NODEDC_MISSION_CORE/deploy/telemetry-plane/telegraf/Update-NdcMissionCoreTelegraf.ps1

[CmdletBinding()]
param(
    [Parameter(Mandatory = $true)]
    [string]$ConfigurationTemplate,
    [string]$PipelineCollector = "$PSScriptRoot\Get-NdcMissionCorePipelineTelemetry.ps1",
    [string]$PipelineJournal = "D:\NDC_MISSIONCORE\runtime\derived\.perception-persistent-publish\pipeline-telemetry.jsonl"
)

$ErrorActionPreference = "Stop"
$serviceName = "telegraf"
$installRoot = "C:\Program Files\NDC\Mission Core\Telegraf"
$configurationRoot = "C:\ProgramData\NDC\MissionCore\telemetry-agent"
$configurationPath = Join-Path $configurationRoot "telegraf.conf"
$collectorPath = Join-Path $configurationRoot "Get-NdcMissionCorePipelineTelemetry.ps1"
$executable = Join-Path $installRoot "telegraf.exe"
$serviceRegistryPath = "HKLM:\SYSTEM\CurrentControlSet\Services\$serviceName"

$service = Get-Service -Name $serviceName -ErrorAction Stop
if (-not (Test-Path -LiteralPath $executable -PathType Leaf)) {
    throw "NDC Mission Core Telegraf executable is missing"
}
if (-not (Test-Path -LiteralPath $ConfigurationTemplate -PathType Leaf)) {
    throw "Configuration template not found: $ConfigurationTemplate"
}
if (-not (Test-Path -LiteralPath $PipelineCollector -PathType Leaf)) {
    throw "Pipeline collector not found: $PipelineCollector"
}
$journalParent = Get-Item -LiteralPath (
    Resolve-Path -LiteralPath (Split-Path $PipelineJournal -Parent)
).Path -Force
if (
    -not $journalParent.PSIsContainer -or
    ($journalParent.Attributes -band [IO.FileAttributes]::ReparsePoint) -or
    [IO.Path]::GetPathRoot($journalParent.FullName).TrimEnd("\") -ine "D:"
) {
    throw "Pipeline journal parent must be a real D: directory"
}
if (Test-Path -LiteralPath $PipelineJournal) {
    $journal = Get-Item -LiteralPath $PipelineJournal -Force
    if (
        $journal.PSIsContainer -or
        ($journal.Attributes -band [IO.FileAttributes]::ReparsePoint)
    ) {
        throw "Pipeline journal must be a regular file"
    }
}
else {
    New-Item -ItemType File -Path $PipelineJournal | Out-Null
}

$serviceEnvironmentBefore = @((Get-ItemProperty -Path $serviceRegistryPath).Environment)
foreach ($entry in $serviceEnvironmentBefore) {
    $name, $value = $entry -split "=", 2
    if ($name -and $value) {
        Set-Item -Path "Env:$name" -Value $value
    }
}
if (-not $env:MISSIONCORE_TELEMETRY_INTERVAL) {
    $env:MISSIONCORE_TELEMETRY_INTERVAL = "2s"
}
$serviceEnvironmentCandidate = @($serviceEnvironmentBefore)
if (-not ($serviceEnvironmentCandidate | Where-Object {
    $_ -like "MISSIONCORE_TELEMETRY_INTERVAL=*"
})) {
    $serviceEnvironmentCandidate += (
        "MISSIONCORE_TELEMETRY_INTERVAL=$($env:MISSIONCORE_TELEMETRY_INTERVAL)"
    )
}

$temporaryRoot = Join-Path $env:TEMP "ndc-mission-core-telegraf-update-$([Guid]::NewGuid().ToString('N'))"
$validationOutput = Join-Path $temporaryRoot "validation.out.log"
$validationError = Join-Path $temporaryRoot "validation.error.log"
$backupRoot = Join-Path $configurationRoot "backups"
$backupPath = Join-Path $backupRoot "telegraf-$([DateTime]::UtcNow.ToString('yyyyMMddTHHmmssZ')).conf"
$backupCollectorPath = Join-Path $backupRoot `
    "pipeline-collector-$([DateTime]::UtcNow.ToString('yyyyMMddTHHmmssZ')).ps1"
$hadCollector = Test-Path -LiteralPath $collectorPath -PathType Leaf
try {
    New-Item -ItemType Directory -Path $temporaryRoot, $backupRoot -Force | Out-Null
    $collectorProbe = & powershell.exe -NoLogo -NoProfile -NonInteractive `
        -ExecutionPolicy Bypass -File $PipelineCollector
    if ($LASTEXITCODE -ne 0) {
        throw "Pipeline collector validation failed"
    }
    $collectorDocument = $collectorProbe | ConvertFrom-Json
    if (@($collectorDocument).Count -ne 9) {
        throw "Pipeline collector must publish exactly nine stage samples"
    }
    if ($hadCollector) {
        Copy-Item -LiteralPath $collectorPath -Destination $backupCollectorPath
    }
    Copy-Item -LiteralPath $PipelineCollector -Destination $collectorPath
    $validation = Start-Process -FilePath $executable `
        -ArgumentList @("--config", $ConfigurationTemplate, "--test") `
        -NoNewWindow -Wait -PassThru `
        -RedirectStandardOutput $validationOutput `
        -RedirectStandardError $validationError
    if ($validation.ExitCode -ne 0) {
        $validationDetail = Get-Content -LiteralPath $validationError -Tail 8 |
            Out-String
        throw "Telegraf configuration validation failed: $validationDetail"
    }

    Copy-Item -LiteralPath $configurationPath -Destination $backupPath
    Set-ItemProperty -Path $serviceRegistryPath -Name Environment `
        -Type MultiString -Value $serviceEnvironmentCandidate
    Stop-Service -Name $serviceName
    $service = Get-Service -Name $serviceName
    $service.WaitForStatus(
        [ServiceProcess.ServiceControllerStatus]::Stopped,
        [TimeSpan]::FromSeconds(20)
    )
    try {
        Copy-Item -LiteralPath $ConfigurationTemplate -Destination $configurationPath
        Start-Service -Name $serviceName
        $service = Get-Service -Name $serviceName
        $service.WaitForStatus(
            [ServiceProcess.ServiceControllerStatus]::Running,
            [TimeSpan]::FromSeconds(20)
        )
        & sc.exe failure $serviceName reset= 86400 `
            actions= restart/5000/restart/30000/restart/60000 | Out-Null
        if ($LASTEXITCODE -ne 0) {
            throw "Failed to configure Telegraf service recovery actions"
        }
        & sc.exe failureflag $serviceName 1 | Out-Null
        if ($LASTEXITCODE -ne 0) {
            throw "Failed to enable Telegraf recovery for non-crash failures"
        }
    }
    catch {
        Stop-Service -Name $serviceName -Force -ErrorAction SilentlyContinue
        $service = Get-Service -Name $serviceName
        if ($service.Status -ne [ServiceProcess.ServiceControllerStatus]::Stopped) {
            $service.WaitForStatus(
                [ServiceProcess.ServiceControllerStatus]::Stopped,
                [TimeSpan]::FromSeconds(20)
            )
        }
        Set-ItemProperty -Path $serviceRegistryPath -Name Environment `
            -Type MultiString -Value $serviceEnvironmentBefore
        Copy-Item -LiteralPath $backupPath -Destination $configurationPath
        if ($hadCollector) {
            Copy-Item -LiteralPath $backupCollectorPath -Destination $collectorPath
        }
        else {
            Remove-Item -LiteralPath $collectorPath -Force -ErrorAction SilentlyContinue
        }
        Start-Service -Name $serviceName
        throw
    }

    [ordered]@{
        SchemaVersion = "missioncore.telemetry-agent-config-result/v1"
        ServiceName = $serviceName
        Status = (Get-Service -Name $serviceName).Status.ToString()
        RecoveryConfigured = $true
        Configuration = $configurationPath
        Backup = $backupPath
        PipelineCollector = $collectorPath
        PipelineCollectorBackup = if ($hadCollector) { $backupCollectorPath } else { $null }
        PipelineJournal = $PipelineJournal
    } | ConvertTo-Json -Compress
}
catch {
    Set-ItemProperty -Path $serviceRegistryPath -Name Environment `
        -Type MultiString -Value $serviceEnvironmentBefore
    if ($hadCollector -and (Test-Path -LiteralPath $backupCollectorPath -PathType Leaf)) {
        Copy-Item -LiteralPath $backupCollectorPath -Destination $collectorPath
    }
    elseif (-not $hadCollector) {
        Remove-Item -LiteralPath $collectorPath -Force -ErrorAction SilentlyContinue
    }
    throw
}
finally {
    Remove-Item -LiteralPath $temporaryRoot -Recurse -Force -ErrorAction SilentlyContinue
}