690619:1128 240 #02
CI / CD Pipeline / build (push) Successful in 7m30s
CI / CD Pipeline / deploy (push) Successful in 14m25s

This commit is contained in:
2026-06-19 11:28:59 +07:00
parent 5f29fdbe8c
commit 12e230332e
3 changed files with 120 additions and 80 deletions
@@ -138,10 +138,31 @@ export class VramMonitorService {
/** /**
* ตรวจสอบว่า VRAM เพียงพอสำหรับความต้องการโหลดโมเดลหรือไม่ * ตรวจสอบว่า VRAM เพียงพอสำหรับความต้องการโหลดโมเดลหรือไม่
* ถ้าไม่มีโมเดลโหลดอยู่เลย จะอนุญาตให้โหลดโมเดลใหม่ได้เสมอ (ป้องกัน false positive)
*/ */
async hasVramCapacity(requiredMb: number): Promise<boolean> { async hasVramCapacity(requiredMb: number): Promise<boolean> {
const headroom = await this.getVramHeadroom(); const headroom = await this.getVramHeadroom();
return headroom.availableMb >= requiredMb; // ถ้าไม่มีโมเดลโหลดอยู่เลย อนุญาตให้โหลดโมเดลใหม่ได้เสมอ
if (headroom.usedMb === 0 && headroom.querySuccess) {
this.logger.log(
`No models loaded in VRAM, allowing model load (required=${requiredMb}MB)`
);
return true;
}
// ถ้า query ล้มเหลว ใช้ optimistic fallback (assume no VRAM used)
if (!headroom.querySuccess) {
this.logger.log(
`VRAM query failed, using optimistic fallback (required=${requiredMb}MB)`
);
return true;
}
const hasCapacity = headroom.availableMb >= requiredMb;
if (!hasCapacity) {
this.logger.warn(
`VRAM insufficient: available=${headroom.availableMb}MB, required=${requiredMb}MB`
);
}
return hasCapacity;
} }
/** /**
@@ -100,6 +100,22 @@ describe('VramMonitorService', () => {
const result = await service.hasVramCapacity(3000); // query available is 2048MB, required 3000MB const result = await service.hasVramCapacity(3000); // query available is 2048MB, required 3000MB
expect(result).toBe(false); expect(result).toBe(false);
}); });
it('ควรคืน true เมื่อไม่มีโมเดลโหลดอยู่เลย (ป้องกัน false positive)', async () => {
mockedAxios.get.mockResolvedValue({
data: {
models: [], // ไม่มีโมเดลโหลด
},
});
const result = await service.hasVramCapacity(5000); // ต้องการ 5GB แม้ availableMb = 8192MB
expect(result).toBe(true);
});
it('ควรคืน true เมื่อ query ล้มเหลว (optimistic fallback)', async () => {
mockedAxios.get.mockRejectedValue(new Error('Connection timeout'));
const result = await service.hasVramCapacity(5000);
expect(result).toBe(true);
});
}); });
describe('getVramStatus', () => { describe('getVramStatus', () => {
it('ควรคืน status ที่ถูกต้องเมื่อ Ollama คืน models', async () => { it('ควรคืน status ที่ถูกต้องเมื่อ Ollama คืน models', async () => {
+82 -79
View File
@@ -615,85 +615,6 @@ export default function AiAdminConsolePage() {
</div> </div>
</div> </div>
<Card>
<CardHeader>
<CardTitle className="flex items-center gap-2 text-lg">
<Power className="h-5 w-5" />
System Toggle
</CardTitle>
</CardHeader>
<CardContent>
<div className="flex flex-col gap-5 sm:flex-row sm:items-center sm:justify-between">
<div className="space-y-1">
<div className="text-base font-medium">
{aiEnabled ? 'AI พร้อมให้ผู้ใช้ทั่วไปใช้งาน' : 'AI ถูกปิดสำหรับผู้ใช้ทั่วไป'}
</div>
<div className="text-sm text-muted-foreground">
Superadmin
</div>
<div className="text-xs text-muted-foreground flex items-center gap-1.5 pt-1 flex-wrap">
<span>Active Models:</span>
<Badge
variant="outline"
className="text-[10px] py-0 px-1.5 border-primary/20 text-primary bg-primary/5 font-semibold"
>
{isHealthLoading ? 'Loading...' : toCanonicalModel(health?.activeModels?.main ?? 'np-dms-ai')}
</Badge>
<span className="text-muted-foreground/50">+</span>
<Badge
variant="outline"
className="text-[10px] py-0 px-1.5 border-purple-500/20 text-purple-600 dark:text-purple-400 bg-purple-500/5 font-semibold"
>
{isHealthLoading ? 'Loading...' : toCanonicalModel(health?.activeModels?.ocr ?? 'np-dms-ocr')}
</Badge>
</div>
</div>
<div className="flex items-center gap-3">
{busy && <Loader2 className="h-4 w-4 animate-spin" />}
<Switch
checked={aiEnabled}
disabled={busy || isError}
aria-label="Toggle AI features"
onCheckedChange={handleToggle}
/>
</div>
</div>
{isError && (
<div className="mt-4 rounded-md border border-destructive/30 bg-destructive/10 p-3 text-sm text-destructive">
AI
</div>
)}
</CardContent>
</Card>
<div className="grid gap-4 md:grid-cols-2">
<Card>
<CardHeader>
<CardTitle className="flex items-center gap-2 text-base">
<ShieldCheck className="h-5 w-5" />
Protection
</CardTitle>
</CardHeader>
<CardContent className="text-sm text-muted-foreground">
AI AI inference endpoints HTTP 503
</CardContent>
</Card>
<Card>
<CardHeader>
<CardTitle className="text-base">Polling</CardTitle>
</CardHeader>
<CardContent className="flex items-center justify-between gap-3 text-sm text-muted-foreground">
<span>
30
{(isFetching || isHealthLoading) && !(isLoading || isHealthLoading) ? ' (กำลังรีเฟรช)' : ''}
</span>
<Button type="button" variant="outline" size="sm" onClick={() => void handleRefreshAll()}>
Refresh
</Button>
</CardContent>
</Card>
</div>
<Tabs defaultValue="overview" className="w-full space-y-6"> <Tabs defaultValue="overview" className="w-full space-y-6">
<TabsList className="grid w-full grid-cols-3 max-w-[500px]"> <TabsList className="grid w-full grid-cols-3 max-w-[500px]">
<TabsTrigger value="overview">System Toggle</TabsTrigger> <TabsTrigger value="overview">System Toggle</TabsTrigger>
@@ -701,6 +622,88 @@ export default function AiAdminConsolePage() {
<TabsTrigger value="sandbox">3-Step Pipeline Sandbox</TabsTrigger> <TabsTrigger value="sandbox">3-Step Pipeline Sandbox</TabsTrigger>
</TabsList> </TabsList>
<TabsContent value="overview" className="space-y-6">
<Card>
<CardHeader>
<CardTitle className="flex items-center gap-2 text-lg">
<Power className="h-5 w-5" />
System Toggle
</CardTitle>
</CardHeader>
<CardContent>
<div className="flex flex-col gap-5 sm:flex-row sm:items-center sm:justify-between">
<div className="space-y-1">
<div className="text-base font-medium">
{aiEnabled ? 'AI พร้อมให้ผู้ใช้ทั่วไปใช้งาน' : 'AI ถูกปิดสำหรับผู้ใช้ทั่วไป'}
</div>
<div className="text-sm text-muted-foreground">
Superadmin
</div>
<div className="text-xs text-muted-foreground flex items-center gap-1.5 pt-1 flex-wrap">
<span>Active Models:</span>
<Badge
variant="outline"
className="text-[10px] py-0 px-1.5 border-primary/20 text-primary bg-primary/5 font-semibold"
>
{isHealthLoading ? 'Loading...' : toCanonicalModel(health?.activeModels?.main ?? 'np-dms-ai')}
</Badge>
<span className="text-muted-foreground/50">+</span>
<Badge
variant="outline"
className="text-[10px] py-0 px-1.5 border-purple-500/20 text-purple-600 dark:text-purple-400 bg-purple-500/5 font-semibold"
>
{isHealthLoading ? 'Loading...' : toCanonicalModel(health?.activeModels?.ocr ?? 'np-dms-ocr')}
</Badge>
</div>
</div>
<div className="flex items-center gap-3">
{busy && <Loader2 className="h-4 w-4 animate-spin" />}
<Switch
checked={aiEnabled}
disabled={busy || isError}
aria-label="Toggle AI features"
onCheckedChange={handleToggle}
/>
</div>
</div>
{isError && (
<div className="mt-4 rounded-md border border-destructive/30 bg-destructive/10 p-3 text-sm text-destructive">
AI
</div>
)}
</CardContent>
</Card>
<div className="grid gap-4 md:grid-cols-2">
<Card>
<CardHeader>
<CardTitle className="flex items-center gap-2 text-base">
<ShieldCheck className="h-5 w-5" />
Protection
</CardTitle>
</CardHeader>
<CardContent className="text-sm text-muted-foreground">
AI AI inference endpoints HTTP 503
</CardContent>
</Card>
<Card>
<CardHeader>
<CardTitle className="text-base">Polling</CardTitle>
</CardHeader>
<CardContent className="flex items-center justify-between gap-3 text-sm text-muted-foreground">
<span>
30
{(isFetching || isHealthLoading) && !(isLoading || isHealthLoading) ? ' (กำลังรีเฟรช)' : ''}
</span>
<Button type="button" variant="outline" size="sm" onClick={() => void handleRefreshAll()}>
Refresh
</Button>
</CardContent>
</Card>
</div>
</TabsContent>
<TabsContent value="playground" className="space-y-6"> <TabsContent value="playground" className="space-y-6">
<Card className="border border-border/50 bg-background/50 backdrop-blur-md"> <Card className="border border-border/50 bg-background/50 backdrop-blur-md">
<CardHeader> <CardHeader>