Mirrors/ProxMenux
1
0
Fork 0
mirror of https://github.com/MacRimi/ProxMenux.git synced 2025-11-18 19:46:18 +00:00
Code Issues Packages Projects Releases Wiki Activity

Update AppImage

Browse Source
This commit is contained in:
MacRimi
2025-10-13 23:50:31 +02:00
parent 75a458f2be
commit c1614e8241
2 changed files with 447 additions and 86 deletions
Download Patch File Download Diff File Expand all files Collapse all files

276
AppImage/components/storage-metrics.tsx
View File

@@ -4,7 +4,19 @@ import { useState, useEffect } from "react"
import { Card, CardContent, CardHeader, CardTitle } from "./ui/card" import { Card, CardContent, CardHeader, CardTitle } from "./ui/card"
import { Progress } from "./ui/progress" import { Progress } from "./ui/progress"
import { Badge } from "./ui/badge" import { Badge } from "./ui/badge"
import { HardDrive, Database, Archive, AlertTriangle, CheckCircle, Activity, AlertCircle } from "lucide-react" import { Button } from "./ui/button"
import { Dialog, DialogContent, DialogDescription, DialogHeader, DialogTitle, DialogTrigger } from "./ui/dialog"
import {
HardDrive,
Database,
Archive,
AlertTriangle,
CheckCircle,
Activity,
AlertCircle,
Info,
Thermometer,
} from "lucide-react"
interface StorageData { interface StorageData {
total: number total: number
@@ -23,11 +35,22 @@ interface DiskInfo {
usage_percent: number usage_percent: number
health: string health: string
temperature: number temperature: number
disk_type?: string
percentage_used?: number
ssd_life_left?: number
wear_leveling_count?: number
media_wearout_indicator?: number
}
interface DiskGroup {
type: string
disks: DiskInfo[]
avgTemp: number
status: "safe" | "warning" | "critical"
} }
const fetchStorageData = async (): Promise<StorageData | null> => { const fetchStorageData = async (): Promise<StorageData | null> => {
try { try {
console.log("[v0] Fetching storage data from Flask server...")
const response = await fetch("/api/storage", { const response = await fetch("/api/storage", {
method: "GET", method: "GET",
headers: { headers: {
@@ -41,7 +64,6 @@ const fetchStorageData = async (): Promise<StorageData | null> => {
} }
const data = await response.json() const data = await response.json()
console.log("[v0] Successfully fetched storage data from Flask:", data)
return data return data
} catch (error) { } catch (error) {
console.error("[v0] Failed to fetch storage data from Flask server:", error) console.error("[v0] Failed to fetch storage data from Flask server:", error)
@@ -49,6 +71,146 @@ const fetchStorageData = async (): Promise<StorageData | null> => {
} }
} }
const getTempStatus = (temp: number, diskType: string): "safe" | "warning" | "critical" => {
if (diskType === "HDD") {
if (temp > 55) return "critical"
if (temp > 45) return "warning"
return "safe"
} else if (diskType === "SSD") {
if (temp > 65) return "critical"
if (temp > 55) return "warning"
return "safe"
} else if (diskType === "NVMe") {
if (temp > 70) return "critical"
if (temp > 60) return "warning"
return "safe"
}
// Umbral genérico
if (temp > 70) return "critical"
if (temp > 60) return "warning"
return "safe"
}
const groupDisksByType = (disks: DiskInfo[]): DiskGroup[] => {
const groups: { [key: string]: DiskInfo[] } = {}
disks.forEach((disk) => {
const type = disk.disk_type || "Unknown"
if (!groups[type]) {
groups[type] = []
}
groups[type].push(disk)
})
return Object.entries(groups).map(([type, disks]) => {
const temps = disks.map((d) => d.temperature).filter((t) => t > 0)
const avgTemp = temps.length > 0 ? Math.round(temps.reduce((a, b) => a + b, 0) / temps.length) : 0
// Determinar el estado más crítico del grupo
let status: "safe" | "warning" | "critical" = "safe"
disks.forEach((disk) => {
const diskStatus = getTempStatus(disk.temperature, type)
if (diskStatus === "critical") status = "critical"
else if (diskStatus === "warning" && status !== "critical") status = "warning"
})
return { type, disks, avgTemp, status }
})
}
function TemperatureThresholdsModal() {
return (
<Dialog>
<DialogTrigger asChild>
<Button variant="outline" size="sm" className="gap-2 bg-transparent">
<Info className="h-4 w-4" />
Umbrales de temperatura
</Button>
</DialogTrigger>
<DialogContent className="max-w-3xl">
<DialogHeader>
<DialogTitle>Umbrales de temperatura por tipo de disco</DialogTitle>
<DialogDescription>
Rangos de temperatura recomendados para cada tipo de dispositivo de almacenamiento
</DialogDescription>
</DialogHeader>
<div className="overflow-x-auto">
<table className="w-full border-collapse">
<thead>
<tr className="border-b border-border">
<th className="text-left p-3 font-semibold">Tipo de disco</th>
<th className="text-left p-3 font-semibold">Temperatura de operación</th>
<th className="text-left p-3 font-semibold">Zona segura</th>
<th className="text-left p-3 font-semibold">Zona de advertencia</th>
<th className="text-left p-3 font-semibold">Zona crítica</th>
</tr>
</thead>
<tbody>
<tr className="border-b border-border">
<td className="p-3 font-medium">HDD</td>
<td className="p-3">0°C – 60°C (común: 5–55°C)</td>
<td className="p-3">
<Badge variant="outline" className="bg-green-500/10 text-green-500 border-green-500/20">
≤ 45°C
</Badge>
</td>
<td className="p-3">
<Badge variant="outline" className="bg-yellow-500/10 text-yellow-500 border-yellow-500/20">
46 – 55°C
</Badge>
</td>
<td className="p-3">
<Badge variant="outline" className="bg-red-500/10 text-red-500 border-red-500/20">
&gt; 55°C
</Badge>
</td>
</tr>
<tr className="border-b border-border">
<td className="p-3 font-medium">SSD</td>
<td className="p-3">0°C – 70°C</td>
<td className="p-3">
<Badge variant="outline" className="bg-green-500/10 text-green-500 border-green-500/20">
≤ 55°C
</Badge>
</td>
<td className="p-3">
<Badge variant="outline" className="bg-yellow-500/10 text-yellow-500 border-yellow-500/20">
56 – 65°C
</Badge>
</td>
<td className="p-3">
<Badge variant="outline" className="bg-red-500/10 text-red-500 border-red-500/20">
&gt; 65°C
</Badge>
</td>
</tr>
<tr>
<td className="p-3 font-medium">NVMe</td>
<td className="p-3">0°C – 70°C</td>
<td className="p-3">
<Badge variant="outline" className="bg-green-500/10 text-green-500 border-green-500/20">
≤ 60°C
</Badge>
</td>
<td className="p-3">
<Badge variant="outline" className="bg-yellow-500/10 text-yellow-500 border-yellow-500/20">
61 – 70°C
</Badge>
</td>
<td className="p-3">
<Badge variant="outline" className="bg-red-500/10 text-red-500 border-red-500/20">
&gt; 70°C
</Badge>
</td>
</tr>
</tbody>
</table>
</div>
</DialogContent>
</Dialog>
)
}
export function StorageMetrics() { export function StorageMetrics() {
const [storageData, setStorageData] = useState<StorageData | null>(null) const [storageData, setStorageData] = useState<StorageData | null>(null)
const [loading, setLoading] = useState(true) const [loading, setLoading] = useState(true)
@@ -105,6 +267,7 @@ export function StorageMetrics() {
} }
const usagePercent = storageData.total > 0 ? (storageData.used / storageData.total) * 100 : 0 const usagePercent = storageData.total > 0 ? (storageData.used / storageData.total) * 100 : 0
const diskGroups = groupDisksByType(storageData.disks)
return ( return (
<div className="space-y-6"> <div className="space-y-6">
@@ -173,6 +336,75 @@ export function StorageMetrics() {
</Card> </Card>
</div> </div>
{diskGroups.length > 0 && (
<div>
<div className="flex items-center justify-between mb-4">
<h3 className="text-lg font-semibold text-foreground">Temperatura por tipo de disco</h3>
<TemperatureThresholdsModal />
</div>
<div className="grid grid-cols-1 md:grid-cols-2 lg:grid-cols-3 gap-6">
{diskGroups.map((group) => (
<Card key={group.type} className="bg-card border-border">
<CardHeader>
<CardTitle className="text-foreground flex items-center justify-between">
<div className="flex items-center">
<Thermometer className="h-5 w-5 mr-2" />
{group.type} Temperature
</div>
<Badge
variant="outline"
className={
group.status === "safe"
? "bg-green-500/10 text-green-500 border-green-500/20"
: group.status === "warning"
? "bg-yellow-500/10 text-yellow-500 border-yellow-500/20"
: "bg-red-500/10 text-red-500 border-red-500/20"
}
>
{group.status === "safe" ? "Seguro" : group.status === "warning" ? "Advertencia" : "Crítico"}
</Badge>
</CardTitle>
</CardHeader>
<CardContent>
<div
className={`text-3xl font-bold ${
group.status === "safe"
? "text-green-500"
: group.status === "warning"
? "text-yellow-500"
: "text-red-500"
}`}
>
{group.avgTemp}°C
</div>
<p className="text-sm text-muted-foreground mt-2">
Promedio de {group.disks.length} disco{group.disks.length > 1 ? "s" : ""}
</p>
<div className="mt-3 space-y-1">
{group.disks.map((disk, idx) => (
<div key={idx} className="flex items-center justify-between text-xs">
<span className="text-muted-foreground">{disk.name}</span>
<span
className={`font-medium ${
getTempStatus(disk.temperature, group.type) === "safe"
? "text-green-500"
: getTempStatus(disk.temperature, group.type) === "warning"
? "text-yellow-500"
: "text-red-500"
}`}
>
{disk.temperature}°C
</span>
</div>
))}
</div>
</CardContent>
</Card>
))}
</div>
</div>
)}
{/* Disk Details */} {/* Disk Details */}
<Card className="bg-card border-border"> <Card className="bg-card border-border">
<CardHeader> <CardHeader>
@@ -191,7 +423,14 @@ export function StorageMetrics() {
<div className="flex items-center space-x-4"> <div className="flex items-center space-x-4">
<HardDrive className="h-5 w-5 text-muted-foreground" /> <HardDrive className="h-5 w-5 text-muted-foreground" />
<div> <div>
<div className="font-medium text-foreground">{disk.name}</div> <div className="font-medium text-foreground flex items-center gap-2">
{disk.name}
{disk.disk_type && (
<Badge variant="outline" className="text-xs">
{disk.disk_type}
</Badge>
)}
</div>
<div className="text-sm text-muted-foreground"> <div className="text-sm text-muted-foreground">
{disk.fstype} • {disk.mountpoint} {disk.fstype} • {disk.mountpoint}
</div> </div>
@@ -208,8 +447,35 @@ export function StorageMetrics() {
<div className="text-center"> <div className="text-center">
<div className="text-sm text-muted-foreground">Temp</div> <div className="text-sm text-muted-foreground">Temp</div>
<div className="text-sm font-medium text-foreground">{disk.temperature}°C</div> <div
className={`text-sm font-medium ${
getTempStatus(disk.temperature, disk.disk_type || "Unknown") === "safe"
? "text-green-500"
: getTempStatus(disk.temperature, disk.disk_type || "Unknown") === "warning"
? "text-yellow-500"
: "text-red-500"
}`}
>
{disk.temperature}°C
</div> </div>
</div>
{(disk.disk_type === "SSD" || disk.disk_type === "NVMe") && disk.ssd_life_left !== undefined && (
<div className="text-center">
<div className="text-sm text-muted-foreground">Vida útil</div>
<div
className={`text-sm font-medium ${
disk.ssd_life_left >= 80
? "text-green-500"
: disk.ssd_life_left >= 50
? "text-yellow-500"
: "text-red-500"
}`}
>
{disk.ssd_life_left}%
</div>
</div>
)}
<Badge <Badge
variant="outline" variant="outline"

253
AppImage/scripts/flask_server.py
View File

@@ -536,7 +536,13 @@ def get_storage_info():
'pending_sectors': smart_data.get('pending_sectors', 0), 'pending_sectors': smart_data.get('pending_sectors', 0),
'crc_errors': smart_data.get('crc_errors', 0), 'crc_errors': smart_data.get('crc_errors', 0),
'rotation_rate': smart_data.get('rotation_rate', 0), # Added 'rotation_rate': smart_data.get('rotation_rate', 0), # Added
'power_cycles': smart_data.get('power_cycles', 0) # Added 'power_cycles': smart_data.get('power_cycles', 0), # Added
# Campos adicionales para SMART
'disk_type': smart_data.get('disk_type', 'unknown'),
'percentage_used': smart_data.get('percentage_used', 0),
'wear_leveling_count': smart_data.get('wear_leveling_count', 0),
'media_wearout_indicator': smart_data.get('media_wearout_indicator', 0),
'ssd_life_left': smart_data.get('ssd_life_left', 100)
} }
storage_data['disk_count'] += 1 storage_data['disk_count'] += 1
@@ -653,36 +659,44 @@ def get_smart_data(disk_name):
'reallocated_sectors': 0, 'reallocated_sectors': 0,
'pending_sectors': 0, 'pending_sectors': 0,
'crc_errors': 0, 'crc_errors': 0,
'rotation_rate': 0, # Added rotation rate (RPM) 'rotation_rate': 0,
'power_cycles': 0, # Added power cycle count 'power_cycles': 0,
'disk_type': 'unknown', # HDD, SSD, or NVMe
'percentage_used': 0, # For NVMe
'wear_leveling_count': 0, # For SSD
'media_wearout_indicator': 0, # For SSD (Intel)
'ssd_life_left': 100, # For SSD (percentage)
} }
print(f"[v0] ===== Starting SMART data collection for /dev/{disk_name} =====") print(f"[v0] ===== Starting SMART data collection for /dev/{disk_name} =====")
if 'nvme' in disk_name:
smart_data['disk_type'] = 'NVMe'
print(f"[v0] Detected NVMe disk")
try: try:
commands_to_try = [ commands_to_try = [
['smartctl', '-a', '-j', f'/dev/{disk_name}'], # JSON output (preferred) ['smartctl', '-a', '-j', f'/dev/{disk_name}'],
['smartctl', '-a', '-j', '-d', 'ata', f'/dev/{disk_name}'], # JSON with ATA device type ['smartctl', '-a', '-j', '-d', 'ata', f'/dev/{disk_name}'],
['smartctl', '-a', '-j', '-d', 'sat', f'/dev/{disk_name}'], # JSON with SAT device type ['smartctl', '-a', '-j', '-d', 'sat', f'/dev/{disk_name}'],
['smartctl', '-a', f'/dev/{disk_name}'], # Text output (fallback) ['smartctl', '-a', f'/dev/{disk_name}'],
['smartctl', '-a', '-d', 'ata', f'/dev/{disk_name}'], # Text with ATA device type ['smartctl', '-a', '-d', 'ata', f'/dev/{disk_name}'],
['smartctl', '-a', '-d', 'sat', f'/dev/{disk_name}'], # Text with SAT device type ['smartctl', '-a', '-d', 'sat', f'/dev/{disk_name}'],
['smartctl', '-i', '-H', '-A', f'/dev/{disk_name}'], # Info + Health + Attributes ['smartctl', '-i', '-H', '-A', f'/dev/{disk_name}'],
['smartctl', '-i', '-H', '-A', '-d', 'ata', f'/dev/{disk_name}'], # With ATA ['smartctl', '-i', '-H', '-A', '-d', 'ata', f'/dev/{disk_name}'],
['smartctl', '-i', '-H', '-A', '-d', 'sat', f'/dev/{disk_name}'], # With SAT ['smartctl', '-i', '-H', '-A', '-d', 'sat', f'/dev/{disk_name}'],
['smartctl', '-a', '-j', '-d', 'scsi', f'/dev/{disk_name}'], # JSON with SCSI device type ['smartctl', '-a', '-j', '-d', 'scsi', f'/dev/{disk_name}'],
['smartctl', '-a', '-j', '-d', 'sat,12', f'/dev/{disk_name}'], # SAT with 12-byte commands ['smartctl', '-a', '-j', '-d', 'sat,12', f'/dev/{disk_name}'],
['smartctl', '-a', '-j', '-d', 'sat,16', f'/dev/{disk_name}'], # SAT with 16-byte commands ['smartctl', '-a', '-j', '-d', 'sat,16', f'/dev/{disk_name}'],
['smartctl', '-a', '-d', 'sat,12', f'/dev/{disk_name}'], # Text SAT with 12-byte commands ['smartctl', '-a', '-d', 'sat,12', f'/dev/{disk_name}'],
['smartctl', '-a', '-d', 'sat,16', f'/dev/{disk_name}'], # Text SAT with 16-byte commands ['smartctl', '-a', '-d', 'sat,16', f'/dev/{disk_name}'],
] ]
process = None # Initialize process to None process = None
for cmd_index, cmd in enumerate(commands_to_try): for cmd_index, cmd in enumerate(commands_to_try):
print(f"[v0] Attempt {cmd_index + 1}/{len(commands_to_try)}: Running command: {' '.join(cmd)}") print(f"[v0] Attempt {cmd_index + 1}/{len(commands_to_try)}: Running command: {' '.join(cmd)}")
try: try:
process = subprocess.Popen(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True) process = subprocess.Popen(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True)
# Use communicate with a timeout to avoid hanging if the process doesn't exit
stdout, stderr = process.communicate(timeout=15) stdout, stderr = process.communicate(timeout=15)
result_code = process.returncode result_code = process.returncode
@@ -697,14 +711,12 @@ def get_smart_data(disk_name):
if has_output: if has_output:
print(f"[v0] Got output ({len(stdout)} bytes), attempting to parse...") print(f"[v0] Got output ({len(stdout)} bytes), attempting to parse...")
# Try JSON parsing first (if -j flag was used)
if '-j' in cmd: if '-j' in cmd:
try: try:
print(f"[v0] Attempting JSON parse...") print(f"[v0] Attempting JSON parse...")
data = json.loads(stdout) data = json.loads(stdout)
print(f"[v0] JSON parse successful!") print(f"[v0] JSON parse successful!")
# Extract model
if 'model_name' in data: if 'model_name' in data:
smart_data['model'] = data['model_name'] smart_data['model'] = data['model_name']
print(f"[v0] Model: {smart_data['model']}") print(f"[v0] Model: {smart_data['model']}")
@@ -712,61 +724,81 @@ def get_smart_data(disk_name):
smart_data['model'] = data['model_family'] smart_data['model'] = data['model_family']
print(f"[v0] Model family: {smart_data['model']}") print(f"[v0] Model family: {smart_data['model']}")
# Extract serial
if 'serial_number' in data: if 'serial_number' in data:
smart_data['serial'] = data['serial_number'] smart_data['serial'] = data['serial_number']
print(f"[v0] Serial: {smart_data['serial']}") print(f"[v0] Serial: {smart_data['serial']}")
if 'rotation_rate' in data: if 'rotation_rate' in data:
smart_data['rotation_rate'] = data['rotation_rate'] rotation = data['rotation_rate']
print(f"[v0] Rotation Rate: {smart_data['rotation_rate']} RPM") smart_data['rotation_rate'] = rotation
if rotation == 0 or 'Solid State Device' in str(rotation):
smart_data['disk_type'] = 'SSD'
print(f"[v0] Detected SSD (rotation_rate: {rotation})")
elif isinstance(rotation, int) and rotation > 0:
smart_data['disk_type'] = 'HDD'
print(f"[v0] Detected HDD (rotation_rate: {rotation} RPM)")
# Extract SMART status
if 'smart_status' in data and 'passed' in data['smart_status']: if 'smart_status' in data and 'passed' in data['smart_status']:
smart_data['smart_status'] = 'passed' if data['smart_status']['passed'] else 'failed' smart_data['smart_status'] = 'passed' if data['smart_status']['passed'] else 'failed'
smart_data['health'] = 'healthy' if data['smart_status']['passed'] else 'critical' smart_data['health'] = 'healthy' if data['smart_status']['passed'] else 'critical'
print(f"[v0] SMART status: {smart_data['smart_status']}, health: {smart_data['health']}") print(f"[v0] SMART status: {smart_data['smart_status']}, health: {smart_data['health']}")
# Extract temperature
if 'temperature' in data and 'current' in data['temperature']: if 'temperature' in data and 'current' in data['temperature']:
smart_data['temperature'] = data['temperature']['current'] smart_data['temperature'] = data['temperature']['current']
print(f"[v0] Temperature: {smart_data['temperature']}°C") print(f"[v0] Temperature: {smart_data['temperature']}°C")
# Parse ATA SMART attributes
if 'ata_smart_attributes' in data and 'table' in data['ata_smart_attributes']: if 'ata_smart_attributes' in data and 'table' in data['ata_smart_attributes']:
print(f"[v0] Parsing ATA SMART attributes...") print(f"[v0] Parsing ATA SMART attributes...")
for attr in data['ata_smart_attributes']['table']: for attr in data['ata_smart_attributes']['table']:
attr_id = attr.get('id') attr_id = attr.get('id')
attr_name = attr.get('name', '')
raw_value = attr.get('raw', {}).get('value', 0) raw_value = attr.get('raw', {}).get('value', 0)
normalized_value = attr.get('value', 0)
if attr_id == 9: # Power_On_Hours if attr_id == 9:
smart_data['power_on_hours'] = raw_value smart_data['power_on_hours'] = raw_value
print(f"[v0] Power On Hours (ID 9): {raw_value}") print(f"[v0] Power On Hours (ID 9): {raw_value}")
elif attr_id == 12: # Power_Cycle_Count elif attr_id == 12:
smart_data['power_cycles'] = raw_value smart_data['power_cycles'] = raw_value
print(f"[v0] Power Cycles (ID 12): {raw_value}") print(f"[v0] Power Cycles (ID 12): {raw_value}")
elif attr_id == 194: # Temperature_Celsius elif attr_id == 194:
if smart_data['temperature'] == 0: if smart_data['temperature'] == 0:
smart_data['temperature'] = raw_value smart_data['temperature'] = raw_value
print(f"[v0] Temperature (ID 194): {raw_value}°C") print(f"[v0] Temperature (ID 194): {raw_value}°C")
elif attr_id == 190: # Airflow_Temperature_Cel elif attr_id == 190:
if smart_data['temperature'] == 0: if smart_data['temperature'] == 0:
smart_data['temperature'] = raw_value smart_data['temperature'] = raw_value
print(f"[v0] Airflow Temperature (ID 190): {raw_value}°C") print(f"[v0] Airflow Temperature (ID 190): {raw_value}°C")
elif attr_id == 5: # Reallocated_Sector_Ct elif attr_id == 5:
smart_data['reallocated_sectors'] = raw_value smart_data['reallocated_sectors'] = raw_value
print(f"[v0] Reallocated Sectors (ID 5): {raw_value}") print(f"[v0] Reallocated Sectors (ID 5): {raw_value}")
elif attr_id == 197: # Current_Pending_Sector elif attr_id == 197:
smart_data['pending_sectors'] = raw_value smart_data['pending_sectors'] = raw_value
print(f"[v0] Pending Sectors (ID 197): {raw_value}") print(f"[v0] Pending Sectors (ID 197): {raw_value}")
elif attr_id == 199: # UDMA_CRC_Error_Count elif attr_id == 199:
smart_data['crc_errors'] = raw_value smart_data['crc_errors'] = raw_value
print(f"[v0] CRC Errors (ID 199): {raw_value}") print(f"[v0] CRC Errors (ID 199): {raw_value}")
elif attr_id == 177: # Wear_Leveling_Count
smart_data['wear_leveling_count'] = normalized_value
smart_data['ssd_life_left'] = normalized_value
print(f"[v0] Wear Leveling Count (ID 177): {normalized_value}%")
elif attr_id == 231: # SSD_Life_Left or Temperature
if 'Life' in attr_name or 'life' in attr_name:
smart_data['ssd_life_left'] = normalized_value
print(f"[v0] SSD Life Left (ID 231): {normalized_value}%")
elif attr_id == 233: # Media_Wearout_Indicator (Intel)
smart_data['media_wearout_indicator'] = normalized_value
smart_data['ssd_life_left'] = normalized_value
print(f"[v0] Media Wearout Indicator (ID 233): {normalized_value}%")
elif attr_id == 202: # Percent_Lifetime_Remain
smart_data['ssd_life_left'] = normalized_value
print(f"[v0] Percent Lifetime Remain (ID 202): {normalized_value}%")
# Parse NVMe SMART data
if 'nvme_smart_health_information_log' in data: if 'nvme_smart_health_information_log' in data:
print(f"[v0] Parsing NVMe SMART data...") print(f"[v0] Parsing NVMe SMART data...")
nvme_data = data['nvme_smart_health_information_log'] nvme_data = data['nvme_smart_health_information_log']
smart_data['disk_type'] = 'NVMe'
if 'temperature' in nvme_data: if 'temperature' in nvme_data:
smart_data['temperature'] = nvme_data['temperature'] smart_data['temperature'] = nvme_data['temperature']
print(f"[v0] NVMe Temperature: {smart_data['temperature']}°C") print(f"[v0] NVMe Temperature: {smart_data['temperature']}°C")
@@ -776,8 +808,12 @@ def get_smart_data(disk_name):
if 'power_cycles' in nvme_data: if 'power_cycles' in nvme_data:
smart_data['power_cycles'] = nvme_data['power_cycles'] smart_data['power_cycles'] = nvme_data['power_cycles']
print(f"[v0] NVMe Power Cycles: {smart_data['power_cycles']}") print(f"[v0] NVMe Power Cycles: {smart_data['power_cycles']}")
if 'percentage_used' in nvme_data:
smart_data['percentage_used'] = nvme_data['percentage_used']
smart_data['ssd_life_left'] = 100 - nvme_data['percentage_used']
print(f"[v0] NVMe Percentage Used: {smart_data['percentage_used']}%")
print(f"[v0] NVMe Life Left: {smart_data['ssd_life_left']}%")
# If we got good data, break out of the loop
if smart_data['model'] != 'Unknown' and smart_data['serial'] != 'Unknown': if smart_data['model'] != 'Unknown' and smart_data['serial'] != 'Unknown':
print(f"[v0] Successfully extracted complete data from JSON (attempt {cmd_index + 1})") print(f"[v0] Successfully extracted complete data from JSON (attempt {cmd_index + 1})")
break break
@@ -786,14 +822,12 @@ def get_smart_data(disk_name):
print(f"[v0] JSON parse failed: {e}, trying text parsing...") print(f"[v0] JSON parse failed: {e}, trying text parsing...")
if smart_data['model'] == 'Unknown' or smart_data['serial'] == 'Unknown' or smart_data['temperature'] == 0: if smart_data['model'] == 'Unknown' or smart_data['serial'] == 'Unknown' or smart_data['temperature'] == 0:
print(f"[v0] Parsing text output (model={smart_data['model']}, serial={smart_data['serial']}, temp={smart_data['temperature']})...") print(f"[v0] Parsing text output...")
output = stdout output = stdout
# Get basic info
for line in output.split('\n'): for line in output.split('\n'):
line = line.strip() line = line.strip()
# Model detection
if (line.startswith('Device Model:') or line.startswith('Model Number:')) and smart_data['model'] == 'Unknown': if (line.startswith('Device Model:') or line.startswith('Model Number:')) and smart_data['model'] == 'Unknown':
smart_data['model'] = line.split(':', 1)[1].strip() smart_data['model'] = line.split(':', 1)[1].strip()
print(f"[v0] Found model: {smart_data['model']}") print(f"[v0] Found model: {smart_data['model']}")
@@ -801,7 +835,6 @@ def get_smart_data(disk_name):
smart_data['model'] = line.split(':', 1)[1].strip() smart_data['model'] = line.split(':', 1)[1].strip()
print(f"[v0] Found model family: {smart_data['model']}") print(f"[v0] Found model family: {smart_data['model']}")
# Serial detection
elif line.startswith('Serial Number:') and smart_data['serial'] == 'Unknown': elif line.startswith('Serial Number:') and smart_data['serial'] == 'Unknown':
smart_data['serial'] = line.split(':', 1)[1].strip() smart_data['serial'] = line.split(':', 1)[1].strip()
print(f"[v0] Found serial: {smart_data['serial']}") print(f"[v0] Found serial: {smart_data['serial']}")
@@ -811,14 +844,15 @@ def get_smart_data(disk_name):
if 'rpm' in rate_str.lower(): if 'rpm' in rate_str.lower():
try: try:
smart_data['rotation_rate'] = int(rate_str.split()[0]) smart_data['rotation_rate'] = int(rate_str.split()[0])
print(f"[v0] Found rotation rate: {smart_data['rotation_rate']} RPM") smart_data['disk_type'] = 'HDD'
print(f"[v0] Found HDD rotation rate: {smart_data['rotation_rate']} RPM")
except (ValueError, IndexError): except (ValueError, IndexError):
pass pass
elif 'Solid State Device' in rate_str: elif 'Solid State Device' in rate_str:
smart_data['rotation_rate'] = 0 # SSD smart_data['rotation_rate'] = 0
smart_data['disk_type'] = 'SSD'
print(f"[v0] Found SSD (no rotation)") print(f"[v0] Found SSD (no rotation)")
# SMART status detection
elif 'SMART overall-health self-assessment test result:' in line: elif 'SMART overall-health self-assessment test result:' in line:
if 'PASSED' in line: if 'PASSED' in line:
smart_data['smart_status'] = 'passed' smart_data['smart_status'] = 'passed'
@@ -829,14 +863,12 @@ def get_smart_data(disk_name):
smart_data['health'] = 'critical' smart_data['health'] = 'critical'
print(f"[v0] SMART status: FAILED") print(f"[v0] SMART status: FAILED")
# NVMe health
elif 'SMART Health Status:' in line: elif 'SMART Health Status:' in line:
if 'OK' in line: if 'OK' in line:
smart_data['smart_status'] = 'passed' smart_data['smart_status'] = 'passed'
smart_data['health'] = 'healthy' smart_data['health'] = 'healthy'
print(f"[v0] NVMe Health: OK") print(f"[v0] NVMe Health: OK")
# Temperature detection (various formats)
elif 'Current Temperature:' in line and smart_data['temperature'] == 0: elif 'Current Temperature:' in line and smart_data['temperature'] == 0:
try: try:
temp_str = line.split(':')[1].strip().split()[0] temp_str = line.split(':')[1].strip().split()[0]
@@ -845,7 +877,15 @@ def get_smart_data(disk_name):
except (ValueError, IndexError): except (ValueError, IndexError):
pass pass
# Parse SMART attributes table elif 'Percentage Used:' in line:
try:
percentage_str = line.split(':')[1].strip().rstrip('%')
smart_data['percentage_used'] = int(percentage_str)
smart_data['ssd_life_left'] = 100 - smart_data['percentage_used']
print(f"[v0] NVMe Percentage Used: {smart_data['percentage_used']}%")
except (ValueError, IndexError):
pass
in_attributes = False in_attributes = False
for line in output.split('\n'): for line in output.split('\n'):
line = line.strip() line = line.strip()
@@ -856,7 +896,6 @@ def get_smart_data(disk_name):
continue continue
if in_attributes: if in_attributes:
# Stop at empty line or next section
if not line or line.startswith('SMART') or line.startswith('==='): if not line or line.startswith('SMART') or line.startswith('==='):
in_attributes = False in_attributes = False
continue continue
@@ -865,40 +904,56 @@ def get_smart_data(disk_name):
if len(parts) >= 10: if len(parts) >= 10:
try: try:
attr_id = parts[0] attr_id = parts[0]
# Raw value is typically the last column attr_name = parts[1]
normalized_value = int(parts[3])
raw_value = parts[-1] raw_value = parts[-1]
# Parse based on attribute ID if attr_id == '9':
if attr_id == '9': # Power On Hours
raw_clean = raw_value.split()[0].replace('h', '').replace(',', '') raw_clean = raw_value.split()[0].replace('h', '').replace(',', '')
smart_data['power_on_hours'] = int(raw_clean) smart_data['power_on_hours'] = int(raw_clean)
print(f"[v0] Power On Hours: {smart_data['power_on_hours']}") print(f"[v0] Power On Hours: {smart_data['power_on_hours']}")
elif attr_id == '12': # Power Cycle Count elif attr_id == '12':
raw_clean = raw_value.split()[0].replace(',', '') raw_clean = raw_value.split()[0].replace(',', '')
smart_data['power_cycles'] = int(raw_clean) smart_data['power_cycles'] = int(raw_clean)
print(f"[v0] Power Cycles: {smart_data['power_cycles']}") print(f"[v0] Power Cycles: {smart_data['power_cycles']}")
elif attr_id == '194' and smart_data['temperature'] == 0: # Temperature elif attr_id == '194' and smart_data['temperature'] == 0:
temp_str = raw_value.split()[0] temp_str = raw_value.split()[0]
smart_data['temperature'] = int(temp_str) smart_data['temperature'] = int(temp_str)
print(f"[v0] Temperature (attr 194): {smart_data['temperature']}°C") print(f"[v0] Temperature (attr 194): {smart_data['temperature']}°C")
elif attr_id == '190' and smart_data['temperature'] == 0: # Airflow Temperature elif attr_id == '190' and smart_data['temperature'] == 0:
temp_str = raw_value.split()[0] temp_str = raw_value.split()[0]
smart_data['temperature'] = int(temp_str) smart_data['temperature'] = int(temp_str)
print(f"[v0] Airflow Temperature (attr 190): {smart_data['temperature']}°C") print(f"[v0] Airflow Temperature (attr 190): {smart_data['temperature']}°C")
elif attr_id == '5': # Reallocated Sectors elif attr_id == '5':
smart_data['reallocated_sectors'] = int(raw_value) smart_data['reallocated_sectors'] = int(raw_value)
print(f"[v0] Reallocated Sectors: {smart_data['reallocated_sectors']}") print(f"[v0] Reallocated Sectors: {smart_data['reallocated_sectors']}")
elif attr_id == '197': # Pending Sectors elif attr_id == '197':
smart_data['pending_sectors'] = int(raw_value) smart_data['pending_sectors'] = int(raw_value)
print(f"[v0] Pending Sectors: {smart_data['pending_sectors']}") print(f"[v0] Pending Sectors: {smart_data['pending_sectors']}")
elif attr_id == '199': # CRC Errors elif attr_id == '199':
smart_data['crc_errors'] = int(raw_value) smart_data['crc_errors'] = int(raw_value)
print(f"[v0] CRC Errors: {smart_data['crc_errors']}") print(f"[v0] CRC Errors: {smart_data['crc_errors']}")
# Parsear atributos de desgaste SSD
elif attr_id == '177': # Wear_Leveling_Count
smart_data['wear_leveling_count'] = normalized_value
smart_data['ssd_life_left'] = normalized_value
print(f"[v0] Wear Leveling Count: {normalized_value}%")
elif attr_id == '231':
if 'Life' in attr_name or 'life' in attr_name:
smart_data['ssd_life_left'] = normalized_value
print(f"[v0] SSD Life Left: {normalized_value}%")
elif attr_id == '233': # Media_Wearout_Indicator
smart_data['media_wearout_indicator'] = normalized_value
smart_data['ssd_life_left'] = normalized_value
print(f"[v0] Media Wearout Indicator: {normalized_value}%")
elif attr_id == '202': # Percent_Lifetime_Remain
smart_data['ssd_life_left'] = normalized_value
print(f"[v0] Percent Lifetime Remain: {normalized_value}%")
except (ValueError, IndexError) as e: except (ValueError, IndexError) as e:
print(f"[v0] Error parsing attribute line '{line}': {e}") print(f"[v0] Error parsing attribute line '{line}': {e}")
continue continue
# If we got complete data, break
if smart_data['model'] != 'Unknown' and smart_data['serial'] != 'Unknown': if smart_data['model'] != 'Unknown' and smart_data['serial'] != 'Unknown':
print(f"[v0] Successfully extracted complete data from text output (attempt {cmd_index + 1})") print(f"[v0] Successfully extracted complete data from text output (attempt {cmd_index + 1})")
break break
@@ -919,7 +974,6 @@ def get_smart_data(disk_name):
process.kill() process.kill()
continue continue
finally: finally:
# Ensure the process is terminated if it's still running
if process and process.poll() is None: if process and process.poll() is None:
try: try:
process.kill() process.kill()
@@ -927,7 +981,6 @@ def get_smart_data(disk_name):
except Exception as kill_err: except Exception as kill_err:
print(f"[v0] Error killing process: {kill_err}") print(f"[v0] Error killing process: {kill_err}")
if smart_data['reallocated_sectors'] > 0 or smart_data['pending_sectors'] > 0: if smart_data['reallocated_sectors'] > 0 or smart_data['pending_sectors'] > 0:
if smart_data['health'] == 'healthy': if smart_data['health'] == 'healthy':
smart_data['health'] = 'warning' smart_data['health'] = 'warning'
@@ -939,14 +992,39 @@ def get_smart_data(disk_name):
smart_data['health'] = 'critical' smart_data['health'] = 'critical'
print(f"[v0] Health: CRITICAL (SMART failed)") print(f"[v0] Health: CRITICAL (SMART failed)")
# Temperature-based health (only if we have a valid temperature)
if smart_data['health'] == 'healthy' and smart_data['temperature'] > 0: if smart_data['health'] == 'healthy' and smart_data['temperature'] > 0:
if smart_data['temperature'] >= 70: disk_type = smart_data['disk_type']
temp = smart_data['temperature']
if disk_type == 'HDD':
if temp > 55:
smart_data['health'] = 'critical' smart_data['health'] = 'critical'
print(f"[v0] Health: CRITICAL (temperature {smart_data['temperature']}°C)") print(f"[v0] Health: CRITICAL (HDD temperature {temp}°C > 55°C)")
elif smart_data['temperature'] >= 60: elif temp > 45:
smart_data['health'] = 'warning' smart_data['health'] = 'warning'
print(f"[v0] Health: WARNING (temperature {smart_data['temperature']}°C)") print(f"[v0] Health: WARNING (HDD temperature {temp}°C > 45°C)")
elif disk_type == 'SSD':
if temp > 65:
smart_data['health'] = 'critical'
print(f"[v0] Health: CRITICAL (SSD temperature {temp}°C > 65°C)")
elif temp > 55:
smart_data['health'] = 'warning'
print(f"[v0] Health: WARNING (SSD temperature {temp}°C > 55°C)")
elif disk_type == 'NVMe':
if temp > 70:
smart_data['health'] = 'critical'
print(f"[v0] Health: CRITICAL (NVMe temperature {temp}°C > 70°C)")
elif temp > 60:
smart_data['health'] = 'warning'
print(f"[v0] Health: WARNING (NVMe temperature {temp}°C > 60°C)")
else:
# Umbral genérico si no se detectó el tipo
if temp >= 70:
smart_data['health'] = 'critical'
print(f"[v0] Health: CRITICAL (temperature {temp}°C)")
elif temp >= 60:
smart_data['health'] = 'warning'
print(f"[v0] Health: WARNING (temperature {temp}°C)")
except FileNotFoundError: except FileNotFoundError:
print(f"[v0] ERROR: smartctl not found - install smartmontools for disk monitoring.") print(f"[v0] ERROR: smartctl not found - install smartmontools for disk monitoring.")
@@ -3419,7 +3497,7 @@ def api_logs():
'level': level, 'level': level,
'service': log_entry.get('_SYSTEMD_UNIT', log_entry.get('SYSLOG_IDENTIFIER', 'system')), 'service': log_entry.get('_SYSTEMD_UNIT', log_entry.get('SYSLOG_IDENTIFIER', 'system')),
'message': log_entry.get('MESSAGE', ''), 'message': log_entry.get('MESSAGE', ''),
'source': 'journalctl', 'source': 'journal',
'pid': log_entry.get('_PID', ''), 'pid': log_entry.get('_PID', ''),
'hostname': log_entry.get('_HOSTNAME', '') 'hostname': log_entry.get('_HOSTNAME', '')
}) })
@@ -3976,6 +4054,23 @@ def api_prometheus():
metrics.append(f'# TYPE proxmox_disk_usage_percent gauge') metrics.append(f'# TYPE proxmox_disk_usage_percent gauge')
metrics.append(f'proxmox_disk_usage_percent{{node="{node}",disk="{disk_name}"}} {disk.get("percent", 0)} {timestamp}') metrics.append(f'proxmox_disk_usage_percent{{node="{node}",disk="{disk_name}"}} {disk.get("percent", 0)} {timestamp}')
# Métricas de SMART para discos SSD/NVMe
if disk.get('disk_type') in ['SSD', 'NVMe']:
if disk.get('ssd_life_left') is not None and disk.get('ssd_life_left') != 100: # Solo agregar si es diferente del default (100)
metrics.append(f'# HELP proxmox_disk_ssd_life_left SSD remaining life percentage')
metrics.append(f'# TYPE proxmox_disk_ssd_life_left gauge')
metrics.append(f'proxmox_disk_ssd_life_left{{node="{node}",disk="{disk_name}"}} {disk["ssd_life_left"]} {timestamp}')
if disk.get('percentage_used') is not None and disk.get('percentage_used') > 0:
metrics.append(f'# HELP proxmox_disk_nvme_percentage_used NVMe disk percentage used')
metrics.append(f'# TYPE proxmox_disk_nvme_percentage_used gauge')
metrics.append(f'proxmox_disk_nvme_percentage_used{{node="{node}",disk="{disk_name}"}} {disk["percentage_used"]} {timestamp}')
if disk.get('wear_leveling_count') is not None and disk.get('wear_leveling_count') > 0:
metrics.append(f'# HELP proxmox_disk_ssd_wear_leveling_count SSD wear leveling count percentage')
metrics.append(f'# TYPE proxmox_disk_ssd_wear_leveling_count gauge')
metrics.append(f'proxmox_disk_ssd_wear_leveling_count{{node="{node}",disk="{disk_name}"}} {disk["wear_leveling_count"]} {timestamp}')
# Network metrics # Network metrics
network_info = get_network_info() network_info = get_network_info()
if 'traffic' in network_info: if 'traffic' in network_info:
@@ -4067,28 +4162,28 @@ def api_prometheus():
# GPU metrics # GPU metrics
pci_devices = hardware_info.get('pci_devices', []) pci_devices = hardware_info.get('pci_devices', [])
for device in pci_devices: for device in pci_devices:
if device.get('type') == 'GPU': if device.get('type') == 'Graphics Card': # Check for Graphics Card type specifically
gpu_name = device.get('device', 'unknown').replace(' ', '_') gpu_name = device.get('device', 'unknown').replace(' ', '_')
gpu_vendor = device.get('vendor', 'unknown') gpu_vendor = device.get('vendor', 'unknown')
# GPU Temperature # GPU Temperature
if device.get('gpu_temperature') is not None: if device.get('temperature') is not None:
metrics.append(f'# HELP proxmox_gpu_temperature_celsius GPU temperature in Celsius') metrics.append(f'# HELP proxmox_gpu_temperature_celsius GPU temperature in Celsius')
metrics.append(f'# TYPE proxmox_gpu_temperature_celsius gauge') metrics.append(f'# TYPE proxmox_gpu_temperature_celsius gauge')
metrics.append(f'proxmox_gpu_temperature_celsius{{node="{node}",gpu="{gpu_name}",vendor="{gpu_vendor}"}} {device["gpu_temperature"]} {timestamp}') metrics.append(f'proxmox_gpu_temperature_celsius{{node="{node}",gpu="{gpu_name}",vendor="{gpu_vendor}"}} {device["temperature"]} {timestamp}')
# GPU Utilization # GPU Utilization
if device.get('gpu_utilization') is not None: if device.get('utilization_gpu') is not None:
metrics.append(f'# HELP proxmox_gpu_utilization_percent GPU utilization percentage') metrics.append(f'# HELP proxmox_gpu_utilization_percent GPU utilization percentage')
metrics.append(f'# TYPE proxmox_gpu_utilization_percent gauge') metrics.append(f'# TYPE proxmox_gpu_utilization_percent gauge')
metrics.append(f'proxmox_gpu_utilization_percent{{node="{node}",gpu="{gpu_name}",vendor="{gpu_vendor}"}} {device["gpu_utilization"]} {timestamp}') metrics.append(f'proxmox_gpu_utilization_percent{{node="{node}",gpu="{gpu_name}",vendor="{gpu_vendor}"}} {device["utilization_gpu"]} {timestamp}')
# GPU Memory # GPU Memory
if device.get('gpu_memory_used') and device.get('gpu_memory_total'): if device.get('memory_used') and device.get('memory_total'):
try: try:
# Extract numeric values from strings like "1024 MiB" # Extract numeric values from strings like "1024 MiB"
mem_used = float(device['gpu_memory_used'].split()[0]) mem_used = float(device['memory_used'].split()[0])
mem_total = float(device['gpu_memory_total'].split()[0]) mem_total = float(device['memory_total'].split()[0])
mem_used_bytes = mem_used * 1024 * 1024 # Convert MiB to bytes mem_used_bytes = mem_used * 1024 * 1024 # Convert MiB to bytes
mem_total_bytes = mem_total * 1024 * 1024 mem_total_bytes = mem_total * 1024 * 1024
@@ -4103,10 +4198,10 @@ def api_prometheus():
pass pass
# GPU Power Draw (NVIDIA only) # GPU Power Draw (NVIDIA only)
if device.get('gpu_power_draw'): if device.get('power_draw'):
try: try:
# Extract numeric value from string like "75.5 W" # Extract numeric value from string like "75.5 W"
power_draw = float(device['gpu_power_draw'].split()[0]) power_draw = float(device['power_draw'].split()[0])
metrics.append(f'# HELP proxmox_gpu_power_draw_watts GPU power draw in watts') metrics.append(f'# HELP proxmox_gpu_power_draw_watts GPU power draw in watts')
metrics.append(f'# TYPE proxmox_gpu_power_draw_watts gauge') metrics.append(f'# TYPE proxmox_gpu_power_draw_watts gauge')
metrics.append(f'proxmox_gpu_power_draw_watts{{node="{node}",gpu="{gpu_name}",vendor="{gpu_vendor}"}} {power_draw} {timestamp}') metrics.append(f'proxmox_gpu_power_draw_watts{{node="{node}",gpu="{gpu_name}",vendor="{gpu_vendor}"}} {power_draw} {timestamp}')
@@ -4114,20 +4209,20 @@ def api_prometheus():
pass pass
# GPU Clock Speeds (NVIDIA only) # GPU Clock Speeds (NVIDIA only)
if device.get('gpu_clock_speed'): if device.get('clock_graphics'):
try: try:
# Extract numeric value from string like "1500 MHz" # Extract numeric value from string like "1500 MHz"
clock_speed = float(device['gpu_clock_speed'].split()[0]) clock_speed = float(device['clock_graphics'].split()[0])
metrics.append(f'# HELP proxmox_gpu_clock_speed_mhz GPU clock speed in MHz') metrics.append(f'# HELP proxmox_gpu_clock_speed_mhz GPU clock speed in MHz')
metrics.append(f'# TYPE proxmox_gpu_clock_speed_mhz gauge') metrics.append(f'# TYPE proxmox_gpu_clock_speed_mhz gauge')
metrics.append(f'proxmox_gpu_clock_speed_mhz{{node="{node}",gpu="{gpu_name}",vendor="{gpu_vendor}"}} {clock_speed} {timestamp}') metrics.append(f'proxmox_gpu_clock_speed_mhz{{node="{node}",gpu="{gpu_name}",vendor="{gpu_vendor}"}} {clock_speed} {timestamp}')
except (ValueError, IndexError): except (ValueError, IndexError):
pass pass
if device.get('gpu_memory_clock'): if device.get('clock_memory'):
try: try:
# Extract numeric value from string like "5001 MHz" # Extract numeric value from string like "5001 MHz"
mem_clock = float(device['gpu_memory_clock'].split()[0]) mem_clock = float(device['clock_memory'].split()[0])
metrics.append(f'# HELP proxmox_gpu_memory_clock_mhz GPU memory clock speed in MHz') metrics.append(f'# HELP proxmox_gpu_memory_clock_mhz GPU memory clock speed in MHz')
metrics.append(f'# TYPE proxmox_gpu_memory_clock_mhz gauge') metrics.append(f'# TYPE proxmox_gpu_memory_clock_mhz gauge')
metrics.append(f'proxmox_gpu_memory_clock_mhz{{node="{node}",gpu="{gpu_name}",vendor="{gpu_vendor}"}} {mem_clock} {timestamp}') metrics.append(f'proxmox_gpu_memory_clock_mhz{{node="{node}",gpu="{gpu_name}",vendor="{gpu_vendor}"}} {mem_clock} {timestamp}')