chore: attempt to release running services on exit (#927)

Author: louis-jan

cortex-js/src/index.ts

@@ -1,4 +1,5 @@
 import {
+  CORTEX_CPP_PROCESS_DESTROY_URL,
   CORTEX_JS_SYSTEM_URL,
   defaultCortexJsHost,
   defaultCortexJsPort,
@@ -36,5 +37,12 @@ export async function start(host?: string, port?: number) {
 export async function stop(host?: string, port?: number) {
   return fetch(CORTEX_JS_SYSTEM_URL(host, port), {
     method: 'DELETE',
-  }).catch(() => {});
+  })
+    .catch(() => {})
+    .then(() =>
+      fetch(CORTEX_CPP_PROCESS_DESTROY_URL(host, port), {
+        method: 'DELETE',
+      }),
+    )
+    .catch(() => {});
 }

cortex-js/src/usecases/cortex/cortex.usecases.ts

@@ -1,4 +1,8 @@
-import { HttpStatus, Injectable } from '@nestjs/common';
+import {
+  BeforeApplicationShutdown,
+  HttpStatus,
+  Injectable,
+} from '@nestjs/common';
 import { ChildProcess, fork } from 'child_process';
 import { delimiter, join } from 'path';
 import { CortexOperationSuccessfullyDto } from '@/infrastructure/dtos/cortex/cortex-operation-successfully.dto';
@@ -16,7 +20,7 @@ import {
 import { openSync } from 'fs';
 
 @Injectable()
-export class CortexUsecases {
+export class CortexUsecases implements BeforeApplicationShutdown {
   private cortexProcess: ChildProcess | undefined;
 
   constructor(
@@ -44,6 +48,9 @@ export class CortexUsecases {
     const dataFolderPath = await this.fileManagerService.getDataFolderPath();
 
     const writer = openSync(await this.fileManagerService.getLogPath(), 'a+');
+
+    // Attempt to stop the process if it's already running
+    await this.stopCortex();
     // go up one level to get the binary folder, have to also work on windows
     this.cortexProcess = fork(join(__dirname, './../../utils/cortex-cpp'), [], {
       detached: true,
@@ -103,8 +110,6 @@ export class CortexUsecases {
           ),
         ),
       );
-    } catch (err) {
-      console.error(err.response.data);
     } finally {
       this.cortexProcess?.kill();
       return {
@@ -211,4 +216,9 @@ export class CortexUsecases {
       cortexCppPort: port,
     });
   }
+
+  async beforeApplicationShutdown(signal: string) {
+    console.log(`Received ${signal}, performing pre-shutdown tasks.`);
+    await this.stopCortex();
+  }
 }

cortex-js/src/usecases/engines/engines.usecase.ts

@@ -120,10 +120,11 @@ export class EnginesUsecases {
     }
 
     if (
-      (engine === Engines.llamaCPP || engine === Engines.tensorrtLLM) &&
-      options?.runMode === 'GPU' &&
-      options?.gpuType === 'Nvidia' &&
-      !options?.vulkan
+      engine === Engines.tensorrtLLM ||
+      (engine === Engines.llamaCPP &&
+        options?.runMode === 'GPU' &&
+        options?.gpuType === 'Nvidia' &&
+        !options?.vulkan)
     )
       await this.installCudaToolkitDependency(
         engine === Engines.tensorrtLLM