radar_system/python_app/scripts/matrix_raw_producer.py

"""Raw acquisition producer for matrix-mode radars (LibreVNA multi-device, SN9000)."""

from __future__ import annotations

import argparse
import logging
from contextlib import suppress
from pathlib import Path
import signal
import threading
import time

from python_app.hardware_full.matrix_radar_service import MatrixRadarService, create_matrix_radar_service
from python_app.models.run_config_model import RunConfigModel
from python_app.orchestration.shm import ShmRingWriter
from python_app.storage.npz.serialize import RAW_MAGIC, serialize_trace_collection

logger = logging.getLogger(__name__)

# The producer waits for the matrix radar forever: a device that is absent at boot
# or disappears mid-run must never kill the producer, only make it wait. Reconnect
# uses a capped exponential backoff so a long absence does not busy-spin, and every
# wait is interruptible by SIGINT/SIGTERM (stop_requested) for a prompt clean exit.
_OPEN_RETRY_MIN_S = 1.0
_OPEN_RETRY_MAX_S = 10.0
# Throttle open-failure logging during a long wait so a permanently absent device
# does not flood the process log: log the first failure, then every Nth attempt.
_OPEN_RETRY_LOG_EVERY = 30


def _open_radar_with_retry(
    config: RunConfigModel,
    previous: MatrixRadarService | None,
    stop_requested: threading.Event,
) -> MatrixRadarService | None:
    """Open+configure the matrix radar, retrying forever until success or stop.

    Used for both the initial open and every in-loop reconnect, so a device that is
    absent at boot or disappears mid-run never kills the producer — it just waits.
    Returns the opened service, or ``None`` if a stop was requested before any device
    became available. Backoff is capped and every wait is interruptible by SIGTERM.
    """
    if previous is not None:
        with suppress(Exception):
            previous.close()

    attempt = 0
    delay = _OPEN_RETRY_MIN_S
    while not stop_requested.is_set():
        radar = create_matrix_radar_service(config)
        try:
            radar.open()
            radar.configure(config.radar.sweep)
        except Exception as exc:  # noqa: BLE001 — waiting for the device is the point
            with suppress(Exception):
                radar.close()  # drop any partial open before the next attempt
            attempt += 1
            if attempt == 1 or attempt % _OPEN_RETRY_LOG_EVERY == 0:
                logger.warning(
                    "Matrix radar not available (attempt %d); retrying every up to %.0fs "
                    "until the device is present: %s",
                    attempt,
                    _OPEN_RETRY_MAX_S,
                    exc,
                )
            if stop_requested.wait(delay):
                with suppress(Exception):
                    radar.close()
                return None
            delay = min(delay * 2.0, _OPEN_RETRY_MAX_S)
            continue

        if attempt > 0:
            logger.info("Matrix radar opened after %d attempt(s).", attempt + 1)
        return radar

    return None


def main() -> int:
    """Run producer process until config or signal requests exit."""
    parser = argparse.ArgumentParser(description="Publish matrix-radar raw sweeps to SHM rings")
    parser.add_argument("--config", required=True, type=Path, help="Path to run_config.json")
    args = parser.parse_args()

    logging.basicConfig(level=logging.INFO, format="%(levelname)s %(name)s: %(message)s")
    stop_requested = threading.Event()

    def request_stop(_signum: int, _frame: object) -> None:
        stop_requested.set()

    signal.signal(signal.SIGINT, request_stop)
    signal.signal(signal.SIGTERM, request_stop)

    config = RunConfigModel.load_from_path(args.config)
    config.apply_device_model_constraints()
    if not config.is_matrix_radar:
        raise RuntimeError(
            "matrix_raw_producer requires a matrix-mode radar.model "
            "(librevna_multi or sn9000)"
        )

    raw_writer = ShmRingWriter(
        config.rings.raw.name,
        config.rings.raw.capacity,
        config.rings.raw.slot_size_bytes,
    )
    raw_tap_writer = ShmRingWriter(
        config.rings.raw_tap.name,
        config.rings.raw_tap.capacity,
        config.rings.raw_tap.slot_size_bytes,
    )

    radar: MatrixRadarService | None = None
    try:
        radar = _open_radar_with_retry(config, previous=None, stop_requested=stop_requested)
        if radar is None:
            return 0  # asked to stop before a device became available
        collection_id = 1
        while not stop_requested.is_set():
            collection_start = time.monotonic()
            try:
                collection = radar.acquire_collection(collection_id=collection_id)
            except Exception as exc:  # noqa: BLE001 — reconnect forever, never give up
                logger.warning(
                    "Matrix radar acquisition failed; reconnecting and waiting for the device: %s",
                    exc,
                    exc_info=True,
                )
                radar = _open_radar_with_retry(config, previous=radar, stop_requested=stop_requested)
                if radar is None:
                    break  # stop requested while waiting to reconnect
                continue

            payload = serialize_trace_collection(collection, RAW_MAGIC)
            if not raw_writer.push(payload):
                raise RuntimeError(
                    f"Raw payload size {len(payload)} exceeds ring slot size {raw_writer.slot_size_bytes}"
                )
            if not raw_tap_writer.push(payload):
                raise RuntimeError(
                    f"Raw tap payload size {len(payload)} exceeds ring slot size {raw_tap_writer.slot_size_bytes}"
                )
            if not config.runtime.continuous:
                break
            collection_duration_s = time.monotonic() - collection_start
            if collection_id == 1 or collection_id % 20 == 0 or collection_duration_s > 2.0:
                logger.info(
                    "matrix radar collection %d acquired in %.3f s",
                    collection_id,
                    collection_duration_s,
                )
            collection_id += 1
    finally:
        if radar is not None:
            with suppress(Exception):
                radar.close()
        raw_tap_writer.close()
        raw_writer.close()

    logger.info("matrix radar raw producer stopped")
    return 0


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