File size: 23,231 Bytes
81121c2 c1a5994 4367adf fa53282 81121c2 fa53282 81121c2 fa53282 81121c2 fa53282 81121c2 fa53282 81121c2 fa53282 81121c2 fa53282 81121c2 fa53282 81121c2 c1a5994 4367adf c1a5994 4367adf c1a5994 81121c2 4367adf 81121c2 4367adf 81121c2 4367adf 81121c2 fa53282 81121c2 fa53282 81121c2 fa53282 81121c2 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 | # -*- coding: utf-8 -*-
"""
建筑机器人绘图系统 · 本地编排服务(多用户并发版)
纯 Python 标准库。默认监听 0.0.0.0:8765,局域网内其他同事可直接访问。
任务进入 FIFO 队列,由单个 Blender 工作线程串行消费,多人同时提交不会互相覆盖。
"""
import json, os, queue, socket, subprocess, sys, threading, time, webbrowser
from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
from urllib.parse import unquote, urlparse, parse_qs
ROOT = os.path.dirname(os.path.abspath(__file__))
sys.path.insert(0, ROOT)
from paths import (STATIC, PLANS, THUMBS, OUTPUT, MEDIA, BLENDER,
BUILD_SCRIPT, BLENDER_DIR_NAME)
import kb
import llm
import chat
import patch
import quickqa
import projects
PORT = int(os.environ.get("ARCHROBOT_PORT", "8765"))
HOST = os.environ.get("ARCHROBOT_HOST", "127.0.0.1")
os.makedirs(OUTPUT, exist_ok=True)
STATUS = {}
STATUS_LOCK = threading.Lock()
TASK_Q = queue.Queue()
RUNNING = {"id": None}
MIME = {".html": "text/html; charset=utf-8", ".js": "text/javascript; charset=utf-8",
".css": "text/css; charset=utf-8", ".png": "image/png", ".jpg": "image/jpeg",
".jpeg": "image/jpeg", ".json": "application/json; charset=utf-8",
".svg": "image/svg+xml", ".mp4": "video/mp4",
".blend": "application/octet-stream", ".ico": "image/x-icon"}
def set_status(pid, d):
with STATUS_LOCK:
STATUS[pid] = d
class RateGate:
"""按 IP 的滑动窗口 + 全局并发上限。
为什么要有:部署到公网后 API Key 在服务端,别人刷对话花的是本账号的钱。
这里不追求精确,只挡住"手抖连点"和"脚本刷"两种最常见的情况。
"""
def __init__(self, per_ip=30, window=600, concurrent=3):
self.per_ip = per_ip # 每个 IP 在窗口内最多问几次
self.window = window # 窗口长度(秒)
self.concurrent = concurrent # 同时进行中的模型调用上限
self._hits = {} # ip -> [时间戳]
self._lock = threading.Lock()
self._inflight = 0
def take(self, ip):
now = time.time()
with self._lock:
if self._inflight >= self.concurrent:
return False, "同时提问的人有点多,请稍等几秒再发", 5
arr = [t for t in self._hits.get(ip, []) if now - t < self.window]
if len(arr) >= self.per_ip:
wait = int(self.window - (now - arr[0])) + 1
self._hits[ip] = arr
return False, "本机提问次数已达上限(%d 次 / %d 分钟)" % (
self.per_ip, self.window // 60), wait
arr.append(now)
self._hits[ip] = arr
self._inflight += 1
return True, "", 0
def done(self):
with self._lock:
self._inflight = max(0, self._inflight - 1)
CHAT_GATE = RateGate()
def get_status(pid):
with STATUS_LOCK:
return dict(STATUS[pid]) if pid in STATUS else None
def cached_state(pid):
"""内存状态优先;服务重启后从磁盘产物恢复,避免"已建模"标记丢失"""
s = get_status(pid)
if s:
return s
if os.path.exists(os.path.join(OUTPUT, pid, "model.blend")):
return {"state": "done", "msg": "已建模(本地已有产物)", "cached": True}
return {"state": "idle"}
def run_blender(plan_id):
out_dir = os.path.join(OUTPUT, plan_id)
json_path = os.path.join(PLANS, plan_id + ".json")
if not os.path.exists(json_path):
set_status(plan_id, {"state": "error", "msg": "图纸不存在"})
return
set_status(plan_id, {"state": "running", "msg": "Blender 引擎启动,解析多视角图纸..."})
cmd = [BLENDER, "--background", "--factory-startup", "--threads", "4", "--python", BUILD_SCRIPT,
"--", "--json", json_path, "--out", out_dir]
# 低优先级:建模跑满 CPU 时,网页和其他同事的请求仍然响应得动
# 注意:creationflags 仅 Windows 可用,Linux(容器/HF Space)必须不传
extra = {"creationflags": 0x00004000 | 0x08000000} if os.name == "nt" else {}
try:
t0 = time.time()
proc = subprocess.run(cmd, capture_output=True, text=True, timeout=1800,
encoding="utf-8", errors="replace", **extra)
log_tail = "\n".join((proc.stdout or "").splitlines()[-8:])
if proc.returncode == 0 and os.path.exists(os.path.join(out_dir, "model.blend")):
set_status(plan_id, {"state": "done",
"msg": "建模完成,耗时 %.0f 秒" % (time.time() - t0),
"log": log_tail})
else:
set_status(plan_id, {"state": "error", "msg": "Blender 返回码 %s" % proc.returncode,
"log": log_tail + "\n" + (proc.stderr or "")[-600:]})
except Exception as e:
set_status(plan_id, {"state": "error", "msg": "调用失败: %s" % e})
def worker_loop():
"""单线程消费队列:Blender 是 CPU 重活,串行最稳;多人同时提交只会排队"""
while True:
pid = TASK_Q.get()
RUNNING["id"] = pid
try:
st = get_status(pid) or {}
if st.get("state") != "done":
run_blender(pid)
except Exception as e:
set_status(pid, {"state": "error", "msg": str(e)})
finally:
RUNNING["id"] = None
TASK_Q.task_done()
class Handler(BaseHTTPRequestHandler):
protocol_version = "HTTP/1.1"
timeout = 30
def log_message(self, *a):
pass
def _client_ip(self):
"""HF / Nginx 之类反代后面,socket 里只有反代 IP,真实来源在 X-Forwarded-For"""
fwd = (self.headers.get("X-Forwarded-For") or "").split(",")[0].strip()
return fwd or (self.client_address[0] if self.client_address else "?")
def _send(self, code, body, ctype="application/json; charset=utf-8"):
if isinstance(body, str):
body = body.encode("utf-8")
self.send_response(code)
self.send_header("Content-Type", ctype)
self.send_header("Content-Length", str(len(body)))
self.send_header("Cache-Control", "no-store, max-age=0")
self.send_header("Access-Control-Allow-Origin", "*")
self.end_headers()
self.wfile.write(body)
def _file(self, path):
if not os.path.isfile(path):
return self._send(404, '{"error":"not found"}')
ext = os.path.splitext(path)[1].lower()
with open(path, "rb") as f:
data = f.read()
self.send_response(200)
self.send_header("Content-Type", MIME.get(ext, "application/octet-stream"))
self.send_header("Content-Length", str(len(data)))
self.send_header("Cache-Control", "no-store, max-age=0")
self.send_header("Access-Control-Allow-Origin", "*")
if ext == ".blend":
self.send_header("Content-Disposition", "attachment; filename=%s_model.blend"
% os.path.basename(os.path.dirname(path)))
self.end_headers()
self.wfile.write(data)
def _safe(self, base, rel):
base = os.path.realpath(base)
p = os.path.realpath(os.path.join(base, rel.replace("/", os.sep)))
try:
return p if os.path.commonpath([p, base]) == base else None
except ValueError:
return None
def do_OPTIONS(self):
self.send_response(204)
self.send_header("Access-Control-Allow-Origin", "*")
self.send_header("Access-Control-Allow-Methods", "GET,POST,OPTIONS")
self.send_header("Access-Control-Allow-Headers", "Content-Type")
self.send_header("Content-Length", "0")
self.end_headers()
def do_GET(self):
try:
self._get()
except Exception as e:
self._send(500, json.dumps({"error": str(e)}, ensure_ascii=False))
def _get(self):
if projects.route(self, 'GET'):
return
url = urlparse(self.path).path
if url in ("/", "/index.html"):
return self._file(os.path.join(STATIC, "workspace.html"))
if url == "/legacy":
return self._file(os.path.join(STATIC, "index.html"))
if url == "/api/health":
nb = 0
idx = os.path.join(PLANS, "index.json")
if os.path.exists(idx):
try:
nb = len(json.load(open(idx, encoding="utf-8")))
except Exception:
nb = 0
thumbs = len(os.listdir(THUMBS)) if os.path.isdir(THUMBS) else 0
return self._send(200, json.dumps(
{"ok": True, "blender": os.path.exists(BLENDER), "buildings": nb,
"plans": thumbs, "media": os.path.isdir(MEDIA),
"three": os.path.exists(os.path.join(STATIC, "lib", "three.min.js")),
"chat": llm.ready(), "engines": len(llm.ENGINES),
"kb": kb.stats()["chunks"],
"running": RUNNING["id"], "queued": TASK_Q.qsize()},
ensure_ascii=False))
if url in ("/api/buildings", "/api/plans"):
idx = os.path.join(PLANS, "index.json")
if not os.path.exists(idx):
return self._send(500, '{"error":"plans/index.json 缺失"}')
return self._file(idx)
if url.startswith("/api/status/"):
pid = url.rsplit("/", 1)[-1]
return self._send(200, json.dumps(cached_state(pid), ensure_ascii=False))
if url.startswith("/api/model/"):
pid = url.rsplit("/", 1)[-1]
info = os.path.join(OUTPUT, pid, "model_info.json")
if not os.path.exists(info):
return self._send(404, '{"error":"no model info"}')
return self._file(info)
if url.startswith("/api/state"):
with STATUS_LOCK:
mem = {k: v.get("state") for k, v in STATUS.items()}
out = {}
for p in os.listdir(OUTPUT):
out[p] = "done" if os.path.exists(os.path.join(OUTPUT, p, "model.blend")) else "idle"
out.update(mem)
return self._send(200, json.dumps(out, ensure_ascii=False))
# ---- 图纸对话 / 知识库 ----
if url == "/api/chat/engines":
return self._send(200, json.dumps({
"engines": llm.engines(), "default": llm.DEFAULT_ENGINE,
"ready": llm.ready(), "token_source": llm.token_source(),
"kb": kb.stats(),
}, ensure_ascii=False))
if url == "/api/chat/suggest":
q = parse_qs(urlparse(self.path).query)
pid = (q.get("id") or [""])[0]
return self._send(200, json.dumps(
{"id": pid, "questions": kb.suggest(pid)}, ensure_ascii=False))
if url == "/api/kb/search":
q = parse_qs(urlparse(self.path).query)
query = (q.get("q") or [""])[0]
pid = (q.get("building") or [None])[0] or None
return self._send(200, json.dumps(
{"query": query, "hits": kb.search(query, pid, 6)}, ensure_ascii=False))
if url == "/api/kb/stats":
return self._send(200, json.dumps(kb.stats(), ensure_ascii=False))
rel = unquote(url).lstrip("/")
# 稳定的 ASCII 别名,避免中文文件名在部分浏览器/代理下取不到
if rel == "media/demo.mp4":
return self._file(os.path.join(MEDIA, "ArchRobot_演示_配音版.mp4"))
if rel.startswith("media/"):
p = self._safe(MEDIA, rel[6:])
return self._file(p) if p else self._send(403, '{"error":"forbidden"}')
if rel.startswith("static/"):
p = self._safe(STATIC, rel[7:])
return self._file(p) if p else self._send(403, '{"error":"forbidden"}')
if rel.startswith("output/"):
p = self._safe(OUTPUT, rel[7:])
return self._file(p) if p else self._send(403, '{"error":"forbidden"}')
if rel.startswith("thumbs/") or rel.startswith("plans/"):
p = self._safe(STATIC, rel)
return self._file(p) if p else self._send(403, '{"error":"forbidden"}')
self._send(404, '{"error":"not found"}')
def do_POST(self):
try:
self._post()
except Exception as e:
self._send(500, json.dumps({"error": str(e)}, ensure_ascii=False))
def _post(self):
if projects.route(self, 'POST'):
return
url = urlparse(self.path).path
try:
ln = int(self.headers.get("Content-Length", 0) or 0)
req = json.loads(self.rfile.read(ln) or b"{}")
except Exception:
return self._send(400, '{"error":"bad request"}')
if url == "/api/generate":
pid = str(req.get("id", ""))
if not pid or not os.path.exists(os.path.join(PLANS, pid + ".json")):
return self._send(400, '{"error":"bad id"}')
if os.path.exists(os.path.join(OUTPUT, pid, "model.blend")):
set_status(pid, {"state": "done", "msg": "已建模(复用本地产物)", "cached": True})
return self._send(200, json.dumps({"state": "done", "cached": True},
ensure_ascii=False))
cur = get_status(pid)
if cur and cur.get("state") in ("queued", "running"):
return self._send(200, json.dumps(cur, ensure_ascii=False))
set_status(pid, {"state": "queued", "msg": "已入队,等待 Blender 空闲..."})
TASK_Q.put(pid)
return self._send(200, json.dumps({"state": "queued", "queued": TASK_Q.qsize()},
ensure_ascii=False))
if url == "/api/open":
pid = str(req.get("id", ""))
blend = os.path.join(OUTPUT, pid, "model.blend")
if not os.path.exists(blend):
return self._send(404, '{"error":"model not found"}')
subprocess.Popen([BLENDER, blend], cwd=os.path.dirname(blend))
return self._send(200, '{"state":"launching"}')
# 对话改参数:大模型只出参数补丁,几何仍由 bpy 确定性执行,改出来是一版新模型
if url == "/api/chat/build":
question = (req.get("question") or "").strip()
pid = (req.get("building") or "").strip()
if not question or not pid:
return self._send(400, '{"error":"缺少问题或楼栋号"}')
if not os.path.exists(os.path.join(PLANS, pid + ".json")):
return self._send(400, '{"error":"图纸不存在"}')
if not llm.ready():
return self._send(200, json.dumps(
{"ok": False, "error": "服务端未找到百炼 API Key",
"hint": "设置环境变量 BAILIAN_API_KEY 后可用"}, ensure_ascii=False))
ok, why, _wait = CHAT_GATE.take(self._client_ip())
if not ok:
return self._send(429, json.dumps({"ok": False, "error": why},
ensure_ascii=False))
try:
with open(os.path.join(PLANS, pid + ".json"), encoding="utf-8") as f:
plan = json.load(f)
got = patch.parse(question, plan, engine=req.get("engine"))
except Exception as e:
return self._send(200, json.dumps({"ok": False, "error": str(e)},
ensure_ascii=False))
finally:
CHAT_GATE.done()
if got["intent"] != "modify" or not got["patch"]:
# 这句话不是改参(比如「洞口开在几个面上」)。以前这里直接报错,
# 客户勾着「改参数并重建」问属性就会被顶回来——所以改成自动回落问答,
# 顺手告诉他一嘴怎么改参。宁可多答一句,不能让客户以为系统不认问题。
try:
answer, meta = chat.ask(question, engine=req.get("engine"),
building=pid)
except Exception as e:
return self._send(200, json.dumps(
{"ok": False, "error": str(e)}, ensure_ascii=False))
return self._send(200, json.dumps(
dict(ok=True, intent="qa", fallback=True, answer=answer,
patch={}, notes=[got.get("explain") or ""],
hint="这句不是改参,已按问答回答;要改参数可以说「层高改成 3.6 米」",
**meta), ensure_ascii=False))
try:
new_id, _new_plan, diffs = patch.prepare(pid, got["patch"])
except Exception as e:
return self._send(200, json.dumps({"ok": False, "error": str(e)},
ensure_ascii=False))
# 同一版号重跑前先清空旧产物,避免界面读到上一版
out_dir = os.path.join(OUTPUT, new_id)
if os.path.isdir(out_dir):
for n in os.listdir(out_dir):
if n.startswith("model.") or n.startswith("render"):
os.remove(os.path.join(out_dir, n))
set_status(new_id, {"state": "queued", "msg": "按对话改参,已入队..."})
TASK_Q.put(new_id)
return self._send(200, json.dumps(
{"ok": True, "task": new_id, "source": pid, "patch": got["patch"],
"diff": diffs, "answer": got["explain"], "notes": got["notes"],
"meta": {k: v for k, v in (got.get("meta") or {}).items()
if k in ("engine", "label", "latency_ms", "total_tokens")}},
ensure_ascii=False))
if url == "/api/chat":
question = (req.get("question") or "").strip()
if not question:
return self._send(400, '{"error":"问题为空"}')
if not llm.ready():
# 没接模型时不要拿「检索片段」冒充答案:先前那条路会让客户问「洞口在几个面上」,
# 屏幕上却弹出两条无关的 FAQ 原文。这里改成规则直答——先正面回答,再说明降级状态。
qa = quickqa.answer(question, req.get("building"))
if qa:
return self._send(200, json.dumps(
{"ok": True, "answer": qa, "degraded": True, "engine": "rule-qa",
"label": "图纸直答(未接模型)", "citations": [],
"note": "当前部署未配置大模型 Key,这条答案是按图纸参数表逐字取数得出的,不是模型生成。"},
ensure_ascii=False))
return self._send(200, json.dumps(
{"ok": False, "degraded": True,
"error": "当前部署未接入大模型,这个问题超出了直答规则能覆盖的范围",
"hint": "已接入的直答范围:层数/层高/总高、墙厚、轮廓与面积、房间分隔、"
"门窗与内墙门洞、屋面坡度出檐山墙、构件总数与材质、图纸视角与约束点、改参口径、导出格式。"
"其余问题需要配置百炼 Key 后才能回答。"},
ensure_ascii=False))
# 公网部署时这道闸必须有:key 在服务端,被刷就是真金白银
ok, why, wait = CHAT_GATE.take(self._client_ip())
if not ok:
return self._send(429, json.dumps(
{"ok": False, "error": why, "hint": "稍后再试;本演示对每个 IP 限流"},
ensure_ascii=False))
try:
answer, meta = chat.ask(question, engine=req.get("engine"),
building=req.get("building") or None,
history=req.get("history") or [])
return self._send(200, json.dumps(dict(ok=True, answer=answer, **meta),
ensure_ascii=False))
except Exception as e:
# 模型侧抖动(限流/超时/配额)时,先用参数表直答保住这次提问,再把原因说清楚
qa = quickqa.answer(question, req.get("building"))
if qa:
return self._send(200, json.dumps(
{"ok": True, "answer": qa, "degraded": True, "engine": "rule-qa",
"label": "图纸直答(模型暂时不可用)", "citations": [],
"note": "大模型这次没调通(%s),这条是按图纸参数表取数直答的。" % str(e)[:120]},
ensure_ascii=False))
return self._send(200, json.dumps(
{"ok": False, "error": str(e)}, ensure_ascii=False))
finally:
CHAT_GATE.done()
self._send(404, '{"error":"not found"}')
def lan_ips():
ips = []
try:
for info in socket.getaddrinfo(socket.gethostname(), None, socket.AF_INET):
ip = info[4][0]
if ip not in ips and not ip.startswith("127."):
ips.append(ip)
except Exception:
pass
return ips
def main():
try:
srv = ThreadingHTTPServer((HOST, PORT), Handler)
except OSError:
print("端口 %d 已被占用:服务可能已在运行。" % PORT)
print("直接访问 http://127.0.0.1:%d/ 即可,无需重复启动。" % PORT)
return
projects.initialize()
srv.daemon_threads = True
srv.request_queue_size = 128
threading.Thread(target=worker_loop, daemon=True).start()
print("=" * 62)
print(" ArchRobot 本地服务已启动(多用户并发)")
print(" 本机 http://127.0.0.1:%d/" % PORT)
if HOST not in ('127.0.0.1','localhost','::1'):
for ip in lan_ips():
print(" 同事 http://%s:%d/" % (ip, PORT))
print(" 健康 http://127.0.0.1:%d/api/health" % PORT)
print(" 提示:不要直接双击 index.html(file:// 无法调用接口)")
if not os.path.isfile(BLENDER):
print("-" * 62)
print(" [!] 未找到 Blender,图纸可浏览但无法建模")
print(" 期望位置:本仓库根目录 %s\\blender.exe" % BLENDER_DIR_NAME)
print(" 或设置环境变量 BLENDER_EXE 指向 blender.exe")
print(" Windows 一键准备:powershell -ExecutionPolicy Bypass -File setup_blender.ps1")
print("=" * 62)
try:
if not os.environ.get("ARCHROBOT_NO_BROWSER"):
webbrowser.open("http://127.0.0.1:%d/" % PORT)
except Exception:
pass
srv.serve_forever()
if __name__ == "__main__":
main()
|