Fix symlink checkpoint1

This commit is contained in:
Will Miao
2025-02-22 20:28:47 +08:00
parent f1d8d5e9b4
commit cb054953aa
2 changed files with 157 additions and 44 deletions

View File

@@ -7,6 +7,7 @@ from watchdog.events import FileSystemEventHandler, FileCreatedEvent, FileDelete
from typing import List
from threading import Lock
from .lora_scanner import LoraScanner
import platform
logger = logging.getLogger(__name__)
@@ -22,14 +23,17 @@ class LoraFileHandler(FileSystemEventHandler):
self._ignore_paths = set() # Add ignore paths set
self._min_ignore_timeout = 5 # minimum timeout in seconds
self._download_speed = 1024 * 1024 # assume 1MB/s as base speed
self._path_mappings = {} # 添加路径映射字典
def _should_ignore(self, path: str) -> bool:
"""Check if path should be ignored"""
return path.replace(os.sep, '/') in self._ignore_paths
real_path = os.path.realpath(path) # Resolve any symbolic links
return real_path.replace(os.sep, '/') in self._ignore_paths
def add_ignore_path(self, path: str, file_size: int = 0):
"""Add path to ignore list with dynamic timeout based on file size"""
self._ignore_paths.add(path.replace(os.sep, '/'))
real_path = os.path.realpath(path) # Resolve any symbolic links
self._ignore_paths.add(real_path.replace(os.sep, '/'))
# Calculate timeout based on file size, with a minimum value
# Assuming average download speed of 1MB/s
@@ -38,14 +42,31 @@ class LoraFileHandler(FileSystemEventHandler):
(file_size / self._download_speed) * 1.5 # Add 50% buffer
)
logger.debug(f"Adding {path} to ignore list for {timeout:.1f} seconds")
logger.debug(f"Adding {real_path} to ignore list for {timeout:.1f} seconds")
asyncio.get_event_loop().call_later(
timeout,
self._ignore_paths.discard,
path
real_path.replace(os.sep, '/')
)
def add_path_mapping(self, link_path: str, target_path: str):
"""添加符号链接路径映射"""
normalized_link = os.path.normpath(link_path).replace(os.sep, '/')
normalized_target = os.path.normpath(target_path).replace(os.sep, '/')
self._path_mappings[normalized_target] = normalized_link
logger.debug(f"Added path mapping: {normalized_target} -> {normalized_link}")
def _map_path_to_link(self, path: str) -> str:
"""将目标路径映射回符号链接路径"""
normalized_path = os.path.normpath(path).replace(os.sep, '/')
for target_prefix, link_prefix in self._path_mappings.items():
if normalized_path.startswith(target_prefix):
mapped_path = normalized_path.replace(target_prefix, link_prefix, 1)
logger.debug(f"Mapped path {normalized_path} to {mapped_path}")
return mapped_path
return path
def on_created(self, event):
if event.is_directory or not event.src_path.endswith('.safetensors'):
return
@@ -65,11 +86,11 @@ class LoraFileHandler(FileSystemEventHandler):
def _schedule_update(self, action: str, file_path: str):
"""Schedule a cache update"""
with self.lock:
# 标准化路径
normalized_path = file_path.replace(os.sep, '/')
# 将目标路径映射回符号链接路径
mapped_path = self._map_path_to_link(file_path)
normalized_path = mapped_path.replace(os.sep, '/')
self.pending_changes.add((action, normalized_path))
# 使用 call_soon_threadsafe 在事件循环中安排任务
self.loop.call_soon_threadsafe(self._create_update_task)
def _create_update_task(self):
@@ -134,24 +155,111 @@ class LoraFileMonitor:
def __init__(self, scanner: LoraScanner, roots: List[str]):
self.scanner = scanner
scanner.set_file_monitor(self)
self.roots = roots
self.observer = Observer()
# 获取当前运行的事件循环
self.loop = asyncio.get_event_loop()
self.handler = LoraFileHandler(scanner, self.loop)
# 存储所有需要监控的路径(包括链接的目标路径)
self.monitor_paths = set()
for root in roots:
real_root = os.path.realpath(root)
self.monitor_paths.add(real_root)
# 扫描根目录下的链接
self._add_link_targets(root)
def _is_link(self, path: str) -> bool:
"""
检查路径是否为链接
支持:
- Windows: Symbolic Links, Junction Points
- Linux: Symbolic Links
"""
try:
# 首先检查通用的符号链接
if os.path.islink(path):
return True
# Windows 特定的 Junction Points 检测
if platform.system() == 'Windows':
try:
import ctypes
FILE_ATTRIBUTE_REPARSE_POINT = 0x400
attrs = ctypes.windll.kernel32.GetFileAttributesW(str(path))
return attrs != -1 and (attrs & FILE_ATTRIBUTE_REPARSE_POINT)
except Exception as e:
logger.error(f"Error checking Windows reparse point: {e}")
return False
except Exception as e:
logger.error(f"Error checking link status for {path}: {e}")
return False
def _get_link_target(self, path: str) -> str:
"""获取链接目标路径"""
try:
return os.path.realpath(path)
except Exception as e:
logger.error(f"Error resolving link target for {path}: {e}")
return path
def _add_link_targets(self, root: str):
"""递归扫描目录,添加链接指向的目标路径"""
try:
with os.scandir(root) as it:
for entry in it:
logger.debug(f"Checking path: {entry.path}")
if self._is_link(entry.path):
# 获取链接的目标路径
target_path = self._get_link_target(entry.path)
if os.path.isdir(target_path):
normalized_target = os.path.normpath(target_path)
self.monitor_paths.add(normalized_target)
# 添加路径映射到处理器
self.handler.add_path_mapping(entry.path, target_path)
logger.info(f"Found link: {entry.path} -> {normalized_target}")
# 递归扫描目标目录中的链接
self._add_link_targets(target_path)
elif entry.is_dir(follow_symlinks=False):
# 递归扫描子目录
self._add_link_targets(entry.path)
except Exception as e:
logger.error(f"Error scanning links in {root}: {e}")
def start(self):
"""Start monitoring"""
for root in self.roots:
for path_info in self.monitor_paths:
try:
self.observer.schedule(self.handler, root, recursive=True)
logger.info(f"Started monitoring: {root}")
if isinstance(path_info, tuple):
# 对于链接,监控目标路径
_, target_path = path_info
self.observer.schedule(self.handler, target_path, recursive=True)
logger.info(f"Started monitoring target path: {target_path}")
else:
# 对于普通路径,直接监控
self.observer.schedule(self.handler, path_info, recursive=True)
logger.info(f"Started monitoring: {path_info}")
except Exception as e:
logger.error(f"Error monitoring {root}: {e}")
logger.error(f"Error monitoring {path_info}: {e}")
self.observer.start()
def stop(self):
"""Stop monitoring"""
self.observer.stop()
self.observer.join()
self.observer.join()
def rescan_links(self):
"""重新扫描链接(当添加新的链接时调用)"""
new_paths = set()
for path in self.monitor_paths.copy():
self._add_link_targets(path)
# 添加新发现的路径到监控
new_paths = self.monitor_paths - set(self.observer.watches.keys())
for path in new_paths:
try:
self.observer.schedule(self.handler, path, recursive=True)
logger.info(f"Added new monitoring path: {path}")
except Exception as e:
logger.error(f"Error adding new monitor for {path}: {e}")

View File

@@ -231,23 +231,35 @@ class LoraScanner:
async def _scan_directory(self, root_path: str) -> List[Dict]:
"""Scan a single directory for LoRA files"""
loras = []
original_root = root_path # 保存原始根路径
# 使用异步安全的目录遍历方式
async def scan_recursive(path: str):
async def scan_recursive(path: str, visited_paths: set):
"""递归扫描目录,避免循环链接"""
try:
real_path = os.path.realpath(path)
if real_path in visited_paths:
logger.debug(f"Skipping already visited path: {path}")
return
visited_paths.add(real_path)
with os.scandir(path) as it:
entries = list(it) # 同步获取目录条目
entries = list(it)
for entry in entries:
if entry.is_file() and entry.name.endswith('.safetensors'):
file_path = entry.path.replace(os.sep, "/")
await self._process_single_file(file_path, root_path, loras)
await asyncio.sleep(0) # 释放事件循环
elif entry.is_dir():
await scan_recursive(entry.path)
try:
if entry.is_file(follow_symlinks=True) and entry.name.endswith('.safetensors'):
# 使用原始路径而不是真实路径
file_path = entry.path.replace(os.sep, "/")
await self._process_single_file(file_path, original_root, loras)
await asyncio.sleep(0)
elif entry.is_dir(follow_symlinks=True):
# 对于目录,使用原始路径继续扫描
await scan_recursive(entry.path, visited_paths)
except Exception as e:
logger.error(f"Error processing entry {entry.path}: {e}")
except Exception as e:
logger.error(f"Error scanning {path}: {e}")
await scan_recursive(root_path)
await scan_recursive(root_path, set())
return loras
async def _process_single_file(self, file_path: str, root_path: str, loras: list):
@@ -316,6 +328,7 @@ class LoraScanner:
def _calculate_folder(self, file_path: str) -> str:
"""Calculate the folder path for a LoRA file"""
# 使用原始路径计算相对路径
for root in config.loras_roots:
if file_path.startswith(root):
rel_path = os.path.relpath(file_path, root)
@@ -323,46 +336,38 @@ class LoraScanner:
return ''
async def move_model(self, source_path: str, target_path: str) -> bool:
"""Move a model and its associated files to a new location
Args:
source_path: Full path to the source lora file
target_path: Full path to the target directory
Returns:
bool: True if successful, False otherwise
"""
"""Move a model and its associated files to a new location"""
try:
# Ensure paths are normalized
# 保持原始路径格式
source_path = source_path.replace(os.sep, '/')
target_path = target_path.replace(os.sep, '/')
# Get base name without extension
# 其余代码保持不变
base_name = os.path.splitext(os.path.basename(source_path))[0]
source_dir = os.path.dirname(source_path)
# Create target directory if it doesn't exist
os.makedirs(target_path, exist_ok=True)
# Calculate target lora path
target_lora = os.path.join(target_path, f"{base_name}.safetensors").replace(os.sep, '/')
# Get source file size for timeout calculation
file_size = os.path.getsize(source_path)
# 使用真实路径进行文件操作
real_source = os.path.realpath(source_path)
real_target = os.path.realpath(target_lora)
file_size = os.path.getsize(real_source)
# Tell file monitor to ignore these paths
if self.file_monitor:
self.file_monitor.handler.add_ignore_path(
source_path,
real_source,
file_size
)
self.file_monitor.handler.add_ignore_path(
target_lora,
real_target,
file_size
)
# Move main lora file
shutil.move(source_path, target_lora)
# 使用真实路径进行文件操作
shutil.move(real_source, real_target)
# Move associated files
source_metadata = os.path.join(source_dir, f"{base_name}.metadata.json")