checkpoint

This commit is contained in:
Will Miao
2025-04-10 09:08:36 +08:00
parent 64c9e4aeca
commit 8fdfb68741
12 changed files with 1397 additions and 484 deletions

View File

@@ -4,13 +4,11 @@ import logging
import asyncio
import shutil
import time
from typing import List, Dict, Optional
from typing import List, Dict, Optional, Set
from ..utils.models import LoraMetadata
from ..config import config
from ..utils.file_utils import load_metadata, get_file_info, normalize_path, find_preview_file, save_metadata
from ..utils.lora_metadata import extract_lora_metadata
from .lora_cache import LoraCache
from .model_scanner import ModelScanner
from .lora_hash_index import LoraHashIndex
from .settings_manager import settings
from ..utils.constants import NSFW_LEVELS
@@ -19,7 +17,7 @@ import sys
logger = logging.getLogger(__name__)
class LoraScanner:
class LoraScanner(ModelScanner):
"""Service for scanning and managing LoRA files"""
_instance = None
@@ -31,20 +29,20 @@ class LoraScanner:
return cls._instance
def __init__(self):
# 确保初始化只执行一次
# Ensure initialization happens only once
if not hasattr(self, '_initialized'):
self._cache: Optional[LoraCache] = None
self._hash_index = LoraHashIndex()
self._initialization_lock = asyncio.Lock()
self._initialization_task: Optional[asyncio.Task] = None
# Define supported file extensions
file_extensions = {'.safetensors'}
# Initialize parent class
super().__init__(
model_type="lora",
model_class=LoraMetadata,
file_extensions=file_extensions,
hash_index=LoraHashIndex()
)
self._initialized = True
self.file_monitor = None # Add this line
self._tags_count = {} # Add a dictionary to store tag counts
def set_file_monitor(self, monitor):
"""Set file monitor instance"""
self.file_monitor = monitor
@classmethod
async def get_instance(cls):
"""Get singleton instance with async support"""
@@ -52,89 +50,74 @@ class LoraScanner:
if cls._instance is None:
cls._instance = cls()
return cls._instance
async def get_cached_data(self, force_refresh: bool = False) -> LoraCache:
"""Get cached LoRA data, refresh if needed"""
async with self._initialization_lock:
def get_model_roots(self) -> List[str]:
"""Get lora root directories"""
return config.loras_roots
async def scan_all_models(self) -> List[Dict]:
"""Scan all LoRA directories and return metadata"""
all_loras = []
# Create scan tasks for each directory
scan_tasks = []
for lora_root in self.get_model_roots():
task = asyncio.create_task(self._scan_directory(lora_root))
scan_tasks.append(task)
# 如果缓存未初始化但需要响应请求,返回空缓存
if self._cache is None and not force_refresh:
return LoraCache(
raw_data=[],
sorted_by_name=[],
sorted_by_date=[],
folders=[]
)
# 如果正在初始化,等待完成
if self._initialization_task and not self._initialization_task.done():
try:
await self._initialization_task
except Exception as e:
logger.error(f"Cache initialization failed: {e}")
self._initialization_task = None
if (self._cache is None or force_refresh):
# Wait for all tasks to complete
for task in scan_tasks:
try:
loras = await task
all_loras.extend(loras)
except Exception as e:
logger.error(f"Error scanning directory: {e}")
# 创建新的初始化任务
if not self._initialization_task or self._initialization_task.done():
self._initialization_task = asyncio.create_task(self._initialize_cache())
return all_loras
async def _scan_directory(self, root_path: str) -> List[Dict]:
"""Scan a single directory for LoRA files"""
loras = []
original_root = root_path # Save original root path
async def scan_recursive(path: str, visited_paths: set):
"""Recursively scan directory, avoiding circular symlinks"""
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)
try:
await self._initialization_task
except Exception as e:
logger.error(f"Cache initialization failed: {e}")
# 如果缓存已存在,继续使用旧缓存
if self._cache is None:
raise # 如果没有缓存,则抛出异常
return self._cache
with os.scandir(path) as it:
entries = list(it)
for entry in entries:
try:
if entry.is_file(follow_symlinks=True) and any(entry.name.endswith(ext) for ext in self.file_extensions):
# Use original path instead of real path
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):
# For directories, continue scanning with original path
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}")
async def _initialize_cache(self) -> None:
"""Initialize or refresh the cache"""
await scan_recursive(root_path, set())
return loras
async def _process_single_file(self, file_path: str, root_path: str, loras: list):
"""Process a single file and add to results list"""
try:
start_time = time.time()
# Clear existing hash index
self._hash_index.clear()
# Clear existing tags count
self._tags_count = {}
# Scan for new data
raw_data = await self.scan_all_loras()
# Build hash index and tags count
for lora_data in raw_data:
if 'sha256' in lora_data and 'file_path' in lora_data:
self._hash_index.add_entry(lora_data['sha256'].lower(), lora_data['file_path'])
# Count tags
if 'tags' in lora_data and lora_data['tags']:
for tag in lora_data['tags']:
self._tags_count[tag] = self._tags_count.get(tag, 0) + 1
# Update cache
self._cache = LoraCache(
raw_data=raw_data,
sorted_by_name=[],
sorted_by_date=[],
folders=[]
)
# Call resort_cache to create sorted views
await self._cache.resort()
self._initialization_task = None
logger.info(f"LoRA Manager: Cache initialization completed in {time.time() - start_time:.2f} seconds, found {len(raw_data)} loras")
result = await self._process_model_file(file_path, root_path)
if result:
loras.append(result)
except Exception as e:
logger.error(f"LoRA Manager: Error initializing cache: {e}")
self._cache = LoraCache(
raw_data=[],
sorted_by_name=[],
sorted_by_date=[],
folders=[]
)
logger.error(f"Error processing {file_path}: {e}")
async def get_paginated_data(self, page: int, page_size: int, sort_by: str = 'name',
folder: str = None, search: str = None, fuzzy_search: bool = False,
base_models: list = None, tags: list = None,
@@ -280,240 +263,14 @@ class LoraScanner:
return result
def invalidate_cache(self):
"""Invalidate the current cache"""
self._cache = None
async def scan_all_loras(self) -> List[Dict]:
"""Scan all LoRA directories and return metadata"""
all_loras = []
# 分目录异步扫描
scan_tasks = []
for loras_root in config.loras_roots:
task = asyncio.create_task(self._scan_directory(loras_root))
scan_tasks.append(task)
for task in scan_tasks:
try:
loras = await task
all_loras.extend(loras)
except Exception as e:
logger.error(f"Error scanning directory: {e}")
return all_loras
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
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)
for entry in entries:
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, set())
return loras
async def _process_single_file(self, file_path: str, root_path: str, loras: list):
"""处理单个文件并添加到结果列表"""
try:
result = await self._process_lora_file(file_path, root_path)
if result:
loras.append(result)
except Exception as e:
logger.error(f"Error processing {file_path}: {e}")
async def _process_lora_file(self, file_path: str, root_path: str) -> Dict:
"""Process a single LoRA file and return its metadata"""
# Try loading existing metadata
metadata = await load_metadata(file_path)
if metadata is None:
# Try to find and use .civitai.info file first
civitai_info_path = f"{os.path.splitext(file_path)[0]}.civitai.info"
if os.path.exists(civitai_info_path):
try:
with open(civitai_info_path, 'r', encoding='utf-8') as f:
version_info = json.load(f)
file_info = next((f for f in version_info.get('files', []) if f.get('primary')), None)
if file_info:
# Create a minimal file_info with the required fields
file_name = os.path.splitext(os.path.basename(file_path))[0]
file_info['name'] = file_name
# Use from_civitai_info to create metadata
metadata = LoraMetadata.from_civitai_info(version_info, file_info, file_path)
metadata.preview_url = find_preview_file(file_name, os.path.dirname(file_path))
await save_metadata(file_path, metadata)
logger.debug(f"Created metadata from .civitai.info for {file_path}")
except Exception as e:
logger.error(f"Error creating metadata from .civitai.info for {file_path}: {e}")
# If still no metadata, create new metadata using get_file_info
if metadata is None:
metadata = await get_file_info(file_path)
# Convert to dict and add folder info
lora_data = metadata.to_dict()
# Try to fetch missing metadata from Civitai if needed
await self._fetch_missing_metadata(file_path, lora_data)
rel_path = os.path.relpath(file_path, root_path)
folder = os.path.dirname(rel_path)
lora_data['folder'] = folder.replace(os.path.sep, '/')
return lora_data
async def _fetch_missing_metadata(self, file_path: str, lora_data: Dict) -> None:
"""Fetch missing description and tags from Civitai if needed
Args:
file_path: Path to the lora file
lora_data: Lora metadata dictionary to update
"""
try:
# Skip if already marked as deleted on Civitai
if lora_data.get('civitai_deleted', False):
logger.debug(f"Skipping metadata fetch for {file_path}: marked as deleted on Civitai")
return
# Check if we need to fetch additional metadata from Civitai
needs_metadata_update = False
model_id = None
# Check if we have Civitai model ID but missing metadata
if lora_data.get('civitai'):
# Try to get model ID directly from the correct location
model_id = lora_data['civitai'].get('modelId')
if model_id:
model_id = str(model_id)
# Check if tags are missing or empty
tags_missing = not lora_data.get('tags') or len(lora_data.get('tags', [])) == 0
# Check if description is missing or empty
desc_missing = not lora_data.get('modelDescription') or lora_data.get('modelDescription') in (None, "")
needs_metadata_update = tags_missing or desc_missing
# Fetch missing metadata if needed
if needs_metadata_update and model_id:
logger.debug(f"Fetching missing metadata for {file_path} with model ID {model_id}")
from ..services.civitai_client import CivitaiClient
client = CivitaiClient()
# Get metadata and status code
model_metadata, status_code = await client.get_model_metadata(model_id)
await client.close()
# Handle 404 status (model deleted from Civitai)
if status_code == 404:
logger.warning(f"Model {model_id} appears to be deleted from Civitai (404 response)")
# Mark as deleted to avoid future API calls
lora_data['civitai_deleted'] = True
# Save the updated metadata back to file
metadata_path = os.path.splitext(file_path)[0] + '.metadata.json'
with open(metadata_path, 'w', encoding='utf-8') as f:
json.dump(lora_data, f, indent=2, ensure_ascii=False)
# Process valid metadata if available
elif model_metadata:
logger.debug(f"Updating metadata for {file_path} with model ID {model_id}")
# Update tags if they were missing
if model_metadata.get('tags') and (not lora_data.get('tags') or len(lora_data.get('tags', [])) == 0):
lora_data['tags'] = model_metadata['tags']
# Update description if it was missing
if model_metadata.get('description') and (not lora_data.get('modelDescription') or lora_data.get('modelDescription') in (None, "")):
lora_data['modelDescription'] = model_metadata['description']
# Save the updated metadata back to file
metadata_path = os.path.splitext(file_path)[0] + '.metadata.json'
with open(metadata_path, 'w', encoding='utf-8') as f:
json.dump(lora_data, f, indent=2, ensure_ascii=False)
except Exception as e:
logger.error(f"Failed to update metadata from Civitai for {file_path}: {e}")
async def update_preview_in_cache(self, file_path: str, preview_url: str) -> bool:
"""Update preview URL in cache for a specific lora
Args:
file_path: The file path of the lora to update
preview_url: The new preview URL
Returns:
bool: True if the update was successful, False if cache doesn't exist or lora wasn't found
"""
if self._cache is None:
return False
return await self._cache.update_preview_url(file_path, preview_url)
async def scan_single_lora(self, file_path: str) -> Optional[Dict]:
"""Scan a single LoRA file and return its metadata"""
try:
if not os.path.exists(os.path.realpath(file_path)):
return None
# 获取基本文件信息
metadata = await get_file_info(file_path)
if not metadata:
return None
folder = self._calculate_folder(file_path)
# 确保 folder 字段存在
metadata_dict = metadata.to_dict()
metadata_dict['folder'] = folder or ''
return metadata_dict
except Exception as e:
logger.error(f"Error scanning {file_path}: {e}")
return None
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)
return os.path.dirname(rel_path).replace(os.path.sep, '/')
return ''
async def move_model(self, source_path: str, target_path: str) -> bool:
"""Move a model and its associated files to a new location"""
try:
# 保持原始路径格式
# Keep original path format
source_path = source_path.replace(os.sep, '/')
target_path = target_path.replace(os.sep, '/')
# 其余代码保持不变
# Rest of the code remains unchanged
base_name = os.path.splitext(os.path.basename(source_path))[0]
source_dir = os.path.dirname(source_path)
@@ -521,7 +278,7 @@ class LoraScanner:
target_lora = os.path.join(target_path, f"{base_name}.safetensors").replace(os.sep, '/')
# 使用真实路径进行文件操作
# Use real paths for file operations
real_source = os.path.realpath(source_path)
real_target = os.path.realpath(target_lora)
@@ -537,7 +294,7 @@ class LoraScanner:
file_size
)
# 使用真实路径进行文件操作
# Use real paths for file operations
shutil.move(real_source, real_target)
# Move associated files
@@ -648,7 +405,7 @@ class LoraScanner:
except Exception as e:
logger.error(f"Error updating metadata paths: {e}", exc_info=True)
# Add new methods for hash index functionality
# Lora-specific hash index functionality
def has_lora_hash(self, sha256: str) -> bool:
"""Check if a LoRA with given hash exists"""
return self._hash_index.has_hash(sha256.lower())
@@ -681,16 +438,8 @@ class LoraScanner:
return None
# Add new method to get top tags
async def get_top_tags(self, limit: int = 20) -> List[Dict[str, any]]:
"""Get top tags sorted by count
Args:
limit: Maximum number of tags to return
Returns:
List of dictionaries with tag name and count, sorted by count
"""
"""Get top tags sorted by count"""
# Make sure cache is initialized
await self.get_cached_data()
@@ -705,14 +454,7 @@ class LoraScanner:
return sorted_tags[:limit]
async def get_base_models(self, limit: int = 20) -> List[Dict[str, any]]:
"""Get base models used in loras sorted by frequency
Args:
limit: Maximum number of base models to return
Returns:
List of dictionaries with base model name and count, sorted by count
"""
"""Get base models used in loras sorted by frequency"""
# Make sure cache is initialized
cache = await self.get_cached_data()