tao.chen/model-platform
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: add schedule package

Browse Source
This commit is contained in:
tao.chen
2026-07-29 13:35:32 +08:00
parent 128aa7563f
commit 233dc90efe
10 changed files with 1038 additions and 830 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docker-compose.yml
+22
View File
@@ -29,10 +29,32 @@ services:
build:
context: .
dockerfile: runtime/Dockerfile
cap_add:
- SYS_ADMIN
devices:
- /dev/fuse:/dev/fuse
security_opt:
- apparmor:unconfined
ports:
- "8001:8001"
environment:
- PUBLIC_BASE_URL=http://runtime
# --- Rclone 动态环境变量配置 (对应名称 rustfs) ---
- RCLONE_CONFIG_RUSTFS_TYPE=s3
- RCLONE_CONFIG_RUSTFS_PROVIDER=Other
- RCLONE_CONFIG_RUSTFS_ACCESS_KEY_ID=BdsXeamEnvSDQnk8tRxh
- RCLONE_CONFIG_RUSTFS_SECRET_ACCESS_KEY=mmBVc3RqzbT2VX3ysKGnirYH5kYD3ww3wFtMVvrb
# 替换为你的 RustFS 服务地址（如果是同 docker-compose 网络下的服务，可以直接填服务名:端口）
- RCLONE_CONFIG_RUSTFS_ENDPOINT=http://8.153.151.51:9000
# 自建 S3 建议强制开启 Path-style 访问 (http://endpoint/bucket)
- RCLONE_CONFIG_RUSTFS_ENV_AUTH=false
- RCLONE_CONFIG_RUSTFS_FORCE_PATH_STYLE=true
- RCLONE_CONFIG_RUSTFS_REGION=other
# --- Runtime 逻辑环境变量 ---
- REMOTE_BUCKET=rustfs:workspaces
- WORKSPACES_ROOT=/app/workspaces
volumes:
- ./runtime:/app/runtime:ro
- ./common:/app/common:ro
pyproject.toml
+4 -1
View File
@@ -2,11 +2,14 @@
name = "model-platform"
version = "0.1.0"
requires-python = ">=3.12"
dependencies = []
dependencies = [
"ipykernel>=7.3.0",
]
[tool.uv.workspace]
members = [
"backend",
"runtime",
"common",
"schedule",
]
runtime/Dockerfile
+12
View File
@@ -3,6 +3,18 @@ FROM python:3.12-slim
WORKDIR /app
# 安装系统依赖（fuse3 是 rclone mount 的核心底层依赖）
RUN apt-get update && apt-get install -y --no-install-recommends \
fuse3 \
ca-certificates \
curl \
procps \
&& sed -i 's/#user_allow_other/user_allow_other/g' /etc/fuse.conf \
&& rm -rf /var/lib/apt/lists/*
# 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
# 安装 uv
COPY --from=ghcr.io/astral-sh/uv:latest /uv /bin/uv
runtime/pyproject.toml
+4
View File
@@ -15,6 +15,10 @@ dependencies = [
"uvicorn>=0.51.0",
]
[project.scripts]
runtime = "runtime:main"
[[tool.uv.index]]
url = "https://pypi.tuna.tsinghua.edu.cn/simple/"
default = true
runtime/src/runtime/main.py
+169 -35
View File
@@ -19,8 +19,147 @@ from loguru import logger
# 全局内存字典：记录运行中的 Jupyter 进程信息
JUPYTER_PROCESSES: Dict[str, dict] = {}
WORKSPACES_ROOT = os.getenv("WORKSPACES_ROOT", "/app/workspaces")
WORKSPACES_ROOT = Path(os.getenv("WORKSPACES_ROOT", "/app/workspaces"))
PUBLIC_BASE_URL = os.getenv("PUBLIC_BASE_URL", "http://localhost")
REMOTE_BUCKET = os.getenv("REMOTE_BUCKET", "rustfs:workspaces")
RCLONE_PROCESS = None
def is_mountpoint(path: Path) -> bool:
"""
判断目录是否已经挂载
"""
result = subprocess.run(
["mountpoint", "-q", str(path)],
stdout=subprocess.DEVNULL,
stderr=subprocess.DEVNULL,
)
return result.returncode == 0
def start_rclone_mount():
"""
启动 rclone mount
"""
global RCLONE_PROCESS
if is_mountpoint(WORKSPACES_ROOT):
logger.info(
f"Mountpoint already exists: {WORKSPACES_ROOT}"
)
return
WORKSPACES_ROOT.mkdir(parents=True, exist_ok=True)
logger.info(
f"Starting rclone mount "
f"{REMOTE_BUCKET} -> {WORKSPACES_ROOT}"
)
log_file = open(
"/tmp/rclone-mount.log",
"a",
buffering=1,
)
cmd = [
"rclone",
"mount",
REMOTE_BUCKET,
WORKSPACES_ROOT.as_posix(),
"--allow-other",
"--vfs-cache-mode","full",
"--vfs-cache-max-size","20G",
"--vfs-write-back","5s",
"--dir-cache-time","30s",
"--poll-interval","30s",
"--log-level","INFO",
]
RCLONE_PROCESS = subprocess.Popen(
cmd,
stdout=log_file,
stderr=subprocess.STDOUT,
start_new_session=True,
)
# 等待 mount ready
timeout = 20
while timeout > 0:
if is_mountpoint(WORKSPACES_ROOT):
logger.info(f"rclone mount ready: {WORKSPACES_ROOT}" )
return
# rclone异常退出
if RCLONE_PROCESS.poll() is not None:
raise RuntimeError( "rclone mount process exited")
time.sleep(1)
timeout -= 1
raise RuntimeError( f"Timeout waiting mount: {WORKSPACES_ROOT}")
def stop_rclone_mount():
global RCLONE_PROCESS
logger.info(
"Stopping rclone mount..."
)
if RCLONE_PROCESS:
if RCLONE_PROCESS.poll() is None:
RCLONE_PROCESS.terminate()
try:
RCLONE_PROCESS.wait(timeout=10)
except subprocess.TimeoutExpired:
logger.warning("Force killing rclone")
RCLONE_PROCESS.kill()
if is_mountpoint(WORKSPACES_ROOT):
logger.info(f"Unmount {WORKSPACES_ROOT}")
result = subprocess.run(
[
"fusermount3",
"-u",
WORKSPACES_ROOT.as_posix(),
]
)
if result.returncode != 0:
subprocess.run(
[
"umount",
"-l",
WORKSPACES_ROOT.as_posix(),
]
)
logger.info("rclone stopped")
def scan_workspaces():
"""
扫描已有 workspace
"""
if not WORKSPACES_ROOT.exists():
return
try:
entries = os.listdir(WORKSPACES_ROOT)
except Exception as e:
logger.error(f"scan workspace failed: {e}")
return
for entry in entries:
path = WORKSPACES_ROOT / entry
if not path.is_dir():
continue
logger.info(f"Found workspace: {entry}" )
try:
full_path = os.path.join(WORKSPACES_ROOT, entry)
if os.path.isdir(full_path):
logger.info(f"Auto-starting Jupyter for workspace ID: '{entry}'")
try:
_handle_start(entry)
except Exception as err:
logger.error(f"Startup failed for workspace '{entry}': {err}")
except Exception as e:
logger.error(f"Start workspace {entry} failed: {e}" )
def get_free_port() -> int:
@@ -76,10 +215,9 @@ def start_process(cmd, workspace_path, log_dir="/tmp/process_logs"):
def _handle_start(ws_id: str):
workspace_path = Path(f"./test/workspaces/{ws_id}")
workspace_path.mkdir(parents=True, exist_ok=True)
workspace_path = WORKSPACES_ROOT / ws_id
# 1. 如果已存在，校验进程状态并复用
# 如果已存在，校验进程状态并复用
if ws_id in JUPYTER_PROCESSES:
p_info = JUPYTER_PROCESSES[ws_id]
if p_info["process"].poll() is None:
@@ -196,6 +334,7 @@ def _handle_list():
}
return {"status": "success", "instances": active_instances}
def _handle_get(ws_id: str):
"""【新增】获取指定 Workspace 的 Jupyter 运行状态与完整 URL"""
if ws_id not in JUPYTER_PROCESSES:
@@ -229,44 +368,39 @@ def _handle_get(ws_id: str):
# ==================== FastAPI Lifespan 定义 ====================
@asynccontextmanager
async def lifespan(app: FastAPI):
# ---------------- 1. 启动阶段 (Startup) ----------------
logger.info(
f"Initializing workspace processes from root: '{WORKSPACES_ROOT}'..."
)
os.makedirs(WORKSPACES_ROOT, exist_ok=True)
global RCLONE_PROCESS
logger.info("Starting up Runtime Service...")
start_rclone_mount()
logger.info(f"Scanning workspaces: {WORKSPACES_ROOT}")
scan_workspaces()
logger.info("Runtime Service started")
# 遍历总路径下的每一个子目录（每个子目录名即 workspace_id）
for entry in os.listdir(WORKSPACES_ROOT):
full_path = os.path.join(WORKSPACES_ROOT, entry)
if os.path.isdir(full_path):
logger.info(f"Auto-starting Jupyter for workspace ID: '{entry}'")
# ==================== 2. 服务运行阶段 (Serving) ====================
try:
yield # 服务保持运行，等待并处理 API 请求
finally:
logger.info("Service is shutting down. Terminating all active Jupyter sub-processes...")
# 优先杀死所有 Jupyter 子进程（确保文件句柄被释放）
active_workspaces = list(JUPYTER_PROCESSES.keys())
for ws_id in active_workspaces:
try:
_handle_start(entry)
_handle_stop(ws_id)
except Exception as err:
logger.error(
f"Startup failed for workspace '{entry}': {err}"
)
logger.error(f"Error terminating Jupyter process for '{ws_id}': {err}")
logger.info("All Jupyter sub-processes have been terminated.")
JUPYTER_PROCESSES.clear()
yield # 服务在此处保持正常运行，等待 API 请求
# ---------------- 2. 停止阶段 (Shutdown) ----------------
logger.info(
"Service is shutting down. Terminating all active Jupyter sub-processes..."
)
# 复制 key 列表，防止字典在迭代清理过程中因大小改变报错
active_workspaces = list(JUPYTER_PROCESSES.keys())
for ws_id in active_workspaces:
# 卸载 Rclone 挂载点（强制将 VFS 缓存刷新同步至对象存储）
try:
_handle_stop(ws_id)
except Exception as err:
logger.error(
f"Error terminating Jupyter process for '{ws_id}': {err}"
)
logger.info("All Jupyter sub-processes have been clean terminated.")
stop_rclone_mount()
except Exception as e:
logger.error(f"Stop rclone failed: {e}" )
logger.info("Runtime Service stopped")
app = FastAPI(lifespan=lifespan)
# ---------------- 统一入口 POST 接口 ----------------
@app.get("/api/v1/health")
def healthz():
schedule/.python-version
+1
View File
@@ -0,0 +1 @@
3.12
schedule/README.md
View File
schedule/pyproject.toml
+24
View File
@@ -0,0 +1,24 @@
[project]
name = "schedule"
version = "0.1.0"
description = "Add your description here"
readme = "README.md"
authors = [
{ name = "tao.chen", email = "93983997+taochen-ct@users.noreply.github.com" }
]
requires-python = ">=3.12"
dependencies = []
[project.scripts]
schedule = "schedule:main"
[[tool.uv.index]]
url = "https://pypi.tuna.tsinghua.edu.cn/simple/"
default = true
[tool.uv.sources]
common = { workspace = true }
[build-system]
requires = ["hatchling"]
build-backend = "hatchling.build"
schedule/src/schedule/__init__.py
+2
View File
@@ -0,0 +1,2 @@
def main() -> None:
print("Hello from schedule!")
uv.lock
Generated
+800 -794
View File
File diff suppressed because it is too large Load Diff