chore: reduce destroy attempts

extensions/inference-cortex-extension/bin/version.txt

@@ -1 +1 @@
-1.0.3-rc1
+1.0.3-rc2

extensions/inference-cortex-extension/src/index.ts

@@ -69,11 +69,11 @@ export default class JanInferenceCortexExtension extends LocalOAIEngine {
 
     super.onLoad()
 
+    await this.queue.add(() => this.clean())
     this.queue.add(() => this.healthz())
     this.queue.add(() => this.setDefaultEngine(systemInfo))
     // Run the process watchdog
     const systemInfo = await systemInformation()
-    await this.clean()
     await executeOnMain(NODE, 'run', systemInfo)
     this.subscribeToEvents()
 
@@ -160,7 +160,8 @@ export default class JanInferenceCortexExtension extends LocalOAIEngine {
     return ky
       .get(`${CORTEX_API_URL}/healthz`, {
         retry: {
-          limit: 10,
+          limit: 20,
+          delay: () => 500,
           methods: ['get'],
         },
       })
@@ -192,6 +193,9 @@ export default class JanInferenceCortexExtension extends LocalOAIEngine {
     return ky
       .delete(`${CORTEX_API_URL}/processmanager/destroy`, {
         timeout: 2000, // maximum 2 seconds
+        retry: {
+          limit: 0,
+        },
       })
       .catch(() => {
         // Do nothing

extensions/model-extension/src/cortex.ts

@@ -1,6 +1,6 @@
 import PQueue from 'p-queue'
 import ky from 'ky'
-import {  extractModelLoadParams, Model } from '@janhq/core'
+import { extractModelLoadParams, Model } from '@janhq/core'
 import { extractInferenceParams } from '@janhq/core'
 /**
  * cortex.cpp Model APIs interface
@@ -155,7 +155,8 @@ export class CortexAPI implements ICortexAPI {
     return ky
       .get(`${API_URL}/healthz`, {
         retry: {
-          limit: 10,
+          limit: 20,
+          delay: () => 500,
           methods: ['get'],
         },
       })