`"} $free")
                $metrics.Add("windows_cluster_csv_utilization{name=`"$csvName`",path=`"$csvPath`"} $util")
            }
        }
    } catch {
        # CSV info unavailable (not all clusters use CSV)
    }

    # --- Networks ---
    try {
        $networks = Get-ClusterNetwork -ErrorAction Stop
        $metrics.AddRange([string[]](Write-MetricHeader 'windows_cluster_network_state' 'gauge' 'Network state (1=Up, 0=other)'))

        foreach ($net in $networks) {
            $netName = Get-PrometheusEscape $net.Name
            $netState = $net.State.ToString()
            $netRole = $net.Role.ToString()
            $netVal = if ($netState -eq 'Up') { 1 } else { 0 }
            $metrics.Add("windows_cluster_network_state{network=`"$netName`",state=`"$netState`",role=`"$netRole`"} $netVal")
        }
    } catch {
        # Network info unavailable
    }

    # --- Failover events ---
    try {
        $events = Get-WinEvent -FilterHashtable @{
            LogName = 'Microsoft-Windows-FailoverClustering/Operational'
            ID = 1069, 1070, 1641
        } -MaxEvents 100 -ErrorAction Stop

        $last24h = ($events | Where-Object { $_.TimeCreated -gt (Get-Date).AddHours(-24) } | Measure-Object).Count

        $metrics.AddRange([string[]](Write-MetricHeader 'windows_cluster_failover_events_total' 'gauge' 'Failover events in last 24h'))
        $metrics.Add("windows_cluster_failover_events_total $last24h")

        if ($events.Count -gt 0) {
            $lastEvent = [math]::Round(($events[0].TimeCreated - [datetime]'1970-01-01').TotalSeconds)
            $metrics.AddRange([string[]](Write-MetricHeader 'windows_cluster_last_failover_timestamp' 'gauge' 'Unix timestamp of last failover event'))
            $metrics.Add("windows_cluster_last_failover_timestamp $lastEvent")
        }
    } catch {
        # Event log unavailable
    }

    # --- Duration and timestamp ---
    $duration = [math]::Round(((Get-Date) - $startTime).TotalSeconds, 2)
    $timestamp = [math]::Round((Get-Date -UFormat %s), 0)

    $metrics.AddRange([string[]](Write-MetricHeader 'windows_cluster_exporter_duration_seconds' 'gauge' 'Script execution time'))
    $metrics.Add("windows_cluster_exporter_duration_seconds $duration")
    $metrics.AddRange([string[]](Write-MetricHeader 'windows_cluster_exporter_last_run_timestamp' 'gauge' 'Unix timestamp of last successful run'))
    $metrics.Add("windows_cluster_exporter_last_run_timestamp $timestamp")

    return ($metrics -join "`n")
}

# ============================================================================
# OUTPUT
# ============================================================================

switch ($Mode) {
    'stdout' {
        Get-ClusterHealthMetrics
    }
    'textfile' {
        if (-not (Test-Path $TextfileDir)) { New-Item -ItemType Directory -Path $TextfileDir -Force | Out-Null }
        $tempFile = Join-Path $TextfileDir "windows-cluster-health-metrics.tmp"
        $finalFile = Join-Path $TextfileDir "windows-cluster-health-metrics.prom"
        Get-ClusterHealthMetrics | Out-File -FilePath $tempFile -Encoding utf8 -NoNewline
        Move-Item -Path $tempFile -Destination $finalFile -Force
        Write-Host "Wrote metrics to $finalFile"
    }
    'http' {
        $prefix = "http://+:$Port/metrics/"
        $listener = [System.Net.HttpListener]::new()
        $listener.Prefixes.Add($prefix)
        $listener.Start()
        Write-Host "Listening on port $Port..."
        try {
            while ($listener.IsListening) {
                $context = $listener.GetContext()
                $response = $context.Response
                $metricsOutput = Get-ClusterHealthMetrics
                $buffer = [System.Text.Encoding]::UTF8.GetBytes($metricsOutput)
                $response.ContentType = 'text/plain; version=0.0.4; charset=utf-8'
                $response.ContentLength64 = $buffer.Length
                $response.OutputStream.Write($buffer, 0, $buffer.Length)
                $response.OutputStream.Close()
            }
        } finally {
            $listener.Stop()
        }
    }
}