Files
FishServer/fish_api/app/services/action_watch.py

279 lines
9.4 KiB
Python
Raw Normal View History

from __future__ import annotations
import asyncio
import shutil
from pathlib import Path
from typing import Dict, List, Set, Tuple
from loguru import logger
from app.compat import to_thread
2026-04-10 10:30:01 +08:00
from app.db import (
add_watch_processed,
health_snapshot_deliverable,
load_watch_processed,
save_health_snapshot,
)
2026-05-13 09:19:31 +08:00
from app.logging_config import new_run_id, stage
from app.services import action as action_svc
from app.services.measure import transcode_src_to_h264_dst
2026-05-14 16:49:05 +08:00
from app.services.video_slice import _get_ffmpeg_path
from app.settings import Settings
2026-04-10 10:30:01 +08:00
from app.state import app_state
from app.watch_idle import IdleWatchWarnState, idle_warn_interval_sec, maybe_warn_idle_watch
_ACTION_IDLE_WARN_INTERVAL_SEC = idle_warn_interval_sec(
"FISH_ACTION_WATCH_IDLE_WARN_INTERVAL_SEC"
)
def _state_path(settings: Settings) -> Path:
"""返回旧版 JSON 状态文件路径(仅用于兼容导入 SQLite"""
assert settings.action_watch_dir is not None
return settings.action_watch_dir / ".fishaction_watch_processed.json"
_VIDEO_SUFFIXES = frozenset({".mp4", ".mkv", ".mov"})
def iter_mp4(watch_dir: Path, recursive: bool) -> List[Path]:
"""List video files (.mp4, .mkv, .mov) in *watch_dir*."""
if recursive:
return sorted(
p
for p in watch_dir.rglob("*")
if p.is_file() and p.suffix.lower() in _VIDEO_SUFFIXES
)
return sorted(
p
for p in watch_dir.iterdir()
if p.is_file() and p.suffix.lower() in _VIDEO_SUFFIXES
)
def _public_media_url(settings: Settings, basename: str) -> str:
base = settings.public_base_url.rstrip("/")
return f"{base}/media/{basename}"
def _slice_media_basename(base_name: str, slice_index: int) -> str:
"""生成切片视频的媒体文件名。"""
base = Path(base_name).stem
return f"{base}_slice_{slice_index:03d}.mp4"
def _publish_slice_video(
src: Path, dst: Path, settings: Settings,
) -> str:
"""Transcode and publish a single video slice to media_root.
Returns the public URL on success, empty string on failure.
"""
tmp = dst.with_name(dst.stem + "_tmp.mp4")
tmp.unlink(missing_ok=True)
2026-05-13 09:19:31 +08:00
log = logger.bind(pipeline="action_watch", source=src.name)
try:
2026-05-14 16:49:05 +08:00
log.info(
"[行为监控] 切片 H.264 转码开始 | ffmpeg={} | {} -> {}",
_get_ffmpeg_path(),
src.name,
tmp.name,
)
ok = transcode_src_to_h264_dst(src, tmp)
if ok and tmp.is_file() and tmp.stat().st_size > 0:
tmp.replace(dst)
2026-05-13 09:19:31 +08:00
log.info("[行为监控] 切片已发布为 H.264{} -> {}", src.name, dst.name)
else:
tmp.unlink(missing_ok=True)
shutil.copy2(src, dst)
2026-05-13 09:19:31 +08:00
log.warning(
2026-05-14 16:49:05 +08:00
"[行为监控] 转码失败,已直接拷贝原文件:{} -> {}(详见 measure 管线日志)",
src.name,
dst.name,
)
return _public_media_url(settings, dst.name)
except Exception:
2026-05-13 09:19:31 +08:00
log.exception("[行为监控] 切片发布异常 | {}", src.name)
tmp.unlink(missing_ok=True)
if dst.is_file():
return _public_media_url(settings, dst.name)
return ""
async def _run_inference_and_state(
mp4: Path,
settings: Settings,
processed: Set[str],
state_file: Path,
) -> None:
key = str(mp4.resolve())
if key in processed:
return
2026-05-13 09:19:31 +08:00
rid = new_run_id("action_watch")
log = logger.bind(pipeline="action_watch", run_id=rid, source=mp4.name)
try:
size_mb = mp4.stat().st_size / (1024 * 1024)
except OSError:
size_mb = 0.0
log.info(
"[行为监控] 触发行为识别 | mp4={} | 大小={:.1f} MB", mp4, size_mb,
)
async with app_state.action_lock:
app_state.action_status = "running"
try:
2026-05-13 09:19:31 +08:00
with stage(
f"行为识别 watch 全流程({mp4.name}",
pipeline="action_watch",
step="file_total",
run_id=rid,
source=mp4.name,
metrics={"size_mb": round(size_mb, 2)},
):
slice_files, duration = await to_thread(
action_svc.prepare_action_slices, mp4, settings, run_id=rid,
)
2026-05-13 09:19:31 +08:00
settings.media_root.mkdir(parents=True, exist_ok=True)
base_name = (
settings.biomass_water_video_media_name or "biomass_water_surface.mp4"
)
2026-04-13 17:13:02 +08:00
2026-05-13 09:19:31 +08:00
saved_count = 0
first_snap = None
total_slices = len(slice_files)
for i, slice_file in enumerate(slice_files):
snap = await to_thread(
action_svc.run_single_slice_inference,
slice_file, i, total_slices, duration, mp4.name, settings,
run_id=rid,
)
if first_snap is None:
first_snap = snap
if not health_snapshot_deliverable(snap):
continue
video_url = ""
media_basename = _slice_media_basename(base_name, i)
dst = settings.media_root / media_basename
video_url = await to_thread(
_publish_slice_video, slice_file, dst, settings,
)
slice_key = f"{key}#slice{i}"
save_health_snapshot(
settings, snap, source_path=slice_key, video_url=video_url,
)
saved_count += 1
if saved_count == 0:
log.warning(
"[行为监控] 无可投递结果,跳过写库 | mp4={}", mp4.name,
)
2026-04-13 17:13:02 +08:00
app_state.action_status = "idle"
processed.add(key)
if settings.action_watch_use_state_file:
add_watch_processed(settings, key, "action")
2026-04-13 17:13:02 +08:00
pred = (first_snap.raw_class_en or "").strip() if first_snap else ""
2026-05-13 09:19:31 +08:00
zh = first_snap.behavior_result if first_snap else ""
health = first_snap.health_result if first_snap else ""
log.info(
"[行为监控] 完成 | mp4={} | 类别={} ({}) | 健康={} | 投递切片={}",
mp4.name, zh or "-", pred or "-", health or "-", saved_count,
2026-04-13 17:13:02 +08:00
)
except Exception as e:
2026-05-13 09:19:31 +08:00
log.exception("[行为监控] 处理异常 | mp4={} | {}", mp4, e)
2026-04-10 10:30:01 +08:00
app_state.action_status = "idle"
processed.add(key)
if settings.action_watch_use_state_file:
add_watch_processed(settings, key, "action")
async def watch_tick(
settings: Settings,
processed: Set[str],
stability: Dict[str, Tuple[int, int]],
state_file: Path,
) -> bool:
"""处理一轮目录扫描;若处理了至少一个文件返回 True。"""
assert settings.action_watch_dir is not None
watch_dir = settings.action_watch_dir
did = False
seen_keys: Set[str] = set()
for mp4 in iter_mp4(watch_dir, settings.action_watch_recursive):
key = str(mp4.resolve())
seen_keys.add(key)
if key in processed:
continue
try:
st = mp4.stat()
except OSError:
continue
size = int(st.st_size)
if size <= 0:
stability.pop(key, None)
continue
last = stability.get(key)
if last is None or last[0] != size:
stability[key] = (size, 1)
else:
_, cnt = last
stability[key] = (size, cnt + 1)
_, cnt = stability[key]
if cnt >= settings.action_watch_stable_polls:
2026-04-10 10:30:01 +08:00
await _run_inference_and_state(mp4, settings, processed, state_file)
stability.pop(key, None)
did = True
for k in list(stability.keys()):
if k not in seen_keys:
del stability[k]
return did
async def run_action_watch_loop(settings: Settings) -> None:
assert settings.action_watch_dir is not None
wd = settings.action_watch_dir
if not wd.is_dir():
2026-05-13 09:19:31 +08:00
logger.bind(pipeline="action_watch").warning(
"[行为监控] 启动跳过:路径不是目录 {}", wd
)
return
state_file = _state_path(settings)
processed: Set[str] = (
load_watch_processed(settings, state_file, "action")
if settings.action_watch_use_state_file
else set()
)
stability = {} # type: Dict[str, Tuple[int, int]]
2026-05-13 09:19:31 +08:00
logger.bind(pipeline="action_watch").info(
"[行为监控] 监控目录已启动 | dir={} | poll={}s | stable_polls={} | state={} {}",
wd,
settings.action_watch_poll_interval,
settings.action_watch_stable_polls,
"on" if settings.action_watch_use_state_file else "off",
state_file if settings.action_watch_use_state_file else "",
)
idle_warn_state = IdleWatchWarnState()
while True:
did = await watch_tick(settings, processed, stability, state_file)
maybe_warn_idle_watch(
did_work=did,
log_tag="action-watch",
algo_name="FishAction",
idle_hint="目录内无 .mp4、已全部处理完毕或文件尚未达到稳定轮询次数",
watch_dir=wd,
state=idle_warn_state,
interval_sec=_ACTION_IDLE_WARN_INTERVAL_SEC,
)
await asyncio.sleep(max(settings.action_watch_poll_interval, 0.1))