add gpu usage alerts

internal/alerts/alerts.go

@@ -40,13 +40,18 @@ type UserNotificationSettings struct {
 }
 
 type SystemAlertStats struct {
-	Cpu          float64            `json:"cpu"`
-	Mem          float64            `json:"mp"`
-	Disk         float64            `json:"dp"`
-	NetSent      float64            `json:"ns"`
-	NetRecv      float64            `json:"nr"`
-	Temperatures map[string]float32 `json:"t"`
-	LoadAvg      [3]float64         `json:"la"`
+	Cpu          float64                       `json:"cpu"`
+	Mem          float64                       `json:"mp"`
+	Disk         float64                       `json:"dp"`
+	NetSent      float64                       `json:"ns"`
+	NetRecv      float64                       `json:"nr"`
+	GPU          map[string]SystemAlertGPUData `json:"g"`
+	Temperatures map[string]float32            `json:"t"`
+	LoadAvg      [3]float64                    `json:"la"`
+}
+
+type SystemAlertGPUData struct {
+	Usage float64 `json:"u"`
 }
 
 type SystemAlertData struct {

internal/alerts/alerts_system.go

@@ -64,6 +64,8 @@ func (am *AlertManager) HandleSystemAlerts(systemRecord *core.Record, data *syst
 		case "LoadAvg15":
 			val = data.Info.LoadAvg[2]
 			unit = ""
+		case "GPU":
+			val = data.Info.GpuPct
 		}
 
 		triggered := alertRecord.GetBool("triggered")
@@ -206,6 +208,17 @@ func (am *AlertManager) HandleSystemAlerts(systemRecord *core.Record, data *syst
 				alert.val += stats.LoadAvg[1]
 			case "LoadAvg15":
 				alert.val += stats.LoadAvg[2]
+			case "GPU":
+				if len(stats.GPU) == 0 {
+					continue
+				}
+				maxUsage := 0.0
+				for _, gpu := range stats.GPU {
+					if gpu.Usage > maxUsage {
+						maxUsage = gpu.Usage
+					}
+				}
+				alert.val += maxUsage
 			default:
 				continue
 			}

internal/migrations/0_collections_snapshot_0_16_0.go

@@ -75,6 +75,7 @@ func init() {
 					"Disk",
 					"Temperature",
 					"Bandwidth",
+					"GPU",
 					"LoadAvg1",
 					"LoadAvg5",
 					"LoadAvg15"

internal/site/src/lib/alerts.ts

@@ -1,7 +1,7 @@
 import { t } from "@lingui/core/macro"
 import { CpuIcon, HardDriveIcon, HourglassIcon, MemoryStickIcon, ServerIcon, ThermometerIcon } from "lucide-react"
 import type { RecordSubscription } from "pocketbase"
-import { EthernetIcon } from "@/components/ui/icons"
+import { EthernetIcon, GpuIcon } from "@/components/ui/icons"
 import { $alerts } from "@/lib/stores"
 import type { AlertInfo, AlertRecord } from "@/types"
 import { pb } from "./api"
@@ -41,6 +41,12 @@ export const alertInfo: Record<string, AlertInfo> = {
 		desc: () => t`Triggers when combined up/down exceeds a threshold`,
 		max: 125,
 	},
+	GPU: {
+		name: () => t`GPU Usage`,
+		unit: "%",
+		icon: GpuIcon,
+		desc: () => t`Triggers when GPU usage exceeds a threshold`,
+	},
 	Temperature: {
 		name: () => t`Temperature`,
 		unit: "°C",

internal/site/src/types.d.ts

@@ -301,18 +301,18 @@ export interface ChartData {
 	chartTime: ChartTimes
 }
 
-// interface AlertInfo {
-// 	name: () => string
-// 	unit: string
-// 	icon: any
-// 	desc: () => string
-// 	max?: number
-// 	min?: number
-// 	step?: number
-// 	start?: number
-// 	/** Single value description (when there's only one value, like status) */
-// 	singleDesc?: () => string
-// }
+export interface AlertInfo {
+	name: () => string
+	unit: string
+	icon: any
+	desc: () => string
+	max?: number
+	min?: number
+	step?: number
+	start?: number
+	/** Single value description (when there's only one value, like status) */
+	singleDesc?: () => string
+}
 
 export type AlertMap = Record<string, Map<string, AlertRecord>>