#!/usr/bin/env python3
"""
Inicia o servidor TensorBoard sobre o diretorio de logs do treino (Hugging Face Trainer).

Uso tipico apos treino local (o script escolhe logs/ ou results/ com eventos):
  python scripts/launch_tensorboard.py

Ou fixar o diretorio:
  python scripts/launch_tensorboard.py --logdir ./results

Equivalente a: tensorboard --logdir=... --host 127.0.0.1 --port 6006
"""

from __future__ import annotations

import argparse
import socket
import subprocess
import sys
import threading
import time
from pathlib import Path


def _has_tfevents(root: Path) -> bool:
    if not root.is_dir():
        return False
    for path in root.rglob("*"):
        if path.is_file() and path.name.startswith("events.out.tfevents"):
            return True
    return False


def _pick_logdir(cwd: Path) -> Path:
    """Prefere ./logs ou ./results quando contem ficheiros events.out.tfevents*."""
    candidates = [cwd / "logs", cwd / "results"]
    for directory in candidates:
        if directory.is_dir() and _has_tfevents(directory):
            return directory
    for directory in candidates:
        if directory.is_dir():
            return directory
    return cwd / "results"


def _can_bind(host: str, port: int) -> bool:
    bind_host = "127.0.0.1" if host in ("127.0.0.1", "localhost") else host
    with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
        s.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
        try:
            s.bind((bind_host, port))
            return True
        except OSError:
            return False


def _resolve_port(host: str, preferred: int, max_attempts: int = 10) -> int:
    for offset in range(max_attempts):
        port = preferred + offset
        if _can_bind(host, port):
            return port
    return preferred


def _tensorboard_module_cmd(logdir: Path, host: str, port: int, reload_interval: int) -> list[str]:
    return [
        sys.executable,
        "-m",
        "tensorboard.main",
        "--logdir",
        str(logdir.resolve()),
        "--host",
        host,
        "--port",
        str(port),
        "--reload_interval",
        str(reload_interval),
    ]


def _stderr_bind_failure(text: str) -> bool:
    low = text.lower()
    return "could not bind" in low or "already in use" in low


def _drain_stderr(stream: object) -> None:
    try:
        for line in stream:  # type: ignore[union-attr]
            sys.stderr.write(line)
    except Exception:
        pass


def _run_tensorboard_process(
    cmd: list[str],
    host: str,
    port: int,
    quick_fail_seconds: float = 1.25,
) -> tuple[int, bool]:
    """
    Devolve (codigo_saida, tentar_proxima_porta). tentar_proxima_porta e True quando a
    falha parece ser bind/porta (corrida com _can_bind ou outro processo).
    """
    proc = subprocess.Popen(
        cmd,
        stdout=subprocess.DEVNULL,
        stderr=subprocess.PIPE,
        text=True,
    )
    assert proc.stderr is not None
    deadline = time.monotonic() + quick_fail_seconds
    while time.monotonic() < deadline:
        if proc.poll() is not None:
            err = proc.stderr.read()
            code = proc.returncode if proc.returncode is not None else 1
            retry_port = code != 0 and _stderr_bind_failure(err)
            if err:
                sys.stderr.write(err)
            return code, retry_port
        time.sleep(0.05)

    threading.Thread(target=_drain_stderr, args=(proc.stderr,), daemon=True).start()
    print(f"Abre no browser: http://{host}:{port}/")
    try:
        return proc.wait(), False
    except KeyboardInterrupt:
        proc.terminate()
        try:
            proc.wait(timeout=8)
        except subprocess.TimeoutExpired:
            proc.kill()
        return 130, False


def _start_tensorboard_with_port_fallback(
    logdir: Path,
    host: str,
    preferred_port: int,
    reload_interval: int,
    max_attempts: int = 10,
) -> int:
    preferred = preferred_port
    if not _can_bind(host, preferred):
        alt = _resolve_port(host, preferred)
        if alt != preferred:
            print(
                f"Porta {preferred} ocupada (ex.: outro TensorBoard ou Docker). A usar {alt}.",
                file=sys.stderr,
            )
            preferred = alt

    for offset in range(max_attempts):
        port = preferred + offset
        if offset > 0 and not _can_bind(host, port):
            continue
        cmd = _tensorboard_module_cmd(
            logdir=logdir,
            host=host,
            port=port,
            reload_interval=reload_interval,
        )
        print(f"Iniciando TensorBoard: {' '.join(cmd)}")
        rc, retry_port = _run_tensorboard_process(cmd, host=host, port=port)
        if rc == 0:
            return 0
        if retry_port and offset + 1 < max_attempts:
            print(
                f"Porta {port} indisponivel ao iniciar. A tentar {port + 1}...",
                file=sys.stderr,
            )
            continue
        return rc

    print("Erro: nao foi possivel abrir o TensorBoard em nenhuma porta tentada.", file=sys.stderr)
    return 1


def main() -> int:
    parser = argparse.ArgumentParser(
        description="Abre TensorBoard nos logs gerados pelo train.py (report_to=tensorboard).",
    )
    parser.add_argument(
        "--logdir",
        type=Path,
        default=None,
        help=(
            "Diretorio com events.out.tfevents.* "
            "(default: auto procura em ./logs e ./results)."
        ),
    )
    parser.add_argument(
        "--host",
        default="127.0.0.1",
        help="Interface de escuta (default: 127.0.0.1).",
    )
    parser.add_argument(
        "--port",
        type=int,
        default=6006,
        help="Porta HTTP preferida (default: 6006). Se estiver ocupada, usa a seguinte livre.",
    )
    parser.add_argument(
        "--reload_interval",
        type=int,
        default=5,
        help="Segundos entre recargas ao detetar novos eventos (default: 5).",
    )
    args = parser.parse_args()

    cwd = Path.cwd()
    logdir = args.logdir if args.logdir is not None else _pick_logdir(cwd)
    logdir = logdir.resolve()

    if args.logdir is None:
        print(f"TensorBoard --logdir (auto): {logdir}", file=sys.stderr)

    if not logdir.exists():
        print(f"Erro: diretorio nao existe: {logdir}", file=sys.stderr)
        return 1

    if not _has_tfevents(logdir):
        print(
            "Aviso: nenhum ficheiro events.out.tfevents* encontrado sob este diretorio. "
            "O TensorBoard pode ficar sem escalares. Se descarregou os logs do servidor, "
            "use --logdir com a pasta onde estao os eventos (ex.: ./logs).",
            file=sys.stderr,
        )

    try:
        import tensorboard  # noqa: F401
    except ImportError:
        print(
            "Erro: pacote 'tensorboard' nao instalado. Execute: pip install tensorboard",
            file=sys.stderr,
        )
        return 1

    return _start_tensorboard_with_port_fallback(
        logdir=logdir,
        host=args.host,
        preferred_port=args.port,
        reload_interval=args.reload_interval,
    )


if __name__ == "__main__":
    raise SystemExit(main())