@remixmate/cli 0.9.27 → 0.9.28

package/dist/manifest.json

@@ -1,7 +1,7 @@
 {
   "schemaVersion": 1,
-  "version": "0.9.27",
-  "generatedAt": "2026-09-13T09:11:30.410Z",
+  "version": "0.9.28",
+  "generatedAt": "2026-09-13T13:40:09.148Z",
   "skills": [
     {
       "id": "export-jianying",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@remixmate/cli",
-  "version": "0.9.27",
+  "version": "0.9.28",
   "description": "AI media generation skills for Claude Code / Codex — 12 skills covering image, video, voice, digital human, web screenshot, web recording, script, template registry, rendering, Jianying export, and video deconstruction.",
   "type": "module",
   "main": "dist/index.js",
@@ -25,6 +25,7 @@
     "test:template-pipeline": "PYTHONDONTWRITEBYTECODE=1 node dist/cli.js exec -- python3 scripts/test-template-pipeline.py",
     "test:props-contract": "PYTHONDONTWRITEBYTECODE=1 python3 scripts/test-props-contract.py",
     "test:narration-speed": "PYTHONDONTWRITEBYTECODE=1 python3 scripts/test-narration-speed.py",
+    "test:render-resume": "PYTHONDONTWRITEBYTECODE=1 python3 scripts/test-render-resume.py",
     "test:render-plan": "PYTHONDONTWRITEBYTECODE=1 python3 scripts/test-render-plan-snapshot.py",
     "test:render-plan:update": "PYTHONDONTWRITEBYTECODE=1 python3 scripts/test-render-plan-snapshot.py --update",
     "test:contracts": "PYTHONDONTWRITEBYTECODE=1 python3 skills/template-registry/scripts/check_contracts.py",

package/skills/render-video/scripts/remote_renderer_client.py

@@ -61,6 +61,19 @@ class RemoteRenderError(RuntimeError):
     """抛出至调用方，由上层写入 render_plan.errors。"""
 
 
+class RemoteRenderTimeout(RemoteRenderError):
+    """轮询超时 —— 任务**没有失败**，只是我们不等了。
+
+    与普通 RemoteRenderError 分开，是因为上层的处置完全相反：
+    真失败 → 可以重新提交一个任务；超时 → 后端还在渲，重新提交等于白烧一遍，
+    必须保留 taskId 让下一次调用续上。
+    """
+
+    def __init__(self, message: str, task_id: str):
+        super().__init__(message)
+        self.task_id = task_id
+
+
 def _build_headers(private_token: str, content_type: str = "application/json", conversation_id: Optional[str] = None) -> dict:
     if not private_token:
         raise RemoteRenderError("PrivToken is not set (PRIV_TOKEN or --priv-token)")
@@ -174,9 +187,10 @@ def poll_render(
     while True:
         elapsed = time.monotonic() - start
         if elapsed > timeout:
-            raise RemoteRenderError(
+            raise RemoteRenderTimeout(
                 f"remote render polling timed out (waited {elapsed:.0f}s, taskId={task_id}); "
-                f"the job may still be running on the backend — query {status_path} later"
+                f"the job may still be running on the backend — query {status_path} later",
+                task_id,
             )
 
         try:

package/skills/render-video/scripts/render_video.py

@@ -1647,6 +1647,23 @@ def render_with_local_cli(render_plan: dict, output_path: str) -> bool:
         return False
 
 
+def _persist_plan_checkpoint(render_plan: dict, job_id: Optional[int], private_token: str) -> None:
+    """把当前 RenderPlan 存个档（尽力而为，失败只记日志）。
+
+    存在的唯一理由：远端 taskId 必须在**渲染结束之前**就落库。脚本可能在任何时刻
+    被外层超时 SIGTERM，那一刻没有任何机会写盘；只在流程末尾 save_plan 的话，
+    taskId 永远存不下来，"续跑"就无从谈起。
+    """
+    if not job_id:
+        return
+    try:
+        import render_job_client  # type: ignore
+        render_job_client.save_plan(job_id, json.dumps(render_plan, ensure_ascii=False), private_token)
+        LogPrint(f"   💾 RenderPlan checkpoint saved (jobId={job_id})", file=sys.stderr)
+    except Exception as exc:  # noqa: BLE001 — 存档失败不该让渲染本身失败
+        LogPrint(f"   ⚠️  RenderPlan checkpoint failed (jobId={job_id}): {exc}", file=sys.stderr)
+
+
 def render_with_remote_api(
     render_plan: dict,
     output_path: str,
@@ -1656,10 +1673,14 @@ def render_with_remote_api(
     poll_interval: float = 5.0,
     upload_title: Optional[str] = None,
     conversation_id: Optional[str] = None,
+    job_id: Optional[int] = None,
 ) -> bool:
     """Submit render to ab-api /tool/renderVideo and poll until completion.
 
     On success, writes upload.fileUrl into render_plan and skips local MP4 download.
+
+    job_id 用于中途存档（见 _persist_plan_checkpoint）：有它才能在被强杀后续跑，
+    没有则退化成旧行为（一次性，杀了就得重渲）。
     """
     import remote_renderer_client
 
@@ -1706,6 +1727,13 @@ def render_with_remote_api(
     # 内置模板即使带 sourceOssKey（builtin-template-registry 回填，仅服务
     # derive_template 派生），渲染仍走 ab-render 预构建的 /render，绝不走 /renderDraft，
     # 否则会用 OSS 上的旧快照 + DraftMainVideo 复刻渲染，与生产 MainVideo 漂移。
+    # 下面这套 payload 构造在"续跑既有任务"时其实用不上（见后面的 resuming 分支），
+    # 所以它的失败不该让续跑也跟着失败 —— 任务还在后端好好渲着。
+    pending_resume = bool(str(render_plan.get("remoteTaskId") or "").strip())
+    # 先给默认值：presign 失败时下面的分支不会走到，续跑分支仍要读 status_path。
+    render_path = "/render"
+    status_path = "/renderStatus"
+
     source_oss_key = None
     template_id = render_plan.get("templateId", "")
     if template_id:
@@ -1719,20 +1747,27 @@ def render_with_remote_api(
 
     if source_oss_key:
         import render_job_client  # type: ignore
+        tarball_url = None
         try:
             src = render_job_client.presign_template_source(template_id, private_token)
             tarball_url = src.get("tarballUrl")
             if not tarball_url:
                 raise RuntimeError(f"presignSource 未返回 tarballUrl: {src}")
         except Exception as exc:
-            LogPrint(f"❌ presignSource failed for private template {template_id}: {exc}", file=sys.stderr)
-            render_plan["status"] = "failed"
-            render_plan["errors"].append({
-                "phase": "render",
-                "message": f"presignSource failed: {exc}",
-                "timestamp": now_iso(),
-            })
-            return False
+            if pending_resume:
+                LogPrint(
+                    f"⚠️  presignSource failed ({exc}) — 但有待续跑的任务，忽略继续",
+                    file=sys.stderr,
+                )
+            else:
+                LogPrint(f"❌ presignSource failed for private template {template_id}: {exc}", file=sys.stderr)
+                render_plan["status"] = "failed"
+                render_plan["errors"].append({
+                    "phase": "render",
+                    "message": f"presignSource failed: {exc}",
+                    "timestamp": now_iso(),
+                })
+                return False
         payload = {
             "tarballUrl": tarball_url,
             "inputProps": input_props,
@@ -1759,8 +1794,10 @@ def render_with_remote_api(
         status_path = "/renderStatus"
 
     total_frames = config.get("totalFrames", 0)
-    LogPrint(f"🎬 Submitting remote render task...", file=sys.stderr)
-    if source_oss_key:
+    if pending_resume:
+        # 续跑：下面那堆"正在提交"的日志会误导人，跳过
+        LogPrint(f"🎬 Remote render task already exists, will resume polling", file=sys.stderr)
+    elif source_oss_key:
         LogPrint(f"   Private template (OSS dynamic bundle): {template_id}", file=sys.stderr)
     else:
         LogPrint(f"   Composition: {composition_id}", file=sys.stderr)
@@ -1770,28 +1807,54 @@ def render_with_remote_api(
     LogPrint(f"   total frames: {total_frames}", file=sys.stderr)
     sys.stderr.flush()
 
-    try:
-        task_id = remote_renderer_client.start_render(payload, private_token=private_token, conversation_id=conversation_id, path=render_path)
-    except Exception as exc:
-        LogPrint(f"❌ remote render submission failed: {exc}", file=sys.stderr)
-        render_plan["status"] = "failed"
-        render_plan["errors"].append({
+    # ─── 续跑既有任务 ────────────────────────────────────────────────────
+    # 上一次调用可能是被外层超时 SIGTERM 掉的（skill 执行有上限），而远端任务还在
+    # ab-render 上跑着。这时重新提交 = 白烧一遍渲染、并且大概率再次撞上同一个上限。
+    # 所以先看 plan 里有没有上次留下的 taskId：有就直接续着轮询。
+    #
+    # 只在"任务不可用"（已失败 / 查不到）时才退回重新提交 —— 轮询超时不算不可用，
+    # 那条路走 RemoteRenderTimeout，保留 taskId 交给下一次调用。
+    task_id = str(render_plan.get("remoteTaskId") or "").strip()
+    if task_id:
+        # 提交时用的状态端点存在 plan 里（内置模板 /renderStatus，私有模板
+        # /renderDraftStatus）。用存的那个，别依赖这次重新推导的结果。
+        status_path = render_plan.get("remoteStatusPath") or status_path
+        LogPrint(f"🔄 resuming existing remote task: taskId={task_id}", file=sys.stderr)
+        render_plan["logs"].append({
             "phase": "render",
-            "message": f"remote start_render failed: {exc}",
+            "message": f"Resuming remote task: {task_id}",
             "timestamp": now_iso(),
         })
-        return False
+    else:
+        try:
+            task_id = remote_renderer_client.start_render(payload, private_token=private_token, conversation_id=conversation_id, path=render_path)
+        except Exception as exc:
+            LogPrint(f"❌ remote render submission failed: {exc}", file=sys.stderr)
+            render_plan["status"] = "failed"
+            render_plan["errors"].append({
+                "phase": "render",
+                "message": f"remote start_render failed: {exc}",
+                "timestamp": now_iso(),
+            })
+            return False
 
-    LogPrint(f"   ✅ task submitted, taskId={task_id}", file=sys.stderr)
-    render_plan["remoteTaskId"] = task_id
-    render_plan["logs"].append({
-        "phase": "render",
-        "message": f"Remote task submitted: {task_id}",
-        "timestamp": now_iso(),
-    })
+        LogPrint(f"   ✅ task submitted, taskId={task_id}", file=sys.stderr)
+        render_plan["remoteTaskId"] = task_id
+        render_plan["remoteStatusPath"] = status_path
+        render_plan["remoteStartedAt"] = time.time()
+        render_plan["logs"].append({
+            "phase": "render",
+            "message": f"Remote task submitted: {task_id}",
+            "timestamp": now_iso(),
+        })
+        # **立刻落库**，不等渲染结束。被 SIGTERM 的那一刻脚本没有任何机会写盘，
+        # 所以 taskId 必须在这里就进 DB，否则下一次调用无从续起（这正是旧代码
+        # 只写内存、`remoteTaskId` 全程没人读的后果）。
+        _persist_plan_checkpoint(render_plan, job_id, private_token)
 
     last_progress = -1.0
-    render_start_time = time.time()
+    # 续跑时用提交时刻算 elapsed，否则 ETA 会按"刚开始"估，一上来就偏小。
+    render_start_time = float(render_plan.get("remoteStartedAt") or time.time())
 
     def _on_progress(status_data: dict):
         nonlocal last_progress
@@ -1834,9 +1897,32 @@ def render_with_remote_api(
             adaptive_interval=True,
             status_path=status_path,
         )
+    except remote_renderer_client.RemoteRenderTimeout as exc:
+        # 等不动了，但任务**没失败**。保留 taskId，让下一次调用接着轮询；
+        # 状态不写 failed（写了模型会当成"渲染失败"而重新走一遍完整流程）。
+        LogPrint(f"⏳ {exc}", file=sys.stderr)
+        render_plan["status"] = "rendering"
+        _persist_plan_checkpoint(render_plan, job_id, private_token)
+        render_plan["logs"].append({
+            "phase": "render",
+            "message": f"poll timeout, task still running: {task_id}",
+            "timestamp": now_iso(),
+        })
+        print(
+            f"\n⏳ 远端渲染仍在进行（taskId={task_id}）。"
+            f"用同一个 job_id 再调一次 render_video 即可续上，不会重新渲染。",
+            flush=True,
+        )
+        return False
     except Exception as exc:
         LogPrint(f"❌ remote render polling failed: {exc}", file=sys.stderr)
         render_plan["status"] = "failed"
+        # 任务确实不可用了（失败 / 查不到）。清掉 taskId，否则下一次调用会一直
+        # 去续一个永远好不了的任务，再也提交不了新的。
+        render_plan.pop("remoteTaskId", None)
+        render_plan.pop("remoteStatusPath", None)
+        render_plan.pop("remoteStartedAt", None)
+        _persist_plan_checkpoint(render_plan, job_id, private_token)
         render_plan["errors"].append({
             "phase": "render",
             "message": f"remote poll_render failed: {exc}",
@@ -2030,8 +2116,11 @@ Examples:
     parser.add_argument(
         "--remote-poll-timeout",
         type=float,
-        default=float(os.environ.get("REMOTION_REMOTE_POLL_TIMEOUT", "1800")),
-        help="Remote-render polling timeout (seconds, default 1800)",
+        default=float(os.environ.get("REMOTION_REMOTE_POLL_TIMEOUT", "2700")),
+        # 实测最长的一支约 30 分钟，加上 ab-render 排队要留余量，所以是 45 分钟。
+        # 这个值必须小于 ab-agent 侧 render_video 的 skill 超时（默认 50 分钟），
+        # 否则外层先 SIGTERM，这里这套"超时但保留 taskId"的续跑逻辑根本轮不到执行。
+        help="Remote-render polling timeout (seconds, default 2700)",
     )
     parser.add_argument(
         "--remote-poll-interval",
@@ -2405,6 +2494,7 @@ Examples:
             poll_interval=args.remote_poll_interval,
             upload_title=upload_title_hint,
             conversation_id=conversation_id,
+            job_id=args.job_id if args.save_job else None,
         )
     else:
         LogPrint(f"💻 Using local render mode", file=sys.stderr)