179 lines
16 KiB
Python
179 lines
16 KiB
Python
from __future__ import annotations
|
||
|
||
import hashlib, hmac, html, json, os, re, time, urllib.parse, uuid
|
||
from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
|
||
from pathlib import Path
|
||
|
||
from asrpipe import review
|
||
from . import db
|
||
from .config import AGENT_TOKEN, DRAFTS, FAILED, INBOX, MAX_UPLOAD, PUBLIC_URL, RESULTS, STATE, SUPPORTED, TRANSCRIPTS, ensure_dirs
|
||
|
||
LABELS = {"queued":"等待处理","transcribing":"正在转写","refining":"正在整理","review":"等待确认","publishing":"正在定稿","complete":"已完成","failed":"处理失败"}
|
||
|
||
def safe(value, limit):
|
||
return re.sub(r"\s+", " ", re.sub(r"[\x00-\x1f\x7f]+", " ", value or "")).strip()[:limit]
|
||
|
||
def stem(value):
|
||
return re.sub(r"[^\w\u3400-\u9fff.-]+", "-", safe(value, 80)).strip("-.") or "录音"
|
||
|
||
def public(row, detailed=False):
|
||
out = {k: row.get(k) for k in ("id","original_name","title","size","status","detail","source","sha256","attempts","error_code","created_at","updated_at")}
|
||
out["status_label"] = LABELS.get(row["status"], row["status"])
|
||
out["web_url"] = f"{PUBLIC_URL}/jobs/{row['id']}/"
|
||
out["review_url"] = f"{PUBLIC_URL}/jobs/{row['id']}/review/"
|
||
out["artifacts"] = {name:f"/api/v1/jobs/{row['id']}/artifacts/{name}" for name in ("transcript","draft","result") if row.get(name)}
|
||
if detailed:
|
||
out["context"] = row.get("context", "")
|
||
out["events"] = db.events(row["id"])
|
||
return out
|
||
|
||
def pending(job):
|
||
found=[]
|
||
for row in review.pending(1000):
|
||
payload=json.loads(row["payload"] or "{}")
|
||
if ((job.get("transcript") and row["recording"]==job["transcript"])
|
||
or (job.get("draft") and payload.get("draft_path")==job["draft"])):
|
||
found.append(row)
|
||
return found
|
||
|
||
def shell(title, body, script=""):
|
||
return f'''<!doctype html><html lang=zh-CN><head><meta charset=utf-8><meta name=viewport content="width=device-width,initial-scale=1"><title>{html.escape(title)}</title><style>
|
||
:root{{--bg:#f4f6f8;--card:#fff;--ink:#18212b;--muted:#647180;--line:#d9dfe6;--accent:#176b5b;--bad:#b33a2b}}@media(prefers-color-scheme:dark){{:root{{--bg:#11161b;--card:#192128;--ink:#eef3f5;--muted:#a8b3bd;--line:#34414b;--accent:#63c7ae}}}}*{{box-sizing:border-box}}body{{margin:0;background:var(--bg);color:var(--ink);font:16px/1.55 system-ui,sans-serif}}header{{background:var(--card);border-bottom:1px solid var(--line)}}header div,main{{max-width:900px;margin:auto;padding:16px}}header div,.row{{display:flex;justify-content:space-between;gap:12px}}a{{color:var(--accent);text-decoration:none}}nav a{{margin-left:16px}}.card{{background:var(--card);border:1px solid var(--line);border-radius:10px;padding:18px;margin:15px 0}}.job{{padding:13px 0;border-top:1px solid var(--line)}}.job:first-child{{border:0}}.muted{{color:var(--muted);font-size:14px}}.badge{{font-weight:700;color:var(--accent)}}.failed{{color:var(--bad)}}input,textarea,button{{font:inherit}}input,textarea{{width:100%;padding:10px;border:1px solid var(--line);border-radius:7px;background:var(--card);color:var(--ink)}}label{{display:block;margin:12px 0 5px;font-weight:650}}button{{border:0;border-radius:7px;padding:11px 16px;background:var(--accent);color:white;font-weight:700}}button.alt{{background:#68747f}}progress{{width:100%}}pre{{white-space:pre-wrap;overflow-wrap:anywhere;max-height:55vh;overflow:auto}}.timeline{{border-left:2px solid var(--line);padding-left:16px}}.choice{{display:block;width:100%;margin:7px 0}}@media(max-width:520px){{.row{{display:block}}}}</style></head><body><header><div><strong>ASR 录音工作台</strong><nav><a href=/>任务</a><a href=/docs>Agent API</a></nav></div></header><main>{body}</main>{script}</body></html>'''.encode()
|
||
|
||
def home():
|
||
body='''<section class=card><h1>上传录音</h1><p class=muted>也可通过 SSH 上传到 tools-pve:/srv/asr/inbox，完成后自动处理。</p><form id=up><label>录音文件</label><input id=file type=file required><label>标题</label><input id=title maxlength=80><label>已知背景</label><textarea id=context maxlength=2000></textarea><p><button id=send>上传并处理</button></p><progress id=bar max=100 hidden></progress><div id=msg class=muted></div></form></section><section class=card><div class=row><h2>最近任务</h2><span class=muted>自动刷新</span></div><div id=jobs>读取中…</div></section>'''
|
||
script='''<script>const E=s=>String(s??'').replace(/[&<>"']/g,c=>({'&':'&','<':'<','>':'>','"':'"',"'":'''}[c]));async function load(){let d=await(await fetch('/ui/jobs')).json();jobs.innerHTML=d.jobs.length?d.jobs.map(j=>`<div class=job><div class=row><strong>${E(j.title||j.original_name)}</strong><span class="badge ${E(j.status)}">${E(j.status_label)}</span></div><div class=muted>${new Date(j.created_at*1000).toLocaleString()} · ${E(j.source)} · ${E(j.detail)}</div><a href="/jobs/${E(j.id)}/">查看任务</a></div>`).join(''):'暂无任务'}load();setInterval(load,8000);up.onsubmit=x=>{x.preventDefault();let f=file.files[0],q=new XMLHttpRequest();send.disabled=true;bar.hidden=false;msg.textContent='上传中…';q.open('POST','/ui/upload');q.setRequestHeader('X-Filename',encodeURIComponent(f.name));q.setRequestHeader('X-Title',encodeURIComponent(title.value));q.setRequestHeader('X-Context',encodeURIComponent(context.value));q.upload.onprogress=v=>{if(v.lengthComputable)bar.value=v.loaded/v.total*100};q.onload=()=>{send.disabled=false;bar.hidden=true;msg.textContent=q.status===201?'已进入队列':'上传失败';if(q.status===201){x.target.reset();load()}};q.send(f)};</script>'''
|
||
return shell("ASR 录音工作台",body,script)
|
||
|
||
def job_page(job):
|
||
events=''.join(f"<p><strong>{html.escape(LABELS.get(e['status'],e['status']))}</strong><br><span class=muted>{time.strftime('%Y-%m-%d %H:%M:%S',time.localtime(e['created_at']))} · {html.escape(e['detail'])}</span></p>" for e in db.events(job["id"]))
|
||
links=' · '.join(f'<a href="/jobs/{job["id"]}/artifact/{n}">{label}</a>' for n,label in (("transcript","转写"),("draft","草稿"),("result","正式稿")) if job.get(n))
|
||
retry=f'<form method=post action="/jobs/{job["id"]}/retry"><button>重新处理</button></form>' if job["status"]=="failed" else ""
|
||
review_link=f'<p><a href="/jobs/{job["id"]}/review/">继续人工确认</a></p>' if pending(job) else ""
|
||
return shell("任务详情",f"<section class=card><div class=row><h1>{html.escape(job.get('title') or job['original_name'])}</h1><span class='badge {job['status']}'>{html.escape(LABELS.get(job['status'],job['status']))}</span></div><p>{html.escape(job.get('detail',''))}</p><p>{links}</p>{review_link}{retry}</section><section class=card><h2>处理时间线</h2><div class=timeline>{events}</div></section>")
|
||
|
||
def review_page(job):
|
||
rows=pending(job)
|
||
if not rows: return shell("人工确认","<section class=card><h1>当前没有待确认问题</h1><p><a href=../>返回任务</a></p></section>")
|
||
row=rows[0]; payload=json.loads(row["payload"] or "{}"); choices=payload.get("candidates") or []
|
||
buttons=''.join(f'<button class=choice name=answer value="{html.escape(str(v),quote=True)}">{html.escape(str(v))}</button>' for v in choices)+'<button class="choice alt" name=answer value="无法确认，保留原文">无法确认 / 保留原文</button>'
|
||
draft=""; path=Path(str(payload.get("draft_path") or ""))
|
||
try: path.resolve().relative_to(DRAFTS.resolve()); draft=path.read_text(encoding="utf-8")
|
||
except (OSError,ValueError): pass
|
||
return shell("人工确认",f"<section class=card><p class=muted>剩余 {len(rows)} 个问题</p><h1>{html.escape(row['question'])}</h1><p>{html.escape(str(payload.get('sentence') or ''))}</p><details><summary>查看草稿</summary><pre>{html.escape(draft)}</pre></details><form method=post><input type=hidden name=id value={row['id']}>{buttons}<label>自定义回答</label><input name=custom maxlength=120><p><button>提交</button></p></form></section>")
|
||
|
||
class Handler(BaseHTTPRequestHandler):
|
||
server_version="ASRService/1"
|
||
def log_message(self,*_): pass
|
||
def send_data(self,data,status=200,ctype="text/html; charset=utf-8",disposition=""):
|
||
self.send_response(status); self.send_header("Content-Type",ctype); self.send_header("Cache-Control","no-store"); self.send_header("X-Content-Type-Options","nosniff"); self.send_header("Content-Security-Policy","default-src 'self'; style-src 'self' 'unsafe-inline'; script-src 'self' 'unsafe-inline'")
|
||
if disposition:self.send_header("Content-Disposition",disposition)
|
||
self.send_header("Content-Length",str(len(data)));self.end_headers();self.wfile.write(data)
|
||
def json(self,obj,status=200):self.send_data(json.dumps(obj,ensure_ascii=False).encode(),status,"application/json; charset=utf-8")
|
||
def auth(self):
|
||
value=self.headers.get("Authorization","")
|
||
return bool(AGENT_TOKEN and value.startswith("Bearer ") and hmac.compare_digest(value[7:],AGENT_TOKEN))
|
||
def need_auth(self):
|
||
if self.auth():return True
|
||
self.json({"error":"unauthorized"},401);return False
|
||
def header(self,name,limit):return safe(urllib.parse.unquote(self.headers.get(name,"")),limit)
|
||
def upload(self,source,protected):
|
||
if protected and not self.auth():return self.json({"error":"unauthorized"},401)
|
||
try:length=int(self.headers.get("Content-Length","0"))
|
||
except ValueError:length=0
|
||
name=self.header("X-Filename",240);suffix=Path(name).suffix.lower()
|
||
if length<=0 or length>MAX_UPLOAD:return self.json({"error":"invalid size"},413)
|
||
if suffix not in SUPPORTED:return self.json({"error":"unsupported type"},415)
|
||
jid=uuid.uuid4().hex[:24]; title=self.header("X-Title",80); stored=f"{time.strftime('%Y-%m-%d_%H_%M_%S')}-{stem(title or Path(name).stem)}-{jid[:8]}{suffix}"; staging=INBOX/".uploading";staging.mkdir(parents=True,exist_ok=True);part=staging/f"{jid}.part";remaining=length;digest=hashlib.sha256()
|
||
try:
|
||
with part.open("xb") as out:
|
||
while remaining:
|
||
chunk=self.rfile.read(min(1024*1024,remaining))
|
||
if not chunk:raise EOFError
|
||
out.write(chunk);digest.update(chunk);remaining-=len(chunk)
|
||
out.flush();os.fsync(out.fileno())
|
||
job=db.create_job(jid,name,stored,title=title,context=self.header("X-Context",2000),size=length,source=source);db.update_job(jid,"queued","录音已落盘，等待 worker",sha256=digest.hexdigest());part.replace(INBOX/stored);self.json(public(job),201)
|
||
except Exception:part.unlink(missing_ok=True);self.json({"error":"upload failed"},500)
|
||
def artifact(self,job,name,api=False):
|
||
if not job:return self.json({"error":"not found"},404)
|
||
roots={"transcript":TRANSCRIPTS,"draft":DRAFTS,"result":RESULTS};path=Path(job.get(name) or "")
|
||
try:path.resolve().relative_to(roots[name].resolve());data=path.read_bytes()
|
||
except (OSError,ValueError):return self.json({"error":"not found"},404)
|
||
self.send_data(data,200,"text/markdown; charset=utf-8",f'inline; filename="{path.name}"')
|
||
def do_GET(self):
|
||
path=urllib.parse.urlsplit(self.path).path
|
||
if path in ("/health","/healthz"):return self.json({"ok":True,"service":"asr-web"})
|
||
if path=="/readyz":
|
||
try:db.connect().close();ok=(STATE/"worker.heartbeat").exists() and time.time()-(STATE/"worker.heartbeat").stat().st_mtime<90
|
||
except Exception:ok=False
|
||
return self.json({"ready":ok},200 if ok else 503)
|
||
if path=="/":return self.send_data(home())
|
||
if path=="/docs":return self.send_data(shell("Agent API","<section class=card><h1>Agent API</h1><p>使用 Bearer Token。POST /api/v1/jobs 的 body 是原始录音，设置 X-Filename、X-Title、X-Context；GET /api/v1/jobs 与 /api/v1/jobs/{id} 查询任务。</p></section>"))
|
||
if path=="/ui/jobs":return self.json({"jobs":[public(x) for x in db.list_jobs()]})
|
||
m=re.fullmatch(r"/jobs/([0-9a-f]{24})/",path)
|
||
if m:
|
||
job=db.get_job(m.group(1));return self.send_data(job_page(job)) if job else self.send_data(b"Not found",404,"text/plain")
|
||
m=re.fullmatch(r"/jobs/([0-9a-f]{24})/review/",path)
|
||
if m:
|
||
job=db.get_job(m.group(1));return self.send_data(review_page(job)) if job else self.send_data(b"Not found",404,"text/plain")
|
||
m=re.fullmatch(r"/jobs/([0-9a-f]{24})/artifact/(transcript|draft|result)",path)
|
||
if m:return self.artifact(db.get_job(m.group(1)),m.group(2))
|
||
if path=="/api/v1/jobs":
|
||
if not self.need_auth():return
|
||
q=urllib.parse.parse_qs(urllib.parse.urlsplit(self.path).query);return self.json({"jobs":[public(x) for x in db.list_jobs(int(q.get('limit',[30])[0]),int(q.get('offset',[0])[0]),q.get('status',[''])[0])]})
|
||
m=re.fullmatch(r"/api/v1/jobs/([0-9a-f]{24})",path)
|
||
if m:
|
||
if not self.need_auth():return
|
||
job=db.get_job(m.group(1));return self.json(public(job,True)) if job else self.json({"error":"not found"},404)
|
||
m=re.fullmatch(r"/api/v1/jobs/([0-9a-f]{24})/artifacts/(transcript|draft|result)",path)
|
||
if m:
|
||
if not self.need_auth():return
|
||
return self.artifact(db.get_job(m.group(1)),m.group(2),True)
|
||
m=re.fullmatch(r"/api/v1/jobs/([0-9a-f]{24})/reviews",path)
|
||
if m:
|
||
if not self.need_auth():return
|
||
job=db.get_job(m.group(1))
|
||
if not job:return self.json({"error":"not found"},404)
|
||
rows=[]
|
||
for row in pending(job):
|
||
payload=json.loads(row["payload"] or "{}")
|
||
rows.append({"id":row["id"],"question":row["question"],
|
||
"kind":payload.get("kind",""),
|
||
"sentence":payload.get("sentence",""),
|
||
"candidates":payload.get("candidates") or []})
|
||
return self.json({"reviews":rows})
|
||
self.send_data(b"Not found",404,"text/plain")
|
||
def do_POST(self):
|
||
path=urllib.parse.urlsplit(self.path).path
|
||
if path=="/ui/upload":return self.upload("web",False)
|
||
if path=="/api/v1/jobs":return self.upload("agent",True)
|
||
m=re.fullmatch(r"/jobs/([0-9a-f]{24})/review/",path)
|
||
if m:
|
||
job=db.get_job(m.group(1));length=min(int(self.headers.get("Content-Length","0")),4096);form=urllib.parse.parse_qs(self.rfile.read(length).decode());qid=int(form.get("id",["0"])[0]);answer=(form.get("custom",[""])[0].strip() or form.get("answer",[""])[0].strip())
|
||
if not job or qid not in {x["id"] for x in pending(job)} or not answer:return self.send_data(b"Invalid answer",400,"text/plain")
|
||
review.answer(qid,answer[:120]);self.send_response(303);self.send_header("Location",path);self.end_headers();return
|
||
m=re.fullmatch(r"/(?:api/v1/)?jobs/([0-9a-f]{24})/retry",path)
|
||
if m:
|
||
api=path.startswith("/api/")
|
||
if api and not self.need_auth():return
|
||
job=db.get_job(m.group(1));source=next((p for p in FAILED.iterdir() if job and p.name==job["stored_name"]),None)
|
||
if not job or job["status"]!="failed" or not source:return self.json({"error":"cannot retry"},409)
|
||
source.replace(INBOX/source.name);db.update_job(job["id"],"queued","重新排队",error_code="")
|
||
if api:return self.json({"ok":True},202)
|
||
self.send_response(303);self.send_header("Location",f"/jobs/{job['id']}/");self.end_headers();return
|
||
m=re.fullmatch(r"/api/v1/jobs/([0-9a-f]{24})/reviews/(\d+)",path)
|
||
if m:
|
||
if not self.need_auth():return
|
||
job=db.get_job(m.group(1));length=min(int(self.headers.get("Content-Length","0")),4096)
|
||
try:answer=str(json.loads(self.rfile.read(length)).get("answer") or "").strip()
|
||
except (ValueError,AttributeError):answer=""
|
||
qid=int(m.group(2))
|
||
if not job or qid not in {x["id"] for x in pending(job)} or not answer:
|
||
return self.json({"error":"invalid review answer"},400)
|
||
return self.json(review.answer(qid,answer[:120]))
|
||
self.send_data(b"Not found",404,"text/plain")
|
||
|
||
def main():
|
||
ensure_dirs();db.connect().close();server=ThreadingHTTPServer(("0.0.0.0",8788),Handler);server.daemon_threads=True;server.serve_forever()
|
||
if __name__=="__main__":main()
|