Mise en place de la pipeline pour github plutot que gitea ; mise en place des images docker sur GHCR plutôt que gitea

Passage version v0.6.13

.gitea/workflows/release.yml

@@ -6,8 +6,10 @@ on:
       - 'v*'
 
 env:
-  REGISTRY: git.igmlcreation.fr
-  REGISTRY_USER: ietm64
+  GITEA_REGISTRY: git.igmlcreation.fr
+  GITEA_REGISTRY_USER: ietm64
+  GHCR_REGISTRY: ghcr.io
+  GHCR_NAMESPACE: igmlcreation
 
 jobs:
   build:
@@ -26,19 +28,39 @@ jobs:
       - name: Login to Gitea Registry
         uses: docker/login-action@v3
         with:
-          registry: ${{ env.REGISTRY }}
-          username: ${{ env.REGISTRY_USER }}
+          registry: ${{ env.GITEA_REGISTRY }}
+          username: ${{ env.GITEA_REGISTRY_USER }}
           password: ${{ secrets.DOCKER_PAT }}
 
+      # Login to GHCR (GitHub Container Registry) pour distribuer les images
+      # publiquement aux utilisateurs finaux. Reputation domaine plus elevee
+      # que git.igmlcreation.fr (mieux pour les antivirus / SmartScreen).
+      - name: Login to GHCR
+        uses: docker/login-action@v3
+        with:
+          registry: ${{ env.GHCR_REGISTRY }}
+          username: ${{ env.GHCR_NAMESPACE }}
+          password: ${{ secrets.GHCR_TOKEN }}
+
       - name: Extract version
         id: meta
         run: echo "version=${GITHUB_REF_NAME#v}" >> $GITHUB_OUTPUT
 
+      # Push vers les deux registries en un seul build (build-push-action
+      # accepte une liste de tags ; aucun build supplementaire necessaire).
+      # Naming :
+      #   - Gitea : conserve l'ancien pattern ietm64/<component> pour ne pas
+      #             casser les installs existantes qui ont REGISTRY=git.igmlcreation.fr
+      #             dans leur .env.
+      #   - GHCR  : nouveau pattern igmlcreation/loremind-<component> qui evite
+      #             la collision avec d'autres projets de l'org.
       - name: Build & push ${{ matrix.component }}
         uses: docker/build-push-action@v5
         with:
           context: ./${{ matrix.component }}
           push: true
           tags: |
-            ${{ env.REGISTRY }}/${{ env.REGISTRY_USER }}/${{ matrix.component }}:latest
-            ${{ env.REGISTRY }}/${{ env.REGISTRY_USER }}/${{ matrix.component }}:${{ steps.meta.outputs.version }}
+            ${{ env.GITEA_REGISTRY }}/${{ env.GITEA_REGISTRY_USER }}/${{ matrix.component }}:latest
+            ${{ env.GITEA_REGISTRY }}/${{ env.GITEA_REGISTRY_USER }}/${{ matrix.component }}:${{ steps.meta.outputs.version }}
+            ${{ env.GHCR_REGISTRY }}/${{ env.GHCR_NAMESPACE }}/loremind-${{ matrix.component }}:latest
+            ${{ env.GHCR_REGISTRY }}/${{ env.GHCR_NAMESPACE }}/loremind-${{ matrix.component }}:${{ steps.meta.outputs.version }}

brain/app/infrastructure/ollama_adapter.py

@@ -61,7 +61,16 @@ class OllamaLLMProvider:
         async with httpx.AsyncClient(timeout=self._timeout) as client:
             try:
                 response = await client.post(url, json=payload)
-                response.raise_for_status()
+                if response.status_code >= 400:
+                    body = response.text
+                    try:
+                        err_obj = json.loads(body)
+                        err_msg = err_obj.get("error") or body
+                    except json.JSONDecodeError:
+                        err_msg = body
+                    raise LLMProviderError(
+                        f"Ollama HTTP {response.status_code} : {err_msg.strip()[:500]}"
+                    )
             except httpx.HTTPError as exc:
                 raise LLMProviderError(
                     f"Erreur lors de l'appel à Ollama : {exc}"
@@ -105,7 +114,20 @@ class OllamaLLMProvider:
         async with httpx.AsyncClient(timeout=self._timeout) as client:
             try:
                 async with client.stream("POST", url, json=payload) as response:
-                    response.raise_for_status()
+                    if response.status_code >= 400:
+                        # On lit le body d'erreur pour le remonter a l'utilisateur,
+                        # sinon on ne voit que "500 Internal Server Error" sans
+                        # savoir POURQUOI Ollama refuse (modele introuvable, OOM,
+                        # num_ctx trop grand pour la VRAM, etc.).
+                        body = (await response.aread()).decode("utf-8", errors="replace")
+                        try:
+                            err_obj = json.loads(body)
+                            err_msg = err_obj.get("error") or body
+                        except json.JSONDecodeError:
+                            err_msg = body
+                        raise LLMProviderError(
+                            f"Ollama HTTP {response.status_code} : {err_msg.strip()[:500]}"
+                        )
                     async for line in response.aiter_lines():
                         if not line.strip():
                             continue

core/pom.xml

@@ -14,7 +14,7 @@
 
     <groupId>com.loremind</groupId>
     <artifactId>loremind-core</artifactId>
-    <version>0.6.9</version>
+    <version>0.6.13</version>
     <name>LoreMind Core</name>
     <description>Backend Core - Architecture Hexagonale</description>
 

core/src/main/java/com/loremind/infrastructure/web/controller/SettingsController.java

@@ -21,12 +21,10 @@ import org.springframework.web.servlet.mvc.method.annotation.StreamingResponseBo
 
 import java.io.IOException;
 import java.io.InputStream;
-import java.io.OutputStream;
 import java.net.URI;
 import java.net.http.HttpClient;
 import java.net.http.HttpRequest;
 import java.net.http.HttpResponse;
-import java.nio.charset.StandardCharsets;
 import java.time.Duration;
 import java.util.Map;
 
@@ -46,13 +44,16 @@ public class SettingsController {
 
     private final RestTemplate restTemplate;
     private final String brainBaseUrl;
+    private final String brainInternalSecret;
     private final boolean demoMode;
 
     public SettingsController(RestTemplate restTemplate,
                               @Value("${brain.base-url}") String brainBaseUrl,
+                              @Value("${brain.internal-secret}") String brainInternalSecret,
                               @Value("${app.demo-mode:false}") boolean demoMode) {
         this.restTemplate = restTemplate;
         this.brainBaseUrl = brainBaseUrl;
+        this.brainInternalSecret = brainInternalSecret;
         this.demoMode = demoMode;
     }
 
@@ -90,15 +91,26 @@ public class SettingsController {
     public ResponseEntity<StreamingResponseBody> pullOllamaModel(@RequestBody Map<String, Object> body) {
         guardDemoMode();
         StreamingResponseBody stream = output -> {
+            // Force HTTP/1.1 : le HttpClient JDK essaie HTTP/2 par defaut,
+            // mais uvicorn (Brain) ne supporte que HTTP/1.1 et rejette la
+            // tentative d'upgrade ("Unsupported upgrade request") -> la
+            // requete n'arrive jamais a notre endpoint Python.
             HttpClient http = HttpClient.newBuilder()
+                    .version(HttpClient.Version.HTTP_1_1)
                     .connectTimeout(Duration.ofSeconds(10))
                     .build();
-            HttpRequest req = HttpRequest.newBuilder()
+            // Le RestTemplate auto-injecte X-Internal-Secret via un interceptor,
+            // mais on bypass RestTemplate pour le streaming -> on doit ajouter
+            // l'entete a la main, sinon le Brain repond 401.
+            HttpRequest.Builder reqBuilder = HttpRequest.newBuilder()
                     .uri(URI.create(brainBaseUrl + "/models/ollama/pull"))
                     .timeout(Duration.ofMinutes(60))
                     .header("Content-Type", "application/json")
-                    .POST(HttpRequest.BodyPublishers.ofString(toJson(body)))
-                    .build();
+                    .POST(HttpRequest.BodyPublishers.ofString(toJson(body)));
+            if (brainInternalSecret != null && !brainInternalSecret.isBlank()) {
+                reqBuilder.header("X-Internal-Secret", brainInternalSecret);
+            }
+            HttpRequest req = reqBuilder.build();
             try {
                 HttpResponse<InputStream> resp = http.send(req, HttpResponse.BodyHandlers.ofInputStream());
                 try (InputStream in = resp.body()) {

core/src/main/resources/application.properties

@@ -5,6 +5,12 @@ server.port=8080
 # de WebFlux (utilisé uniquement pour WebClient côté adapter SSE vers le Brain).
 spring.main.web-application-type=servlet
 
+# Pas de timeout sur les requetes async (StreamingResponseBody, SSE).
+# Le defaut Tomcat coupe a 30s, ce qui interrompt le streaming d'un pull
+# de modele Ollama (peut durer des dizaines de minutes pour un GGUF de 10+ Go).
+# -1 = pas de timeout, on s'appuie sur la fermeture cote client ou cote upstream.
+spring.mvc.async.request-timeout=-1
+
 # Configuration de la base de donnees PostgreSQL
 # Valeurs surchargeables via variables d'env (cf. docker-compose.yml).
 # En dev local, creez un fichier .env a la racine de core/ OU definissez les

docker-compose.yml

@@ -60,7 +60,12 @@ services:
       "
 
   core:
-    image: ${REGISTRY:-git.igmlcreation.fr}/ietm64/core:${TAG:-latest}
+    # Defaut : GHCR (registry public, reputation domaine elevee).
+    # Pour les anciennes installs qui pointaient sur Gitea, REGISTRY et
+    # IMAGE_NAMESPACE peuvent etre overrides dans .env :
+    #   REGISTRY=git.igmlcreation.fr
+    #   IMAGE_NAMESPACE=ietm64/  (le slash final est important : voir image: ci-dessous)
+    image: ${REGISTRY:-ghcr.io}/${IMAGE_NAMESPACE:-igmlcreation/loremind-}core:${TAG:-latest}
     container_name: loremind-core
     labels:
       - "com.centurylinklabs.watchtower.enable=true"
@@ -84,8 +89,8 @@ services:
       # Detection des mises a jour : interroge le registry et delegue le pull/restart
       # a Watchtower. Si WATCHTOWER_TOKEN est vide, la feature est desactivee
       # (l'UI masque le badge et le bouton).
-      UPDATE_CHECK_REGISTRY: ${REGISTRY:-git.igmlcreation.fr}
-      UPDATE_CHECK_IMAGES: ietm64/core,ietm64/brain,ietm64/web
+      UPDATE_CHECK_REGISTRY: ${REGISTRY:-ghcr.io}
+      UPDATE_CHECK_IMAGES: ${IMAGE_NAMESPACE:-igmlcreation/loremind-}core,${IMAGE_NAMESPACE:-igmlcreation/loremind-}brain,${IMAGE_NAMESPACE:-igmlcreation/loremind-}web
       UPDATE_CHECK_TAG: ${TAG:-latest}
       WATCHTOWER_URL: http://watchtower:8080
       WATCHTOWER_TOKEN: ${WATCHTOWER_TOKEN:-}
@@ -115,7 +120,7 @@ services:
     restart: unless-stopped
 
   brain:
-    image: ${REGISTRY:-git.igmlcreation.fr}/ietm64/brain:${TAG:-latest}
+    image: ${REGISTRY:-ghcr.io}/${IMAGE_NAMESPACE:-igmlcreation/loremind-}brain:${TAG:-latest}
     container_name: loremind-brain
     labels:
       - "com.centurylinklabs.watchtower.enable=true"
@@ -138,7 +143,7 @@ services:
     restart: unless-stopped
 
   web:
-    image: ${REGISTRY:-git.igmlcreation.fr}/ietm64/web:${TAG:-latest}
+    image: ${REGISTRY:-ghcr.io}/${IMAGE_NAMESPACE:-igmlcreation/loremind-}web:${TAG:-latest}
     container_name: loremind-web
     labels:
       - "com.centurylinklabs.watchtower.enable=true"

installers/README.md

@@ -29,7 +29,7 @@ déclaratif et auditable en quelques lignes.
 ## Linux (Debian / Ubuntu / Fedora / Arch)
 
 ```bash
-curl -fsSL https://git.igmlcreation.fr/ietm64/loremind/raw/branch/main/installers/install.sh | bash
+curl -fsSL https://raw.githubusercontent.com/IGMLcreation/LoreMind/main/installers/install.sh | bash
 ```
 
 Le script :

installers/install.ps1

@@ -40,16 +40,16 @@
   Auteur       : ietm64
   Licence      : AGPL-3.0
   Projet       : LoreMindMJ - assistant pour Maitres de Jeu de JDR
-  Version      : 0.6.9
+  Version      : 0.6.13
 
 .LINK
-  https://git.igmlcreation.fr/ietm64/loremind
+  https://github.com/IGMLcreation/LoreMind
 #>
 
 [CmdletBinding()]
 param(
     [string]$InstallDir = "$env:LOCALAPPDATA\LoreMind",
-    [string]$ComposeUrl = "https://git.igmlcreation.fr/ietm64/loremind/raw/branch/main/docker-compose.yml",
+    [string]$ComposeUrl = "https://raw.githubusercontent.com/IGMLcreation/LoreMind/main/docker-compose.yml",
     [int]$WebPort      = 8081,
     [switch]$NonInteractive
 )
@@ -316,7 +316,8 @@ $composeProfiles = $profilesList -join ','
 
 $envContent = @"
 # Genere par install.ps1 le $(Get-Date -Format 'yyyy-MM-dd HH:mm')
-REGISTRY=git.igmlcreation.fr
+REGISTRY=ghcr.io
+IMAGE_NAMESPACE=igmlcreation/loremind-
 TAG=latest
 
 WEB_PORT=$WebPort

installers/install.sh

@@ -2,12 +2,12 @@
 # ==========================================================================
 # Installeur LoreMindMJ pour Linux (Debian/Ubuntu/Fedora/Arch)
 # Usage :
-#   curl -fsSL https://git.igmlcreation.fr/ietm64/loremind/raw/branch/main/installers/install.sh | bash
+#   curl -fsSL https://raw.githubusercontent.com/IGMLcreation/LoreMind/main/installers/install.sh | bash
 # ==========================================================================
 set -euo pipefail
 
 INSTALL_DIR="${INSTALL_DIR:-$HOME/.local/share/loremind}"
-COMPOSE_URL="${COMPOSE_URL:-https://git.igmlcreation.fr/ietm64/loremind/raw/branch/main/docker-compose.yml}"
+COMPOSE_URL="${COMPOSE_URL:-https://raw.githubusercontent.com/IGMLcreation/LoreMind/main/docker-compose.yml}"
 WEB_PORT="${WEB_PORT:-8081}"
 NON_INTERACTIVE="${NON_INTERACTIVE:-0}"
 
@@ -190,7 +190,8 @@ COMPOSE_PROFILES="$(IFS=,; echo "${PROFILES_ARR[*]}")"
 
 cat > .env <<EOF
 # Genere par install.sh le $(date '+%Y-%m-%d %H:%M')
-REGISTRY=git.igmlcreation.fr
+REGISTRY=ghcr.io
+IMAGE_NAMESPACE=igmlcreation/loremind-
 TAG=latest
 
 WEB_PORT=${WEB_PORT}

installers/secure-host-ollama.ps1

@@ -24,7 +24,7 @@
   faciliter leur identification et suppression ulterieure.
 
 .LINK
-  https://git.igmlcreation.fr/ietm64/loremind
+  https://github.com/IGMLcreation/LoreMind
 #>
 
 [CmdletBinding()]

web/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "loremind-web",
-  "version": "0.6.9",
+  "version": "0.6.13",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "loremind-web",
-      "version": "0.6.9",
+      "version": "0.6.13",
       "dependencies": {
         "@angular/animations": "^17.0.0",
         "@angular/common": "^17.0.0",

web/package.json

@@ -1,6 +1,6 @@
 {
   "name": "loremind-web",
-  "version": "0.6.9",
+  "version": "0.6.13",
   "description": "LoreMind Frontend - Angular",
   "scripts": {
     "ng": "ng",

web/src/app/services/settings.service.ts

@@ -1,4 +1,4 @@
-import { Injectable } from '@angular/core';
+import { Injectable, NgZone } from '@angular/core';
 import { HttpClient } from '@angular/common/http';
 import { Observable } from 'rxjs';
 
@@ -43,7 +43,7 @@ export class SettingsService {
   // suivants en cross-origin (dev Angular sur :4200 -> core sur :8080).
   private readonly authOptions = { withCredentials: true };
 
-  constructor(private http: HttpClient) {}
+  constructor(private http: HttpClient, private zone: NgZone) {}
 
   getSettings(): Observable<AppSettings> {
     return this.http.get<AppSettings>(this.apiUrl, this.authOptions);
@@ -78,8 +78,16 @@ export class SettingsService {
    * bufferise les reponses XHR, ce qui empeche le streaming en temps reel.
    */
   pullOllamaModel(name: string): Observable<OllamaPullEvent> {
+    // fetch() s'execute hors zone Angular -> les emissions du subscriber
+    // doivent etre forcees dans la zone via NgZone.run() pour que le change
+    // detection se declenche et que la barre de progression se mette a jour
+    // en temps reel.
     return new Observable<OllamaPullEvent>((subscriber) => {
       const controller = new AbortController();
+      const emit = (ev: OllamaPullEvent) => this.zone.run(() => subscriber.next(ev));
+      const fail = (e: unknown) => this.zone.run(() => subscriber.error(e));
+      const done = () => this.zone.run(() => subscriber.complete());
+
       (async () => {
         try {
           const response = await fetch(`${this.apiUrl}/models/ollama/pull`, {
@@ -89,16 +97,22 @@ export class SettingsService {
             body: JSON.stringify({ name }),
             signal: controller.signal,
           });
-          if (!response.ok || !response.body) {
-            subscriber.error(new Error(`HTTP ${response.status}`));
+          if (!response.ok) {
+            const text = await response.text();
+            fail(new Error(`HTTP ${response.status} : ${text || 'reponse vide'}`));
+            return;
+          }
+          if (!response.body) {
+            fail(new Error('Reponse sans corps : streaming non supporte par le navigateur'));
             return;
           }
           const reader = response.body.getReader();
           const decoder = new TextDecoder();
           let buffer = '';
+          let eventsReceived = 0;
           while (true) {
-            const { value, done } = await reader.read();
-            if (done) break;
+            const { value, done: streamDone } = await reader.read();
+            if (streamDone) break;
             buffer += decoder.decode(value, { stream: true });
             // Decoupage NDJSON : chaque ligne est un objet JSON complet.
             let nl: number;
@@ -107,15 +121,27 @@ export class SettingsService {
               buffer = buffer.slice(nl + 1);
               if (!line) continue;
               try {
-                subscriber.next(JSON.parse(line) as OllamaPullEvent);
+                emit(JSON.parse(line) as OllamaPullEvent);
+                eventsReceived++;
               } catch {
                 // Ligne non JSON (rare) : on l'ignore.
               }
             }
           }
-          subscriber.complete();
+          // Reste eventuel dans le buffer (derniere ligne sans '\n').
+          if (buffer.trim()) {
+            try {
+              emit(JSON.parse(buffer.trim()) as OllamaPullEvent);
+              eventsReceived++;
+            } catch { /* ignore */ }
+          }
+          if (eventsReceived === 0) {
+            fail(new Error('Aucun evenement recu du serveur. Verifiez les logs du Brain (docker logs loremind-brain).'));
+            return;
+          }
+          done();
         } catch (err) {
-          if ((err as Error).name !== 'AbortError') subscriber.error(err);
+          if ((err as Error).name !== 'AbortError') fail(err);
         }
       })();
       return () => controller.abort();

web/src/app/settings/settings.component.ts

@@ -59,6 +59,10 @@ export class SettingsComponent implements OnInit {
   pullTotal = 0;
   /** Souscription au flux de pull pour pouvoir l'annuler. */
   private pullSubscription: Subscription | null = null;
+  /** True si on a recu un evenement {status:"success"} d'Ollama. Sans ca,
+   *  une fermeture de stream (timeout proxy, perte reseau) ne doit PAS etre
+   *  interpretee comme une reussite. */
+  private pullSucceeded = false;
 
   /** Modele en cours de suppression (nom) pour disabler son bouton. */
   deletingModel: string | null = null;
@@ -293,6 +297,9 @@ export class SettingsComponent implements OnInit {
         if (event.status) this.pullStatus = event.status;
         if (event.completed != null) this.pullCompleted = event.completed;
         if (event.total != null) this.pullTotal = event.total;
+        // Marqueur explicite : Ollama emet "success" en derniere ligne quand
+        // le pull est reellement complet (manifest + layers + verify).
+        if (event.status === 'success') this.pullSucceeded = true;
       },
       error: (err) => {
         this.errorMessage = this.extractError(err, `Echec du telechargement de ${name}.`);
@@ -300,6 +307,14 @@ export class SettingsComponent implements OnInit {
       },
       complete: () => {
         this.pullInProgress = false;
+        if (!this.pullSucceeded) {
+          // Stream ferme sans 'success' final = connexion coupee
+          // (timeout proxy, perte reseau, ...). Le modele est probablement
+          // partiellement telecharge ; Ollama gardera les couches deja DL.
+          this.errorMessage = `Telechargement de ${name} interrompu avant la fin. Relancez pour reprendre.`;
+          this.refreshModels();
+          return;
+        }
         this.successMessage = `Modele ${name} telecharge.`;
         this.refreshModels();
         // Si l'utilisateur n'avait aucun modele, on selectionne celui-ci.
@@ -326,6 +341,7 @@ export class SettingsComponent implements OnInit {
     this.pullStatus = '';
     this.pullCompleted = 0;
     this.pullTotal = 0;
+    this.pullSucceeded = false;
     if (this.pullSubscription) {
       this.pullSubscription.unsubscribe();
       this.pullSubscription = null;