fix: revision completa de rutas Docker, logica SQL y configuracion

Backend Go:
- backend/cmd/server/main.go: ruta wiki_images configurable via WIKI_IMAGES_PATH
- backend/cmd/wiki_worker/main.go: default /opt/rss2 en lugar de /app, leer env
- workers/ctranslator_worker.py: default CT2_MODEL_PATH /opt/rss2 en lugar de /app
- workers/llm_categorizer_worker.py: default LLM_MODEL_PATH /opt/rss2
- workers/{langdetect,simple_translator,translation_scheduler}.py: DB_HOST default 'localhost' en lugar de 'db' (hostname Docker)

SQL / esquema:
- poc/seed.sql: corregir logica de auto-traducciones ES (id LIKE md5() era incorrecto)
- init-db/06-tags.sql: eliminar columna wiki_checked duplicada

Documentacion y configuracion:
- docs/DEPLOY_DEBIAN.md: usar ct2-transformers-converter (lo que usa el worker real)
- deploy/debian/env.example: agregar WIKI_IMAGES_PATH
- deploy/debian/systemd/rss2-cluster.service: agregar HF_HOME faltante
- deploy/debian/install.sh: comparacion numerica correcta de version Go
- scripts/generate_secure_credentials.sh: ruta CT2_MODEL_PATH corregida
- frontend/nginx.conf: advertencia de que es configuracion Docker legacy
- docs/QUICKSTART_LLM.md: nota de deprecacion Docker
- README.md: renombrar backend-go a backend en diagrama

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

README.md

@@ -34,7 +34,7 @@ Internet (RSS/Atom)
                              │             │
                              │         qdrant-worker ──→ Qdrant
                              │
-                    backend-go (API REST :8080)
+                    backend (API REST :8080)
                              │
                     nginx (:8001)
                              │

backend/cmd/server/main.go

@@ -109,7 +109,11 @@ func main() {
 	api := r.Group("/api")
 	{
 		// Serve static images downloaded by wiki_worker
-		api.StaticFS("/wiki-images", gin.Dir("/app/data/wiki_images", false))
+		wikiImagesDir := os.Getenv("WIKI_IMAGES_PATH")
+		if wikiImagesDir == "" {
+			wikiImagesDir = "/opt/rss2/data/wiki_images"
+		}
+		api.StaticFS("/wiki-images", gin.Dir(wikiImagesDir, false))
 
 		api.POST("/auth/login", handlers.Login)
 		api.POST("/auth/register", handlers.Register)

backend/cmd/wiki_worker/main.go

@@ -24,7 +24,7 @@ var (
 	pool          *pgxpool.Pool
 	sleepInterval = 30
 	batchSize     = 50
-	imagesDir     = "/app/data/wiki_images"
+	imagesDir     = "/opt/rss2/data/wiki_images"
 )
 
 type WikiSummary struct {
@@ -210,6 +210,9 @@ func processTag(ctx context.Context, tag Tag) {
 }
 
 func main() {
+	if val := os.Getenv("WIKI_IMAGES_PATH"); val != "" {
+		imagesDir = val
+	}
 	if val := os.Getenv("WIKI_SLEEP"); val != "" {
 		if sleep, err := fmt.Sscanf(val, "%d", &sleepInterval); err == nil && sleep > 0 {
 			sleepInterval = sleep

deploy/debian/env.example

@@ -81,6 +81,7 @@ MAX_FEEDS_PER_URL=5
 
 # --- Wiki Worker ---
 WIKI_SLEEP=10
+WIKI_IMAGES_PATH=/opt/rss2/data/wiki_images
 
 # --- Topics ---
 TOPICS_SLEEP=10

deploy/debian/install.sh

@@ -33,7 +33,17 @@ apt-get install -y --no-install-recommends \
     libpq-dev
 
 # Go (rss-ingestor-go requiere Go 1.25)
-if ! command -v go &>/dev/null || [[ "$(go version | awk '{print $3}' | tr -d 'go')" < "1.25" ]]; then
+_need_go=false
+if ! command -v go &>/dev/null; then
+    _need_go=true
+else
+    _gover=$(go version | awk '{print $3}' | tr -d 'go')
+    IFS='.' read -ra _gv <<< "$_gover"
+    if [[ "${_gv[0]:-0}" -lt 1 ]] || [[ "${_gv[0]:-0}" -eq 1 && "${_gv[1]:-0}" -lt 25 ]]; then
+        _need_go=true
+    fi
+fi
+if [[ "$_need_go" == "true" ]]; then
     info "Instalando Go 1.25..."
     GO_VERSION="1.25.0"
     ARCH=$(dpkg --print-architecture)

deploy/debian/systemd/rss2-cluster.service

@@ -11,6 +11,7 @@ WorkingDirectory=/opt/rss2/src
 EnvironmentFile=/opt/rss2/.env
 Environment=EVENT_DIST_THRESHOLD=0.35
 Environment=EMB_MODEL=sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2
+Environment=HF_HOME=/opt/rss2/hf_cache
 ExecStart=/opt/rss2/venv/bin/python -m workers.cluster_worker
 Restart=always
 RestartSec=10

docs/DEPLOY_DEBIAN.md

@@ -76,16 +76,23 @@ python3 -m venv /opt/rss2/venv
 /opt/rss2/venv/bin/pip install ctranslate2 transformers sentencepiece
 
 # Convertir modelo NLLB-200 a formato CTranslate2 (tarda 10-30 min)
-/opt/rss2/venv/bin/python - <<'EOF'
-from ctranslate2.converters import OpusMTConverter
-converter = OpusMTConverter("facebook/nllb-200-distilled-600M")
-converter.convert("/opt/rss2/models/nllb-ct2", quantization="int8", force=True)
-print("Modelo convertido OK en /opt/rss2/models/nllb-ct2")
-EOF
+mkdir -p /opt/rss2/models/nllb-ct2
+HF_HOME=/opt/rss2/hf_cache \
+/opt/rss2/venv/bin/ct2-transformers-converter \
+    --model facebook/nllb-200-distilled-600M \
+    --output_dir /opt/rss2/models/nllb-ct2 \
+    --quantization int8 \
+    --force
+
+# Verificar que se generó correctamente
+ls /opt/rss2/models/nllb-ct2/model.bin && echo "Modelo OK"
 ```
 
-> El modelo ocupa ~600 MB convertido. Si la descarga de HuggingFace falla, exporta
-> `HF_ENDPOINT=https://huggingface.co` o usa un mirror.
+> El modelo ocupa ~600 MB convertido. Si la descarga de HuggingFace falla:
+> `export HF_ENDPOINT=https://huggingface.co` antes del comando de conversión.
+
+> **Nota:** El worker convierte el modelo automáticamente si no lo encuentra,
+> pero hacerlo a mano evita que el primer arranque tarde 30 minutos.
 
 ### 4. Ejecutar el instalador
 

docs/QUICKSTART_LLM.md

@@ -1,4 +1,8 @@
-# 🚀 Guía Rápida: Sistema LLM Categorizer
+> **NOTA:** Esta guía está basada en la configuración Docker original. En el despliegue
+> Debian nativo, el LLM categorizer se controla con `systemctl start rss2-categorizer`
+> y el modelo se coloca en `/opt/rss2/models/llm` (var `LLM_MODEL_PATH`).
+
+# Guía Rápida: Sistema LLM Categorizer
 
 ## ✅ Estado Actual
 

frontend/nginx.conf

@@ -1,3 +1,8 @@
+# =============================================================================
+# NOTA: Este nginx.conf es la configuración del contenedor Docker del frontend.
+# NO usar para despliegue nativo Debian — usar deploy/debian/nginx.conf
+# =============================================================================
+
 events {
     worker_connections 1024;
 }

init-db/06-tags.sql

@@ -14,4 +14,3 @@ ALTER TABLE tags ADD COLUMN IF NOT EXISTS wiki_summary TEXT;
 ALTER TABLE tags ADD COLUMN IF NOT EXISTS wiki_url TEXT;
 ALTER TABLE tags ADD COLUMN IF NOT EXISTS image_path TEXT;
 ALTER TABLE tags ADD COLUMN IF NOT EXISTS wiki_checked BOOLEAN DEFAULT FALSE;
-ALTER TABLE tags ADD COLUMN IF NOT EXISTS wiki_checked BOOLEAN DEFAULT FALSE;

poc/seed.sql

@@ -189,7 +189,7 @@ ON CONFLICT (noticia_id, lang_to) DO NOTHING;
 -- Traducciones "self" para artículos en español (necesarias para que aparezcan en filtro translated_only)
 INSERT INTO traducciones (noticia_id,lang_from,lang_to,titulo_trad,resumen_trad,status,vectorized)
 SELECT id,'es','es',titulo,resumen,'done',false
-FROM noticias WHERE lang='es' AND id LIKE md5('poc-es-%')
+FROM noticias WHERE lang='es'
 ON CONFLICT (noticia_id, lang_to) DO NOTHING;
 
 -- ---------------------------------------------------------------------------

scripts/generate_secure_credentials.sh

@@ -129,7 +129,7 @@ URL_DISCOVERY_BATCH_SIZE=10
 MAX_FEEDS_PER_URL=5
 
 # CTranslate2 / AI Model Paths
-CT2_MODEL_PATH=/app/models/nllb-ct2
+CT2_MODEL_PATH=/opt/rss2/models/nllb-ct2
 CT2_DEVICE=cuda
 CT2_COMPUTE_TYPE=int8_float16
 UNIVERSAL_MODEL=facebook/nllb-200-distilled-600M

workers/ctranslator_worker.py

@@ -62,7 +62,7 @@ BATCH_SIZE = _env_int("TRANSLATOR_BATCH", 8)
 MAX_SRC_TOKENS = _env_int("MAX_SRC_TOKENS", 512)
 MAX_NEW_TOKENS = _env_int("MAX_NEW_TOKENS", 512)
 
-CT2_MODEL_PATH = _env_str("CT2_MODEL_PATH", "/app/models/nllb-ct2")
+CT2_MODEL_PATH = _env_str("CT2_MODEL_PATH", "/opt/rss2/models/nllb-ct2")
 CT2_DEVICE = _env_str("CT2_DEVICE", "cpu")
 CT2_COMPUTE_TYPE = _env_str("CT2_COMPUTE_TYPE", "int8")
 UNIVERSAL_MODEL = _env_str("UNIVERSAL_MODEL", "facebook/nllb-200-distilled-600M")

workers/langdetect_worker.py

@@ -22,7 +22,7 @@ logging.basicConfig(
 LOG = logging.getLogger(__name__)
 
 DB_CONFIG = {
-    'host': os.getenv('DB_HOST', 'db'),
+    'host': os.getenv('DB_HOST', 'localhost'),
     'port': int(os.getenv('DB_PORT', 5432)),
     'database': os.getenv('DB_NAME', 'rss'),
     'user': os.getenv('DB_USER', 'rss'),

workers/llm_categorizer_worker.py

@@ -41,7 +41,7 @@ DB_CONFIG = {
 # Configuración del worker
 BATCH_SIZE = int(os.environ.get("LLM_BATCH_SIZE", 10))  # 10 noticias por lote
 SLEEP_IDLE = int(os.environ.get("LLM_SLEEP_IDLE", 30))  # segundos
-MODEL_PATH = os.environ.get("LLM_MODEL_PATH", "/app/models/llm")
+MODEL_PATH = os.environ.get("LLM_MODEL_PATH", "/opt/rss2/models/llm")
 GPU_SPLIT = os.environ.get("LLM_GPU_SPLIT", "auto")
 MAX_SEQ_LEN = int(os.environ.get("LLM_MAX_SEQ_LEN", 4096))
 CACHE_MODE = os.environ.get("LLM_CACHE_MODE", "FP16")

workers/simple_translator_worker.py

@@ -22,7 +22,7 @@ logging.basicConfig(
 logger = logging.getLogger(__name__)
 
 DB_CONFIG = {
-    'host': os.getenv('DB_HOST', 'db'),
+    'host': os.getenv('DB_HOST', 'localhost'),
     'port': int(os.getenv('DB_PORT', 5432)),
     'database': os.getenv('DB_NAME', 'rss'),
     'user': os.getenv('DB_USER', 'rss'),

workers/translation_scheduler.py

@@ -22,7 +22,7 @@ logging.basicConfig(
 logger = logging.getLogger(__name__)
 
 DB_CONFIG = {
-    'host': os.getenv('DB_HOST', 'db'),
+    'host': os.getenv('DB_HOST', 'localhost'),
     'port': int(os.getenv('DB_PORT', 5432)),
     'database': os.getenv('DB_NAME', 'rss'),
     'user': os.getenv('DB_USER', 'rss'),