690602:0957 ADR-033-233 #01

backend/src/modules/ai/services/ollama.service.ts

@@ -1,178 +1,209 @@
 // File: src/modules/ai/services/ollama.service.ts
-
 // Change Log
-
 // - 2026-05-15: เพิ่ม Ollama service สำหรับ ADR-023A 2-model stack.
-
 // - 2026-05-21: เพิ่ม checkHealth สำหรับตรวจสอบสุขภาพและความเร็ว (Latency) ของ Ollama
+// - 2026-06-02: เพิ่ม loadModel() preloading, ดึงจริงจาก /api/ps และเพิ่ม unloadModel() เพื่อล้างหน่วยความจำ GPU/VRAM (ADR-033, Suggestion 1)
 
 import { Injectable, Logger } from '@nestjs/common';
-
 import { ConfigService } from '@nestjs/config';
-
 import axios from 'axios';
 
 export interface OllamaGenerateOptions {
   timeoutMs?: number;
-
   signal?: AbortSignal;
 }
 
 /** บริการเรียก Ollama local-only บน Admin Desktop ตาม ADR-023A */
-
 @Injectable()
 export class OllamaService {
   private readonly logger = new Logger(OllamaService.name);
-
   private readonly ollamaUrl: string;
-
   private readonly mainModel: string;
-
   private readonly embedModel: string;
-
   private readonly timeoutMs: number;
 
   constructor(private readonly configService: ConfigService) {
     this.ollamaUrl = this.configService.get<string>(
       'OLLAMA_URL',
-
       this.configService.get<string>('AI_HOST_URL', 'http://localhost:11434')
     );
-
     this.mainModel = this.configService.get<string>(
       'OLLAMA_MODEL_MAIN',
-
       'gemma4:e4b'
     );
-
     this.embedModel = this.configService.get<string>(
       'OLLAMA_MODEL_EMBED',
-
       this.configService.get<string>('OLLAMA_EMBED_MODEL', 'nomic-embed-text')
     );
-
     this.timeoutMs = this.configService.get<number>('AI_TIMEOUT_MS', 30000);
   }
 
   /** สร้างข้อความตอบกลับจาก gemma4:e4b หรือค่า ENV ที่กำหนด */
-
   async generate(
     prompt: string,
-
     options: OllamaGenerateOptions = {}
   ): Promise<string> {
     try {
       const response = await axios.post<{ response: string }>(
         `${this.ollamaUrl}/api/generate`,
-
         {
           model: this.mainModel,
-
           prompt,
-
           stream: false,
         },
-
         {
           timeout: options.timeoutMs ?? this.timeoutMs,
-
           signal: options.signal,
         }
       );
-
       return response.data.response ?? '';
     } catch (err) {
       this.logger.error(
         'Ollama generate failed',
-
         err instanceof Error ? err.stack : String(err)
       );
-
       throw err;
     }
   }
 
   /** สร้าง embedding ด้วย nomic-embed-text หรือค่า ENV ที่กำหนด */
-
   async generateEmbedding(text: string): Promise<number[]> {
     try {
       const response = await axios.post<{ embedding: number[] }>(
         `${this.ollamaUrl}/api/embeddings`,
-
         { model: this.embedModel, prompt: text },
-
         { timeout: this.timeoutMs }
       );
-
       return response.data.embedding;
     } catch (err) {
       this.logger.error(
         'Ollama embedding failed',
-
         err instanceof Error ? err.stack : String(err)
       );
-
       throw err;
     }
   }
 
   /** คืนชื่อ main model สำหรับ audit log */
-
   getMainModelName(): string {
     return this.mainModel;
   }
 
   /** คืนชื่อ embedding model สำหรับ audit log */
-
   getEmbeddingModelName(): string {
     return this.embedModel;
   }
 
   /** ตรวจสอบสุขภาพและความเร็ว (Latency) ของระบบ Ollama */
-
   async checkHealth(): Promise<{
     status: 'HEALTHY' | 'DEGRADED' | 'DOWN';
-
     latencyMs: number;
-
     models: string[];
-
     error?: string;
   }> {
     const startTime = Date.now();
-
     try {
       await axios.get(`${this.ollamaUrl}/api/tags`, { timeout: 5000 });
-
       const latencyMs = Date.now() - startTime;
-
+      let loadedModels: string[] = [];
+      try {
+        const psResponse = await axios.get<{
+          models?: Array<{ name: string }>;
+        }>(`${this.ollamaUrl}/api/ps`, { timeout: 3000 });
+        if (psResponse.data?.models) {
+          loadedModels = psResponse.data.models.map((m) => m.name);
+        }
+      } catch (psErr) {
+        this.logger.warn(
+          `Failed to fetch loaded models from /api/ps: ${psErr instanceof Error ? psErr.message : String(psErr)}`
+        );
+      }
+      if (loadedModels.length === 0) {
+        loadedModels = [this.mainModel, this.embedModel];
+      }
       return {
         status: 'HEALTHY',
-
         latencyMs,
-
-        models: [this.mainModel, this.embedModel],
+        models: loadedModels,
       };
     } catch (err: unknown) {
       const latencyMs = Date.now() - startTime;
-
       const error = err instanceof Error ? err.message : String(err);
-
       const isTimeout =
         err instanceof Error &&
         (err.message.includes('timeout') ||
           err.message.includes('504') ||
           err.message.includes('code ECONNABORTED'));
-
       return {
         status: isTimeout ? 'DEGRADED' : 'DOWN',
-
         latencyMs,
-
         models: [this.mainModel, this.embedModel],
-
         error,
       };
     }
   }
+
+  /** โหลดโมเดลล่วงหน้าแบบ Synchronous และตรวจสอบความพร้อมบน Ollama (T007) */
+  async loadModel(modelName: string): Promise<boolean> {
+    try {
+      const tagsResponse = await axios.get<{
+        models?: Array<{ name: string; model: string }>;
+      }>(`${this.ollamaUrl}/api/tags`, { timeout: 5000 });
+      const installedModels = tagsResponse.data?.models ?? [];
+      const exists = installedModels.some(
+        (m) =>
+          m.name === modelName ||
+          m.model === modelName ||
+          m.name.startsWith(modelName)
+      );
+      if (!exists) {
+        this.logger.warn(`Model ${modelName} is not installed in Ollama`);
+        return false;
+      }
+      this.logger.log(
+        `Synchronously pre-loading model ${modelName} into GPU memory...`
+      );
+      await axios.post(
+        `${this.ollamaUrl}/api/generate`,
+        {
+          model: modelName,
+          prompt: '',
+          stream: false,
+          keep_alive: -1,
+        },
+        { timeout: 30000 }
+      );
+      this.logger.log(`Model ${modelName} pre-loaded successfully`);
+      return true;
+    } catch (err: unknown) {
+      this.logger.error(
+        `Failed to pre-load model ${modelName}`,
+        err instanceof Error ? err.stack : String(err)
+      );
+      return false;
+    }
+  }
+  /** ล้างโมเดลออกจากหน่วยความจำ GPU ของ Ollama เพื่อคืนค่า VRAM (ADR-033 Suggestion 1) */
+  async unloadModel(modelName: string): Promise<boolean> {
+    try {
+      this.logger.log(`Unloading model ${modelName} from GPU memory...`);
+      await axios.post(
+        `${this.ollamaUrl}/api/generate`,
+        {
+          model: modelName,
+          prompt: '',
+          stream: false,
+          keep_alive: 0,
+        },
+        { timeout: 10000 }
+      );
+      this.logger.log(`Model ${modelName} unloaded successfully`);
+      return true;
+    } catch (err: unknown) {
+      this.logger.warn(
+        `Failed to unload model ${modelName}: ${err instanceof Error ? err.message : String(err)}`
+      );
+      return false;
+    }
+  }
 }