from __future__ import annotations

import os
from concurrent.futures import ThreadPoolExecutor

# Shared executor for CPU-bound analytics (pandas/sklearn) and sync MSSQL I/O
# (pyodbc is inherently synchronous and blocks the event loop if called directly).
#
# Workers are capped at 8 to avoid overwhelming the MSSQL connection pools.
# In K8s: set ANALYTICS_WORKERS to match the pod's CPU limit.
_WORKERS = min(8, int(os.environ.get("ANALYTICS_WORKERS", "0")) or (os.cpu_count() or 2) * 2)

_executor: ThreadPoolExecutor | None = None


def get_executor() -> ThreadPoolExecutor:
    global _executor
    if _executor is None:
        _executor = ThreadPoolExecutor(max_workers=_WORKERS, thread_name_prefix="analytics")
    return _executor


def shutdown_executor() -> None:
    global _executor
    if _executor is not None:
        _executor.shutdown(wait=False)
        _executor = None