Initial release: DictIA v0.8.14-alpha (fork de Speakr, AGPL-3.0)
This commit is contained in:
InnovA AI
177
deployment/docs/TROUBLESHOOTING.md
Normal file
177
deployment/docs/TROUBLESHOOTING.md
Normal file
@@ -0,0 +1,177 @@
|
||||
# Troubleshooting — DictIA
|
||||
|
||||
## WhisperX OOM (Out of Memory)
|
||||
|
||||
**Symptome**: Container `whisperx-asr` crash ou restart en boucle.
|
||||
|
||||
**Cause**: Modele trop gros pour la RAM/VRAM disponible.
|
||||
|
||||
**Solutions**:
|
||||
```bash
|
||||
# Utiliser un modele plus petit dans .env
|
||||
ASR_MODEL=medium # au lieu de large-v3
|
||||
|
||||
# Augmenter la limite memoire (local-cpu)
|
||||
# Editer docker-compose.local-cpu.yml
|
||||
deploy:
|
||||
resources:
|
||||
limits:
|
||||
memory: 24G # au lieu de 18G
|
||||
```
|
||||
|
||||
## Diarisation 403 Forbidden
|
||||
|
||||
**Symptome**: Erreur 403 lors de la transcription avec diarisation.
|
||||
|
||||
**Cause**: Token HuggingFace manquant ou conditions non acceptees.
|
||||
|
||||
**Solution**:
|
||||
1. Creer un token: https://huggingface.co/settings/tokens
|
||||
2. Accepter les conditions: https://huggingface.co/pyannote/speaker-diarization-3.1
|
||||
3. Ajouter dans `.env`:
|
||||
```bash
|
||||
HF_TOKEN=hf_votre_token
|
||||
```
|
||||
4. Redemarrer: `docker compose -f deployment/docker/docker-compose.<profil>.yml restart`
|
||||
|
||||
## GPU non detecte (local-gpu)
|
||||
|
||||
**Symptome**: `nvidia-smi` fonctionne mais Docker ne voit pas le GPU.
|
||||
|
||||
**Solution**:
|
||||
```bash
|
||||
# Installer nvidia-container-toolkit
|
||||
sudo apt install -y nvidia-container-toolkit
|
||||
sudo nvidia-ctk runtime configure --runtime=docker
|
||||
sudo systemctl restart docker
|
||||
|
||||
# Verifier
|
||||
docker run --rm --gpus all nvidia/cuda:12.0-base nvidia-smi
|
||||
```
|
||||
|
||||
## Upload echoue (fichiers volumineux)
|
||||
|
||||
**Symptome**: Upload de gros fichiers (>100MB) echoue.
|
||||
|
||||
**Causes possibles**:
|
||||
- Timeout Nginx/reverse proxy
|
||||
- Limite upload trop basse
|
||||
|
||||
**Solutions**:
|
||||
```bash
|
||||
# Si Nginx: verifier client_max_body_size dans dictia.conf
|
||||
client_max_body_size 500M;
|
||||
|
||||
# Si Tailscale Serve: pas de limite cote Tailscale
|
||||
|
||||
# Timeout gunicorn (dans le Dockerfile, deja a 600s)
|
||||
# Pour des fichiers tres longs, augmenter dans docker-compose:
|
||||
environment:
|
||||
- GUNICORN_TIMEOUT=1200
|
||||
```
|
||||
|
||||
## Container dictia "unhealthy"
|
||||
|
||||
**Symptome**: `docker ps` montre "unhealthy" pour le container dictia.
|
||||
|
||||
**Diagnostic**:
|
||||
```bash
|
||||
# Voir les logs
|
||||
docker logs dictia --tail 50
|
||||
|
||||
# Tester manuellement
|
||||
docker exec dictia python3 -c "import urllib.request; urllib.request.urlopen('http://localhost:8899/health')"
|
||||
```
|
||||
|
||||
**Causes courantes**:
|
||||
- `.env` mal configure (SECRET_KEY manquant)
|
||||
- Base de donnees corrompue (restaurer backup)
|
||||
- Port 8899 deja utilise
|
||||
|
||||
## ASR Proxy: "No GPU available"
|
||||
|
||||
**Symptome**: Transcription echoue avec "No GPU available in any Canadian zone".
|
||||
|
||||
**Causes**:
|
||||
- GCP n'a pas de GPU disponible (capacite epuisee)
|
||||
- Credentials GCP expirees
|
||||
- Budget mensuel atteint
|
||||
|
||||
**Diagnostic**:
|
||||
```bash
|
||||
# Verifier le statut du proxy
|
||||
curl -s http://localhost:9090/health | python3 -m json.tool
|
||||
|
||||
# Verifier les stats (budget)
|
||||
curl -s http://localhost:9090/stats | python3 -m json.tool
|
||||
|
||||
# Voir les logs
|
||||
journalctl -u asr-proxy --since "1 hour ago"
|
||||
```
|
||||
|
||||
**Solutions**:
|
||||
- Attendre (GCP libere des GPUs regulierement)
|
||||
- Le proxy reessaie automatiquement apres un cooldown de 3 minutes
|
||||
- Verifier le dashboard: http://localhost:9090
|
||||
|
||||
## Build Docker lent/echoue
|
||||
|
||||
**Symptome**: `docker build` prend trop de temps ou echoue.
|
||||
|
||||
**Solutions**:
|
||||
```bash
|
||||
# Limiter les ressources si le VPS est petit
|
||||
docker build --memory=2g --cpus=2 -t innova-ai/dictia:latest .
|
||||
|
||||
# Nettoyer le cache Docker si le disque est plein
|
||||
docker builder prune -f
|
||||
docker image prune -f
|
||||
```
|
||||
|
||||
## Base de donnees corrompue
|
||||
|
||||
**Symptome**: Erreur SQLite au demarrage.
|
||||
|
||||
**Solution**:
|
||||
```bash
|
||||
# Restaurer le dernier backup
|
||||
bash deployment/tools/restore.sh backups/dictia-LATEST.tar.gz
|
||||
|
||||
# Ou recreer la base (perd les donnees)
|
||||
rm data/instance/transcriptions.db
|
||||
docker compose -f deployment/docker/docker-compose.<profil>.yml restart
|
||||
```
|
||||
|
||||
## Port 8899 deja utilise
|
||||
|
||||
```bash
|
||||
# Trouver qui utilise le port
|
||||
sudo lsof -i :8899
|
||||
# ou
|
||||
sudo ss -tlnp | grep 8899
|
||||
|
||||
# Arreter le processus ou changer le port dans docker-compose
|
||||
ports:
|
||||
- "8900:8899" # utiliser 8900 a la place
|
||||
```
|
||||
|
||||
## Mise a jour qui casse tout
|
||||
|
||||
```bash
|
||||
# Rollback: revenir au commit precedent
|
||||
cd dictia
|
||||
git log --oneline -5 # trouver le bon commit
|
||||
git checkout <commit-hash>
|
||||
|
||||
# Rebuild et redemarrer
|
||||
docker build -t innova-ai/dictia:latest .
|
||||
docker compose -f deployment/docker/docker-compose.<profil>.yml down
|
||||
docker compose -f deployment/docker/docker-compose.<profil>.yml up -d
|
||||
```
|
||||
|
||||
## Commande de diagnostic rapide
|
||||
|
||||
```bash
|
||||
# Tout verifier d'un coup
|
||||
bash deployment/tools/health-check.sh --json | python3 -m json.tool
|
||||
```
|
||||
Reference in New Issue
Block a user