#!/bin/bash
# Download benchmark data from HF Bucket on startup, then launch the API.
# Bucket is the single source of truth for parquet/registry files.

set -euo pipefail

BUCKET="${HF_BUCKET:-stasaking/blood-brain-benchmark}"
DATA_DIR="/app/data"

mkdir -p "$DATA_DIR"

python -c "
from huggingface_hub import list_bucket_tree, download_bucket_files
import os

bucket = '$BUCKET'
data_dir = '$DATA_DIR'

files = list(list_bucket_tree(bucket))
download_list = []
for f in files:
    if f.path.endswith(('.parquet', '.json')):
        local_path = os.path.join(data_dir, f.path)
        os.makedirs(os.path.dirname(local_path) or data_dir, exist_ok=True)
        download_list.append((f.path, local_path))

print(f'Downloading {len(download_list)} files from bucket {bucket}...')
download_bucket_files(bucket, files=download_list)
for _, lp in download_list:
    print(f'  {lp}: {os.path.getsize(lp)/1e6:.2f} MB')
print('Bucket sync complete.')
"

export BENCHMARK_DATA_DIR="$DATA_DIR"
exec uvicorn main:app --host 0.0.0.0 --port 7860 --workers 1 --timeout-keep-alive 5