feat: 完善模型平台相关功能

runtime/Dockerfile

@@ -1,6 +1,9 @@
 FROM python:3.12-slim-bookworm
 
-ENV PYTHONDONTWRITEBYTECODE=1 PYTHONUNBUFFERED=1 PYTHONPATH=/app
+ENV PYTHONDONTWRITEBYTECODE=1 \
+    PYTHONUNBUFFERED=1 \
+    PYTHONPATH=/app \
+    PATH="/app/.venv/bin:${PATH}"
 WORKDIR /app
 # 从官方 Rclone 镜像直接复制 rclone 二进制文件（高效、稳定）
 COPY --from=rclone/rclone:latest /usr/local/bin/rclone /usr/local/bin/rclone

runtime/src/runtime/mount.py

@@ -21,18 +21,34 @@ REMOTE_BUCKET = settings.remote_bucket
 RCLONE_PROCESS: subprocess.Popen | None = None
 
 
-def is_mountpoint(path: Path) -> bool:
-    result = subprocess.run(
-        ["mountpoint", "-q", str(path)],
-        stdout=subprocess.DEVNULL,
-        stderr=subprocess.DEVNULL,
-    )
-    return result.returncode == 0
+def is_rclone_mount(path: Path) -> bool:
+    """Return True only for the FUSE mount created by this service.
+
+    Docker volumes and bind mounts are mount points too, so ``mountpoint -q``
+    cannot distinguish the writable mount target from an active rclone mount.
+    Linux exposes the filesystem type after `` - `` in ``mountinfo``.
+    """
+    target = str(path.resolve())
+    try:
+        lines = Path("/proc/self/mountinfo").read_text().splitlines()
+    except OSError:
+        return False
+    for line in lines:
+        fields = line.split()
+        if len(fields) < 10 or fields[4].replace("\\040", " ") != target:
+            continue
+        try:
+            separator = fields.index("-")
+        except ValueError:
+            continue
+        if separator + 1 < len(fields) and fields[separator + 1] == "fuse.rclone":
+            return True
+    return False
 
 
 def start_rclone_mount() -> None:
     global RCLONE_PROCESS
-    if is_mountpoint(WORKSPACES_ROOT):
+    if is_rclone_mount(WORKSPACES_ROOT):
         logger.info(f"Mountpoint already exists: {WORKSPACES_ROOT}")
         return
 
@@ -62,7 +78,7 @@ def start_rclone_mount() -> None:
 
     timeout = 20
     while timeout > 0:
-        if is_mountpoint(WORKSPACES_ROOT):
+        if is_rclone_mount(WORKSPACES_ROOT):
             logger.info(f"rclone mount ready: {WORKSPACES_ROOT}")
             return
         if RCLONE_PROCESS.poll() is not None:
@@ -85,10 +101,10 @@ def stop_rclone_mount() -> None:
                 logger.warning("Force killing rclone")
                 RCLONE_PROCESS.kill()
 
-    if is_mountpoint(WORKSPACES_ROOT):
+    if is_rclone_mount(WORKSPACES_ROOT):
         logger.info(f"Unmount {WORKSPACES_ROOT}")
         result = subprocess.run(["fusermount3", "-u", str(WORKSPACES_ROOT)])
         if result.returncode != 0:
             subprocess.run(["umount", "-l", str(WORKSPACES_ROOT)])
 
-    logger.info("rclone stopped")
+    logger.info("rclone stopped")

runtime/src/runtime/process.py

@@ -13,6 +13,7 @@ import os
 import secrets
 import subprocess
 import time
+from pathlib import Path
 from typing import TypedDict
 
 from fastapi import HTTPException
@@ -23,6 +24,7 @@ from common.utils import get_free_port, start_process
 from runtime.mount import WORKSPACES_ROOT
 
 PUBLIC_BASE_URL = settings.public_base_url
+JUPYTER_PROCESS_CWD = Path("/app")
 
 
 class JupyterProcessRecord(TypedDict):
@@ -106,7 +108,7 @@ async def start_workspace(ws_id: str) -> dict:
             "--allow-root",
             f"--ServerApp.token={token}",
             f"--ServerApp.base_url={base_path}",
-            "--notebook-dir=.",
+            f"--ServerApp.root_dir={workspace_path}",
             "--ServerApp.terminals_enabled=False",
             "--NotebookApp.terminals_enabled=False",
             "--ServerApp.allow_origin=*",
@@ -116,7 +118,11 @@ async def start_workspace(ws_id: str) -> dict:
         ]
 
         try:
-            process, log_file = start_process(cmd, workspace_path)
+            # Keep the server process cwd off the rclone/FUSE mount. Remote
+            # S3 directory refreshes can replace a virtual directory inode,
+            # leaving a long-lived cwd marked "(deleted)" and making
+            # os.getcwd() fail while Jupyter starts kernels or nbconvert.
+            process, log_file = start_process(cmd, JUPYTER_PROCESS_CWD)
         except Exception as e:
             logger.error(f"Failed to start Jupyter for {ws_id}: {e}")
             raise HTTPException(
@@ -264,4 +270,4 @@ async def scan_workspaces() -> None:
             except Exception as err:
                 logger.error(f"Startup failed for workspace '{entry}': {err}")
 
-    await asyncio.gather(*[_start(entry) for entry in entries])
+    await asyncio.gather(*[_start(entry) for entry in entries])