Fix GPU replacement preflight with local reference images

2026-11-12 11:06:23 +08:00
parent c151a056de
commit 3526ed9b31
4 changed files with 139 additions and 18 deletions
@@ -145,13 +145,19 @@ export async function generateVideo(params: GenerateVideoParams): Promise<number
      previewParams.referenceFileUrl = undefined; previewParams.referenceLinkUrl = undefined
    }
    // Validate the new request before touching the previous job.
-    const validatedRequest = getVideoAdapter(config.provider).buildGenerateRequest(config, { ...previewParams, id: 0,
+    getVideoAdapter(config.provider).buildGenerateRequest(config, { ...previewParams, id: 0,
      referenceImageUrls: JSON.stringify(previewParams.referenceImageUrls || []),
      referenceVideoUrls: JSON.stringify(previewParams.referenceVideoUrls || []),
      referenceAudioUrls: JSON.stringify(previewParams.referenceAudioUrls || []) })
    if (params.replaceTaskId && ['minimax', 'localvideo'].includes(config.provider)) {
      const target = new URL(config.baseUrl)
      if (target.hostname === '192.168.110.111' && ['18792', '18794'].includes(target.port)) {
+        // Preflight must see the same inline images as submission. A static/
+        // path is local to this computer and cannot be read by the GPU server.
+        const references = await prepareVideoReferences(previewParams, config.provider)
+        const validatedRequest = getVideoAdapter(config.provider).buildGenerateRequest(config, {
+          ...previewParams, id: 0, ...references,
+        })
        const response = await fetch(`${config.baseUrl.replace(/\/$/, '')}/v2/video_generation/validate`, {
          method: 'POST', headers: { Authorization: `Bearer ${config.apiKey}`, 'Content-Type': 'application/json' },
          body: JSON.stringify(validatedRequest.body), signal: AbortSignal.timeout(35_000) })
@@ -369,26 +375,13 @@ async function processTask(id: number, config: AIConfig) {
      }))
    } else {
      const adapter = getVideoAdapter(config.provider)
-      const resolvedImageUrl = await normalizeVideoReferenceUrl(params.imageUrl)
-      const resolvedFirstFrameUrl = await normalizeVideoReferenceUrl(params.firstFrameUrl)
-      const resolvedLastFrameUrl = await normalizeVideoReferenceUrl(params.lastFrameUrl)
-      const resolvedReferenceImageUrls = await normalizeVideoReferenceUrls(params.referenceImageUrls)
-      // 视频使用公网地址；短音色按厂商能力使用 data URL 或公网地址。
-      const resolvedReferenceVideoUrls = resolvePublicMediaUrls(params.referenceVideoUrls, 'video')
-      const resolvedReferenceAudioUrls = await resolveReferenceAudioUrls(params.referenceAudioUrls, config.provider)
-      const resolvedReferenceFileUrl = resolvePublicMediaUrl(params.referenceFileUrl, 'file')
+      const references = await prepareVideoReferences(params, config.provider)
      ;({ url, method, headers, body } = adapter.buildGenerateRequest(config, {
        id: record.id,
        model: record.model,
        prompt: record.prompt,
        referenceMode: params.referenceMode,
-        imageUrl: resolvedImageUrl,
-        firstFrameUrl: resolvedFirstFrameUrl,
-        lastFrameUrl: resolvedLastFrameUrl,
-        referenceImageUrls: resolvedReferenceImageUrls.length ? JSON.stringify(resolvedReferenceImageUrls) : null,
-        referenceVideoUrls: resolvedReferenceVideoUrls.length ? JSON.stringify(resolvedReferenceVideoUrls) : null,
-        referenceAudioUrls: resolvedReferenceAudioUrls.length ? JSON.stringify(resolvedReferenceAudioUrls) : null,
-        referenceFileUrl: resolvedReferenceFileUrl,
+        ...references,
        referenceLinkUrl: params.referenceLinkUrl,
        generateAudio: params.generateAudio,
        duration: params.duration,
@@ -872,6 +865,21 @@ async function normalizeReferenceImages(refs: string[] | null | undefined): Prom
  return normalized
 }

+async function prepareVideoReferences(params: Pick<GenerateVideoParams, 'imageUrl' | 'firstFrameUrl' | 'lastFrameUrl' |
+  'referenceImageUrls' | 'referenceVideoUrls' | 'referenceAudioUrls' | 'referenceFileUrl'>, provider: string) {
+  const imageUrl = await normalizeVideoReferenceUrl(params.imageUrl)
+  const firstFrameUrl = await normalizeVideoReferenceUrl(params.firstFrameUrl)
+  const lastFrameUrl = await normalizeVideoReferenceUrl(params.lastFrameUrl)
+  const images = await normalizeVideoReferenceUrls(params.referenceImageUrls)
+  const videos = resolvePublicMediaUrls(params.referenceVideoUrls, 'video')
+  const audios = await resolveReferenceAudioUrls(params.referenceAudioUrls, provider)
+  return { imageUrl, firstFrameUrl, lastFrameUrl,
+    referenceImageUrls: images.length ? JSON.stringify(images) : null,
+    referenceVideoUrls: videos.length ? JSON.stringify(videos) : null,
+    referenceAudioUrls: audios.length ? JSON.stringify(audios) : null,
+    referenceFileUrl: resolvePublicMediaUrl(params.referenceFileUrl, 'file') }
+}
+
 async function normalizeVideoReferenceUrl(value: string | null | undefined): Promise<string | null> {
  const raw = String(value || '').trim()
  if (!raw) return null
@@ -883,7 +891,7 @@ async function normalizeVideoReferenceUrl(value: string | null | undefined): Pro
      return readImageAsDataUrl(localPath)
    } catch (err) {
      logTaskWarn('VideoTask', 'reference-read-failed', { path: localPath, error: (err as Error).message })
-      return null
+      throw new Error('参考图读取失败，请重新上传后再生成')
    }
  }
  return raw
@@ -52,6 +52,7 @@ export function mapError(err: unknown, opts?: { fallback?: string }): string {
  // Known local worker condition: preserve the actionable GPU/port hint without
  // exposing arbitrary upstream JSON, URLs or credentials in the task list.
  if (/GPU1\s*\/\s*8189 暂不可用：旧 GPU1 任务或显存尚未释放/.test(msg)) return t('errors.gpu1Busy')
+  if (msg.startsWith('新显卡未就绪，原任务保持运行：')) return '新显卡检查未通过，原任务仍在运行，请检查显卡和模型设置'
  if (SERVER5XX_RE.test(msg)) return t('errors.unavailable')
  if (TIMEOUT_RE.test(msg)) return t('errors.timeout')

@@ -0,0 +1,112 @@
+"""Exercise GPU replacement preflight/cancellation with local images, no real GPU jobs."""
+import base64, json, os, pathlib, re, sqlite3, subprocess, threading, time, urllib.error, urllib.request
+from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
+
+root = pathlib.Path(__file__).resolve().parent.parent
+data = root / 'integration-work' / ('replacement-test-' + time.strftime('%Y%m%d-%H%M%S'))
+data.mkdir(); (data / 'static/images').mkdir(parents=True)
+png = base64.b64decode('iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAAC1HAwCAAAAC0lEQVR42mP8/x8AAwMCAO+aN1kAAAAASUVORK5CYII=')
+(data / 'static/images/reference.png').write_bytes(png)
+stage = root / 'integration-work/stage-4.0.11/app'
+source = (root / 'huobao-source/desktop/build/backend.mjs').read_text('utf-8')
+source = source.replace('const require = __hbCreateRequire(import.meta.url)', 'const require = __hbCreateRequire(' + json.dumps(str(stage / 'resources/app.asar/build/backend.mjs')) + ')', 1)
+source = re.sub(r'^import (\w+) from "(better-sqlite3|sharp)";', r'const \1 = require("\2");', source, flags=re.M)
+needle = 'target.hostname === "192.168.110.111" && ["18792", "18794"].includes(target.port)'
+assert source.count(needle) == 1
+source = source.replace(needle, 'target.hostname === "127.0.0.1" && target.port === "18935"')
+(data / 'backend.mjs').write_text(source, 'utf-8')
+events = []; reject = False; process = None
+
+class Mock(BaseHTTPRequestHandler):
+    def log_message(self, *args): pass
+    def reply(self, status, body):
+        raw = json.dumps(body).encode(); self.send_response(status)
+        self.send_header('Content-Type', 'application/json'); self.end_headers(); self.wfile.write(raw)
+    def do_POST(self):
+        body = json.loads(self.rfile.read(int(self.headers['Content-Length'])))
+        events.append((self.path, body))
+        if self.path.endswith('/validate'):
+            images = [x['image_url']['url'] for x in body['content'] if x['type'] == 'image_url']
+            if reject or not images or any(not x.startswith('data:image/') for x in images):
+                self.reply(400, {'error': {'message': 'Local video requires inline image data'}})
+            else: self.reply(200, {'valid': True})
+        elif self.path.endswith('/cancel'):
+            self.reply(200, {'task': {'id': self.path.split('/')[-2], 'status': 'cancelled'}})
+        else:
+            self.reply(200, {'task_id': 'new-' + str(len(events)), 'routing': {'gpu': body['gpu'], 'port': 8191, 'providerId': 'gpu3', 'legacy': False}})
+    def do_GET(self):
+        self.reply(200, {'task': {'id': self.path.split('/')[-1], 'status': 'running', 'worker_state': 'running'}})
+
+server = ThreadingHTTPServer(('127.0.0.1', 18935), Mock)
+threading.Thread(target=server.serve_forever, daemon=True).start()
+env = {**os.environ, 'ELECTRON_RUN_AS_NODE': '1', 'PORT': '18934', 'HUOBAO_DATA_DIR': str(data),
+       'SQLITE_PATH': str(data / 'huobao.sqlite3'), 'FRONTEND_DIST': str(root / 'huobao-source/frontend/.output/public'),
+       'WORKSPACE_PATH': str(stage / 'resources/workspace-template')}
+log = (data / 'backend.log').open('w', encoding='utf-8')
+def api(method, path, body=None, expected=200):
+    request = urllib.request.Request('http://127.0.0.1:18934/api/v1' + path, method=method,
+        data=None if body is None else json.dumps(body).encode(), headers={'Content-Type': 'application/json'})
+    try:
+        with urllib.request.urlopen(request, timeout=40) as response: status = response.status; result = json.load(response)
+    except urllib.error.HTTPError as error: status = error.code; result = json.load(error)
+    assert status == expected, (path, status, result)
+    return result.get('data', result)
+def sql(statement, args=()):
+    with sqlite3.connect(data / 'huobao.sqlite3') as db:
+        db.row_factory = sqlite3.Row
+        return [dict(row) for row in db.execute(statement, args)]
+def replace(shot, old, **extra):
+    return api('POST', '/tasks', {'type': 'video', 'storyboard_id': shot, 'config_id': 1,
+        'model': 'MiniMax-H3', 'gpu': 'gpu3', 'replace_task_id': old, 'prompt': 'reference fixture',
+        'reference_image_urls': ['static/images/reference.png'], **extra.pop('body', {})}, **extra)
+try:
+    process = subprocess.Popen([str(stage / 'HuobaoDrama.exe'), str(data / 'backend.mjs')], cwd=data,
+        env=env, stdout=log, stderr=log, creationflags=subprocess.CREATE_NO_WINDOW)
+    for _ in range(150):
+        assert process.poll() is None, 'Backend exited'
+        try: api('GET', '/health'); break
+        except OSError: time.sleep(.1)
+    ts = '2026-11-12T00:00:00Z'
+    sql('INSERT INTO dramas(id,title,created_at,updated_at) VALUES(1,?,?,?)', ('fixture',ts,ts))
+    sql('INSERT INTO episodes(id,drama_id,episode_number,title,created_at,updated_at) VALUES(1,1,1,?,?,?)', ('fixture',ts,ts))
+    sql('INSERT INTO ai_service_configs(id,service_type,provider,name,base_url,api_key,model,is_active,created_at,updated_at) VALUES(1,?,?,?,?,?,?,1,?,?)',
+        ('video','minimax','fixture','http://127.0.0.1:18935','test-key',json.dumps(['MiniMax-H3']),ts,ts))
+    for i in range(1,5):
+        sql('INSERT INTO storyboards(id,episode_id,storyboard_number,created_at,updated_at) VALUES(?,1,?,?,?)',(i,i,ts,ts))
+        sql('INSERT INTO sys_task(id,type,storyboard_id,provider,model,status,stage,task_id,config_id,config_base_url,created_at,updated_at) VALUES(?,?,?,?,?,?,?,?,?,?,?,?)',
+            (i,'video',i,'minimax','MiniMax-H3','processing','running','old-'+str(i),1,'http://127.0.0.1:18935',ts,ts))
+    # Real local image bytes must reach preflight before cancelling exactly one old job.
+    result = replace(1,1,expected=201)
+    for _ in range(100):
+        if len(events) >= 3: break
+        time.sleep(.1)
+    assert [p for p,b in events[:3]] == ['/v2/video_generation/validate','/v2/video_generation/old-1/cancel','/v2/video_generation']
+    assert events[0][1]['content'] == events[2][1]['content']
+    assert base64.b64decode(events[0][1]['content'][1]['image_url']['url'].split(',')[1]) == png
+    assert events[0][1]['gpu'] == events[2][1]['gpu'] == 'gpu3'
+    assert sql('SELECT error_code FROM sys_task WHERE id=1')[0]['error_code'] == 'user_cancelled'
+    assert sql('SELECT status FROM sys_task WHERE id=2')[0]['status'] == 'processing'
+    # Missing reference fails before any upstream call, leaving the old job intact.
+    before = len(events)
+    response = replace(2,2,expected=400,body={'reference_image_urls':['static/images/missing.png']})
+    assert '参考图读取失败' in response['message'] and len(events) == before
+    assert sql('SELECT status FROM sys_task WHERE id=2')[0]['status'] == 'processing'
+    # Unready destination also leaves old task intact; no silent duplicate submission.
+    reject = True; before = len(events)
+    replace(3,3,expected=400)
+    assert len(events) == before+1 and events[-1][0].endswith('/validate')
+    assert sql('SELECT status FROM sys_task WHERE id=3')[0]['status'] == 'processing'
+    # First-frame replacements use the same conversion and role.
+    reject = False; before = len(events)
+    replace(4,4,expected=201,body={'reference_image_urls':[], 'first_frame_url':'/static/images/reference.png'})
+    for _ in range(100):
+        if len(events) >= before+3: break
+        time.sleep(.1)
+    image = events[before][1]['content'][1]
+    assert image['role'] == 'first_frame' and image['image_url']['url'].startswith('data:image/')
+    assert events[before][1]['content'] == events[before+2][1]['content']
+    print(json.dumps({'passed':True,'inline_reference_preflight':True,'first_frame':True,'cancel_only_selected_shot':True,
+        'preserve_old_on_missing_image':True,'preserve_old_on_validation_failure':True,'fixture':str(data)}))
+finally:
+    if process and process.poll() is None: process.terminate(); process.wait(timeout=10)
+    server.shutdown(); log.close()
@@ -31,7 +31,7 @@ for directory in ('prompts', 'skills'):
    shutil.copytree(root/'huobao-source/backend/workspace'/directory,
                    stage/'resources/workspace-template'/directory,dirs_exist_ok=True)
 manifest={'upstream_version':'4.0.11','source_commit':'b48e6b2312166e25b1ff141832d2e82e5d63c142',
-          'custom_features':['persistent-data','local-video-pool','multi-gpu-images','vdn-hd-versions','per-model-gpu-selection','fixed-gpu-models','cancel-and-regenerate','per-shot-routing','direct-comfy-panels','confirmed-worker-labels','asset-image-references','durable-local-task-tracking','cross-episode-continuity','tail-frame-dependency-queue','storyboard-timeline-validation'],
+          'custom_features':['persistent-data','local-video-pool','multi-gpu-images','vdn-hd-versions','per-model-gpu-selection','fixed-gpu-models','cancel-and-regenerate','per-shot-routing','direct-comfy-panels','confirmed-worker-labels','asset-image-references','durable-local-task-tracking','cross-episode-continuity','tail-frame-dependency-queue','storyboard-timeline-validation','gpu-replacement-inline-references'],
          'archive_sha256':hashlib.sha256(archive.read_bytes()).hexdigest()}
 (stage/'resources/huobao-local-build.json').write_text(json.dumps(manifest,indent=2))
 print('Staged 4.0.11 with persistent data and VDN HD:',manifest['archive_sha256'])