#!/bin/bash
#
# Start reranker service from isolated venv (.venv-reranker).
#
set -euo pipefail

PROJECT_ROOT="$(cd "$(dirname "$0")/.." && pwd)"
cd "${PROJECT_ROOT}"

RERANKER_VENV="${RERANKER_VENV:-${PROJECT_ROOT}/.venv-reranker}"
PYTHON_BIN="${RERANKER_VENV}/bin/python"

if [[ ! -x "${PYTHON_BIN}" ]]; then
  echo "ERROR: reranker venv not found: ${RERANKER_VENV}" >&2
  echo "Please run: ./scripts/setup_reranker_venv.sh" >&2
  exit 1
fi

# Load .env if present (without activating main venv)
ENV_FILE="${PROJECT_ROOT}/.env"
if [ -f "${ENV_FILE}" ]; then
  while IFS= read -r line || [ -n "${line}" ]; do
    line="${line%$'\r'}"
    [[ -z "${line//[[:space:]]/}" ]] && continue
    [[ "${line}" =~ ^[[:space:]]*# ]] && continue
    [[ "${line}" != *=* ]] && continue

    key="${line%%=*}"
    value="${line#*=}"
    key="${key#"${key%%[![:space:]]*}"}"
    key="${key%"${key##*[![:space:]]}"}"
    value="${value#"${value%%[![:space:]]*}"}"

    if [[ ${#value} -ge 2 ]]; then
      first="${value:0:1}"
      last="${value: -1}"
      if [[ ("${first}" == '"' && "${last}" == '"') || ("${first}" == "'" && "${last}" == "'") ]]; then
        value="${value:1:${#value}-2}"
      fi
    fi

    export "${key}=${value}"
  done < "${ENV_FILE}"
fi

RERANKER_HOST="${RERANKER_HOST:-0.0.0.0}"
RERANKER_PORT="${RERANKER_PORT:-6007}"
RERANK_BACKEND=$("${PYTHON_BIN}" -c "from config.services_config import get_rerank_backend_config; print(get_rerank_backend_config()[0])")

# Keep vLLM/triton/torch caches out of system disk.
RERANKER_RUNTIME_DIR="${RERANKER_RUNTIME_DIR:-${PROJECT_ROOT}/.runtime/reranker}"
mkdir -p "${RERANKER_RUNTIME_DIR}/home" \
         "${RERANKER_RUNTIME_DIR}/cache" \
         "${RERANKER_RUNTIME_DIR}/config" \
         "${RERANKER_RUNTIME_DIR}/triton" \
         "${RERANKER_RUNTIME_DIR}/torch_compile" \
         "${RERANKER_RUNTIME_DIR}/tmp"
export HOME="${RERANKER_RUNTIME_DIR}/home"
export XDG_CACHE_HOME="${RERANKER_RUNTIME_DIR}/cache"
export XDG_CONFIG_HOME="${RERANKER_RUNTIME_DIR}/config"
export TRITON_CACHE_DIR="${RERANKER_RUNTIME_DIR}/triton"
export TORCHINDUCTOR_CACHE_DIR="${RERANKER_RUNTIME_DIR}/torch_compile"
export TMPDIR="${RERANKER_RUNTIME_DIR}/tmp"
export VLLM_NO_USAGE_STATS="${VLLM_NO_USAGE_STATS:-1}"
export PATH="${RERANKER_VENV}/bin:${PATH}"

if [[ "${RERANK_BACKEND}" == "qwen3_vllm" ]]; then
  if ! command -v nvidia-smi >/dev/null 2>&1 || ! nvidia-smi >/dev/null 2>&1; then
    echo "ERROR: qwen3_vllm backend requires NVIDIA GPU, but nvidia-smi is unavailable." >&2
    exit 1
  fi
  if ! "${PYTHON_BIN}" - <<'PY'
try:
    import vllm  # noqa: F401
    import torch
    if not torch.cuda.is_available():
        raise SystemExit(1)
except Exception:
    raise SystemExit(1)
PY
  then
    echo "ERROR: qwen3_vllm backend requires vllm + CUDA runtime in ${RERANKER_VENV}." >&2
    echo "Please run: ./scripts/setup_reranker_venv.sh and verify CUDA is available." >&2
    exit 1
  fi
fi

echo "========================================"
echo "Starting Reranker Service"
echo "========================================"
echo "Python: ${PYTHON_BIN}"
echo "Host: ${RERANKER_HOST}"
echo "Port: ${RERANKER_PORT}"
echo "Backend: ${RERANK_BACKEND}"
echo "Runtime dir: ${RERANKER_RUNTIME_DIR}"
echo

exec "${PYTHON_BIN}" -m uvicorn reranker.server:app \
  --host "${RERANKER_HOST}" \
  --port "${RERANKER_PORT}" \
  --workers 1