Initial commit: Nat20 Notes — TTRPG session transcription & summarization

backend/app/diagnostics.py

@@ -0,0 +1,73 @@
+import shutil
+import subprocess
+
+import requests
+
+from . import database as db, config
+
+
+def _check_ffmpeg() -> dict:
+    path = shutil.which("ffmpeg")
+    if not path:
+        return {"name": "ffmpeg", "ok": False, "message": "ffmpeg not found on PATH. This should be baked into the Docker image - if you're seeing this, the image build is broken."}
+    return {"name": "ffmpeg", "ok": True, "message": path}
+
+
+def _check_gpu() -> dict:
+    try:
+        import torch
+        if torch.cuda.is_available():
+            name = torch.cuda.get_device_name(0)
+            free, total = torch.cuda.mem_get_info(0)
+            free_gb, total_gb = free / 1e9, total / 1e9
+            return {"name": "gpu", "ok": True, "message": f"{name} — {free_gb:.1f}GB free of {total_gb:.1f}GB"}
+        return {"name": "gpu", "ok": False, "message": "No CUDA GPU visible. Check that the container was started with --gpus all / GPU passthrough in compose, and that host NVIDIA drivers + Container Toolkit are installed. Transcription will fall back to CPU, which is much slower."}
+    except Exception as e:
+        return {"name": "gpu", "ok": False, "message": f"Could not query GPU: {e}"}
+
+
+def _check_hf_token() -> dict:
+    settings = db.get_settings()
+    token = settings.get("hf_token", "")
+    if not token:
+        return {"name": "huggingface_token", "ok": False, "message": "No HF token set. Diarization (speaker separation) will fail without one. Add it in Settings."}
+    return {"name": "huggingface_token", "ok": True, "message": "Token is set (not validated against HF until first diarization run)."}
+
+
+def _check_llm_backend() -> dict:
+    settings = db.get_settings()
+    if settings.get("llm_mode") == "api":
+        if not settings.get("api_key"):
+            return {"name": "llm_backend", "ok": False, "message": "Hosted API selected but no API key set."}
+        return {"name": "llm_backend", "ok": True, "message": f"Hosted API configured ({settings.get('api_base_url')})"}
+
+    host = settings.get("ollama_host", "http://localhost:11434")
+    model = settings.get("ollama_model", "")
+    try:
+        resp = requests.get(f"{host.rstrip('/')}/api/tags", timeout=5)
+        resp.raise_for_status()
+        models = [m["name"] for m in resp.json().get("models", [])]
+        if model not in models:
+            return {"name": "llm_backend", "ok": False, "message": f"Reached Ollama at {host}, but model '{model}' isn't pulled there. Available: {', '.join(models) or '(none)'}. Run: ollama pull {model}"}
+        return {"name": "llm_backend", "ok": True, "message": f"Ollama reachable at {host}, model '{model}' available"}
+    except requests.exceptions.ConnectionError:
+        return {"name": "llm_backend", "ok": False, "message": f"Could not connect to Ollama at {host}. Check the host/port in Settings, and that Ollama is running and reachable from this container."}
+    except Exception as e:
+        return {"name": "llm_backend", "ok": False, "message": f"Error checking Ollama: {e}"}
+
+
+def _check_disk_space() -> dict:
+    usage = shutil.disk_usage(config.DATA_DIR)
+    free_gb = usage.free / 1e9
+    ok = free_gb > 5
+    return {"name": "disk_space", "ok": ok, "message": f"{free_gb:.1f}GB free in {config.DATA_DIR}" + ("" if ok else " — this is low, recordings and model caches need room")}
+
+
+def run_diagnostics() -> list[dict]:
+    return [
+        _check_ffmpeg(),
+        _check_gpu(),
+        _check_hf_token(),
+        _check_llm_backend(),
+        _check_disk_space(),
+    ]