Add POST /media/verify - ffprobe-based truncated import detection #59

Merged
sascha merged 3 commits from feature/media-verify-endpoint into main 2026-09-17 15:50:29 +02:00
Showing only changes of commit cadcf8766f - Show all commits

83
app.py
View file

@ -12,7 +12,7 @@ from contextlib import asynccontextmanager
from contextvars import ContextVar
log = logging.getLogger("butler")
VERSION = "2.4.6"
VERSION = "2.4.7"
API_DIR = os.environ.get("API_KEY_DIR", "/data/api")
VAULT_CACHE_DIR = os.environ.get("VAULT_CACHE_DIR", "/data/vault-cache")
@ -1546,6 +1546,87 @@ async def vault_reload(_=Depends(_verify)):
return {"reloaded": True, "items": len(_vault_cache)}
# --- Media file integrity verification ---
MEDIA_VERIFY_PROBES = {
"arrapps": ("bazarrUHD", "sascha"),
"arr-chris": (None, "chris"),
"arr-chris-live": (None, "chris"),
}
class MediaVerifyRequest(BaseModel):
host: str = Field(..., max_length=64)
path: str = Field(..., max_length=1000)
expected_minutes: float | None = Field(None, gt=0, le=1440)
tolerance_pct: float = Field(20.0, gt=0, le=100)
@app.post("/media/verify")
async def media_verify(payload: MediaVerifyRequest, _=Depends(_verify)):
"""Probe a media file's real duration via ffprobe to catch truncated imports.
Runs `ffprobe` inside an existing container (bazarrUHD on arrapps, which already
mounts /data and ships ffmpeg) so no new service is required. Compares the
measured duration against an expected runtime (minutes) supplied by the caller
(e.g. Sonarr/Radarr's runtime field) within tolerance_pct.
"""
if not re.fullmatch(r"[A-Za-z0-9_.-]+", payload.host):
raise HTTPException(400, "Invalid host name")
if payload.host not in MEDIA_VERIFY_PROBES:
raise HTTPException(400, f"No ffprobe container configured for host {payload.host}")
if not re.fullmatch(r"/data/[^\x00]+\.(mkv|mp4|avi|m4v|ts)", payload.path):
raise HTTPException(400, "Path must be an absolute /data media file")
if ".." in payload.path or "'" in payload.path:
raise HTTPException(400, "Path contains unsafe characters")
container, _default_user = MEDIA_VERIFY_PROBES[payload.host]
if not container:
raise HTTPException(400, f"Host {payload.host} has no configured ffprobe container yet")
target = _find_inventory_host(payload.host)
if not target:
raise HTTPException(404, f"Host {payload.host} not found in inventory")
probe_cmd = (
f"docker exec {container} ffprobe -v error "
f"-show_entries format=duration -of default=noprint_wrappers=1:nokey=1 '{payload.path}'"
)
rc, out, err = await asyncio.to_thread(
_ssh, f'{target["user"]}@{target["ip"]}', f"sudo -n {probe_cmd} || {probe_cmd}", 30
)
if rc != 0:
_audit("/media/verify", "POST", 502, f"host={payload.host} path={payload.path}")
raise HTTPException(502, (err or out).strip()[:500] or "ffprobe failed")
raw_duration = out.strip()
try:
duration_seconds = float(raw_duration)
except ValueError:
_audit("/media/verify", "POST", 502, f"host={payload.host} unparsable duration")
raise HTTPException(502, "ffprobe returned no parsable duration; file is likely corrupt")
duration_minutes = duration_seconds / 60
result = {
"host": payload.host,
"path": payload.path,
"duration_seconds": round(duration_seconds, 1),
"duration_minutes": round(duration_minutes, 2),
"expected_minutes": payload.expected_minutes,
"verified": True,
"suspect": False,
}
if payload.expected_minutes:
deviation_pct = abs(duration_minutes - payload.expected_minutes) / payload.expected_minutes * 100
result["deviation_pct"] = round(deviation_pct, 1)
result["suspect"] = deviation_pct > payload.tolerance_pct
_audit(
"/media/verify", "POST", 200,
f"host={payload.host} dur={duration_minutes:.1f}m suspect={result['suspect']}",
)
return result
# --- VPS reverse-proxy and DNS management ---
VPS_SSH = "root@46.225.230.72"