Merge pull request #1134 from willmiao/perf/bulk-rename-apply

perf(rename): make bulk filename-template apply O(n) instead of O(n²)

py/services/model_lifecycle_service.py

@@ -2,10 +2,12 @@
 
 from __future__ import annotations
 
+import asyncio
 import json
 import logging
 import os
-from typing import Any, Awaitable, Callable, Dict, Iterable, List, Mapping, Optional, TYPE_CHECKING, cast
+from contextlib import asynccontextmanager
+from typing import Any, AsyncIterator, Awaitable, Callable, Dict, Iterable, List, Mapping, Optional, TYPE_CHECKING, cast
 
 from ..services.service_registry import ServiceRegistry
 from ..services.pending_delete_service import get_pending_delete_service
@@ -107,6 +109,36 @@ def _require_path_in_library_roots(file_path: str, scanner, *, label: str = "pat
     )
 
 
+class BulkRenameContext:
+    """Per-session state threaded through ``rename_model`` calls of a bulk rename.
+
+    Holds the lazily built recipe hash index so a bulk rename loop pays the
+    O(recipes) index build at most once (on the first recipe-touching rename)
+    instead of rescanning every recipe per renamed file. Also tracks whether
+    any recipe was re-pointed so the session finalizes recipe maintenance only
+    when needed.
+    """
+
+    def __init__(self, recipe_scanner: Any) -> None:
+        self._recipe_scanner = recipe_scanner
+        self._recipe_hash_index: Optional[Dict[str, List[Dict[str, Any]]]] = None
+        self.recipes_touched = False
+
+    @property
+    def recipe_scanner(self) -> Any:
+        return self._recipe_scanner
+
+    async def get_recipe_hash_index(self) -> Optional[Dict[str, List[Dict[str, Any]]]]:
+        """Return the lora-hash → recipes index, building it on first use."""
+        if self._recipe_scanner is None:
+            return None
+        if self._recipe_hash_index is None:
+            self._recipe_hash_index = (
+                await self._recipe_scanner.build_lora_hash_index()
+            )
+        return self._recipe_hash_index
+
+
 class ModelLifecycleService:
     """Co-ordinate destructive and mutating model operations."""
 
@@ -365,10 +397,45 @@ class ModelLifecycleService:
 
         return await self._scanner.bulk_delete_models(file_paths)
 
+    @asynccontextmanager
+    async def bulk_rename_session(self) -> AsyncIterator[BulkRenameContext]:
+        """Context for bulk rename loops (filename-template "Apply to Library").
+
+        While active, the per-file ``update_single_model_cache`` resort/persist
+        chain and the per-file recipe folder-metadata refresh/resort are
+        deferred; both run exactly once when the outermost session exits — see
+        ``ModelScanner.defer_cache_persist`` and
+        ``RecipeScanner.finalize_bulk_filename_updates``. The finalize steps run
+        even on cancellation or mid-loop errors, because files are already
+        renamed on disk and the caches must not be left diverging.
+
+        Yields a :class:`BulkRenameContext` to pass as ``bulk_context`` into
+        each ``rename_model`` call of the loop.
+        """
+        recipe_scanner = await self._recipe_scanner_factory()
+        context = BulkRenameContext(recipe_scanner)
+        async with self._scanner.defer_cache_persist():
+            try:
+                yield context
+            finally:
+                if recipe_scanner is not None and context.recipes_touched:
+                    try:
+                        await recipe_scanner.finalize_bulk_filename_updates()
+                    except Exception as exc:  # pragma: no cover - defensive logging
+                        logger.error(
+                            "Error finalizing bulk recipe updates: %s", exc
+                        )
+
     async def rename_model(
-        self, *, file_path: str, new_file_name: str
+        self, *, file_path: str, new_file_name: str, bulk_context: Optional[BulkRenameContext] = None
     ) -> Dict[str, object]:
-        """Rename a model and its companion artefacts."""
+        """Rename a model and its companion artefacts.
+
+        When ``bulk_context`` is given (bulk rename loop), the recipe
+        re-pointing uses the session's prebuilt hash index and defers recipe
+        maintenance to the session finalize; the scanner cache persist is
+        likewise deferred by the surrounding ``bulk_rename_session``.
+        """
 
         if not file_path or not new_file_name:
             raise ValueError("File path and new file name are required")
@@ -419,20 +486,11 @@ class ModelLifecycleService:
             raw_hash = metadata.get("sha256") if isinstance(metadata, dict) else None
             hash_value = raw_hash if isinstance(raw_hash, str) else None
 
-        renamed_files: List[str] = []
-        new_metadata_path: Optional[str] = None
         new_preview: Optional[str] = None
 
-        for old_path, pattern in existing_files:
-            ext = self._get_multipart_ext(pattern)
-            new_path = os.path.join(
-                os.path.dirname(old_path), f"{new_file_name}{ext}"
-            ).replace(os.sep, "/")
-            os.rename(old_path, new_path)
-            renamed_files.append(new_path)
-
-            if ext == ".metadata.json":
-                new_metadata_path = new_path
+        renamed_files, new_metadata_path = await asyncio.to_thread(
+            self._rename_companion_files, existing_files, new_file_name
+        )
 
         if metadata and new_metadata_path:
             metadata["file_name"] = new_file_name
@@ -457,12 +515,26 @@ class ModelLifecycleService:
             )
 
             if hash_value and getattr(self._scanner, "model_type", "") == "lora":
-                recipe_scanner = await self._recipe_scanner_factory()
+                if bulk_context is not None:
+                    recipe_scanner = bulk_context.recipe_scanner
+                    hash_index = await bulk_context.get_recipe_hash_index()
+                    defer_maintenance = True
+                else:
+                    recipe_scanner = await self._recipe_scanner_factory()
+                    hash_index = None
+                    defer_maintenance = False
                 if recipe_scanner:
                     try:
-                        await recipe_scanner.update_lora_filename_by_hash(
-                            hash_value, new_file_name
+                        file_count, cache_count = (
+                            await recipe_scanner.update_lora_filename_by_hash(
+                                hash_value,
+                                new_file_name,
+                                hash_index=hash_index,
+                                defer_maintenance=defer_maintenance,
+                            )
                         )
+                        if bulk_context is not None and (file_count or cache_count):
+                            bulk_context.recipes_touched = True
                     except Exception as exc:  # pragma: no cover - defensive logging
                         logger.error(
                             "Error updating recipe references for %s: %s",
@@ -478,6 +550,34 @@ class ModelLifecycleService:
             "reload_required": False,
         }
 
+    def _rename_companion_files(
+        self,
+        existing_files: List[tuple[str, str]],
+        new_file_name: str,
+    ) -> tuple[List[str], Optional[str]]:
+        """Rename all companion files, off the event loop thread.
+
+        Runs the blocking ``os.rename`` sequence for one model in a worker
+        thread so a single file's HDD I/O does not stall the event loop.
+        Never parallelized across files: one model's renames stay sequential
+        and the helper holds no locks.
+        """
+        renamed_files: List[str] = []
+        new_metadata_path: Optional[str] = None
+
+        for old_path, pattern in existing_files:
+            ext = self._get_multipart_ext(pattern)
+            new_path = os.path.join(
+                os.path.dirname(old_path), f"{new_file_name}{ext}"
+            ).replace(os.sep, "/")
+            os.rename(old_path, new_path)
+            renamed_files.append(new_path)
+
+            if ext == ".metadata.json":
+                new_metadata_path = new_path
+
+        return renamed_files, new_metadata_path
+
     @staticmethod
     def _get_multipart_ext(filename: str) -> str:
         """Return the extension for files with compound suffixes."""

py/services/model_scanner.py

@@ -4,6 +4,7 @@ import logging
 import asyncio
 import time
 import shutil
+from contextlib import asynccontextmanager
 from dataclasses import dataclass
 from typing import Any, Awaitable, Callable, Dict, List, Mapping, Optional, Sequence, Set, Tuple, Type, Union, cast
 
@@ -162,6 +163,13 @@ class ModelScanner:
         self._name_display_mode = self._resolve_name_display_mode()
         self._cancel_requested = False  # Flag for cancellation
         self._move_locks: Dict[str, asyncio.Lock] = {}  # Per-source-file move locks
+        # Bulk-operation deferral: while _defer_persist_depth > 0,
+        # update_single_model_cache() skips the per-call resort/persist and
+        # only marks _deferred_persist_pending; the exit of the outermost
+        # defer_cache_persist() context finalizes once (see
+        # _finalize_deferred_cache_persist).
+        self._defer_persist_depth = 0
+        self._deferred_persist_pending = False
         self._autov3_backfill_scheduled = False  # One-time AutoV3 backfill trigger per process
         # Guard against concurrent all-folders backfill walks (cold fallback
         # for persisted snapshots that predate folder recording).
@@ -773,11 +781,11 @@ class ModelScanner:
         except Exception as exc:
             logger.warning("AutoV3 backfill failed: %s", exc)
 
-    async def _save_persistent_cache(self, scan_result: CacheBuildResult) -> None:
+    async def _save_persistent_cache(self, scan_result: CacheBuildResult, *, force: bool = False) -> None:
         if not scan_result or not getattr(self, '_persistent_cache', None):
             return
 
-        if self.is_cancelled():
+        if self.is_cancelled() and not force:
             logger.info(
                 f"{self.model_type.capitalize()} Scanner: Skipping _save_persistent_cache "
                 "after cancellation"
@@ -836,7 +844,7 @@ class ModelScanner:
                 bucket.append(path)
         return snapshot
 
-    async def _persist_current_cache(self) -> None:
+    async def _persist_current_cache(self, *, force: bool = False) -> None:
         if self._cache is None or not getattr(self, '_persistent_cache', None):
             return
 
@@ -851,7 +859,7 @@ class ModelScanner:
                 else None
             ),
         )
-        await self._save_persistent_cache(snapshot)
+        await self._save_persistent_cache(snapshot, force=force)
         await self._sync_download_history(snapshot.raw_data, source='scan')
     def _count_model_files(self) -> int:
         """Count all model files with supported extensions in all roots
@@ -2646,11 +2654,85 @@ class ModelScanner:
             logger.error(f"Error updating metadata paths: {e}", exc_info=True)
             return None
 
+    @asynccontextmanager
+    async def defer_cache_persist(self):
+        """Defer heavyweight cache maintenance for a bulk operation.
+
+        While at least one ``defer_cache_persist`` context is active,
+        :meth:`update_single_model_cache` performs only the in-memory entry
+        swap plus incremental index updates — it skips the full version-index
+        rebuild, the natsort resort, and the whole-table SQLite persist plus
+        download-history sync that normally run per call. When the outermost
+        context exits, the pending maintenance runs **once** (resort, persist,
+        download-history sync).
+
+        The final persist is forced: it runs even when the scanner's
+        cancellation flag is set or the wrapped block raised, because callers
+        use this around operations that already mutated files on disk and the
+        cache must not be left diverging from reality.
+
+        Intended for bulk rename/move loops (e.g. the filename-template "Apply
+        to Library" flow). Single-shot callers keep the immediate per-call
+        behavior by not entering this context.
+        """
+        self._defer_persist_depth = getattr(self, "_defer_persist_depth", 0) + 1
+        try:
+            yield
+        finally:
+            self._defer_persist_depth -= 1
+            if self._defer_persist_depth == 0:
+                await self._finalize_deferred_cache_persist()
+
+    @property
+    def _cache_persist_deferred(self) -> bool:
+        """True while cache resort/persist is deferred to a bulk finalize."""
+        return getattr(self, "_defer_persist_depth", 0) > 0
+
+    async def _finalize_deferred_cache_persist(self) -> None:
+        """Run the resort + persist deferred by ``defer_cache_persist``.
+
+        Best-effort: failures are logged, never raised, so an error here
+        cannot mask the outcome of the bulk operation itself (including
+        cancellation).
+        """
+        if not getattr(self, "_deferred_persist_pending", False):
+            return
+        self._deferred_persist_pending = False
+        if self._cache is None:
+            return
+        try:
+            # resort() rebuilds the version index and folder list, so the
+            # per-call rebuilds skipped during deferral are covered here.
+            await self._cache.resort()
+            await self._persist_current_cache(force=True)
+            self.bump_cache_version()
+        except Exception:
+            logger.error(
+                "%s Scanner: failed to finalize deferred cache persist",
+                self.model_type.capitalize(),
+                exc_info=True,
+            )
+
     async def update_single_model_cache(self, original_path: str, new_path: str, metadata: Optional[Dict[str, Any]], recalculate_type: bool = False) -> Union[bool, Dict[str, Any]]:
-        """Update cache after a model has been moved or modified"""
+        """Update cache after a model has been moved or modified.
+
+        Performs the full maintenance chain (version-index rebuild, resort,
+        whole-table persist, download-history sync) unless the scanner is
+        inside a :meth:`defer_cache_persist` context, in which case only
+        the in-memory entry swap and incremental index updates run and the
+        heavy chain executes once at context exit.
+        """
+        deferred = self._cache_persist_deferred
         cache = await self.get_cached_data()
 
-        existing_item = next((item for item in cache.raw_data if item['file_path'] == original_path), None)
+        existing_index: Optional[int] = None
+        existing_item = None
+        for idx, item in enumerate(cache.raw_data):
+            if item['file_path'] == original_path:
+                existing_item = item
+                existing_index = idx
+                break
+
         if existing_item:
             cache.remove_from_version_index(existing_item)
 
@@ -2662,11 +2744,18 @@ class ModelScanner:
                         del self._tags_count[tag]
         
         self._hash_index.remove_by_path(original_path)
-        
-        cache.raw_data = [
-            item for item in cache.raw_data
-            if item['file_path'] != original_path
-        ]
+
+        if deferred:
+            # In-place swap avoids the O(n) list rebuild per renamed file;
+            # indexes were already updated incrementally above/below, and the
+            # folder recompute happens in the single finalize resort().
+            if existing_index is not None:
+                cache.raw_data.pop(existing_index)
+        else:
+            cache.raw_data = [
+                item for item in cache.raw_data
+                if item['file_path'] != original_path
+            ]
 
         cache_modified = bool(existing_item) or bool(metadata)
         cache_entry: Optional[Dict[str, Any]] = None
@@ -2707,8 +2796,11 @@ class ModelScanner:
                     cache_entry.get('autov3') or None,
                 )
 
-            all_folders = set(item['folder'] for item in cache.raw_data)
-            cache.folders = sorted(list(all_folders), key=lambda x: x.lower())
+            if not deferred:
+                # O(n) over raw_data; the finalize resort() recomputes the
+                # folder list once, so bulk callers skip it per file.
+                all_folders = set(item['folder'] for item in cache.raw_data)
+                cache.folders = sorted(list(all_folders), key=lambda x: x.lower())
 
             # The move target may live in directories the last scan never saw;
             # record the destination folder (and its parents) in the known
@@ -2723,13 +2815,18 @@ class ModelScanner:
             for tag in cache_entry.get('tags', []):
                 self._tags_count[tag] = self._tags_count.get(tag, 0) + 1
 
-        cache.rebuild_version_index()
+        if deferred:
+            if cache_modified:
+                self._deferred_persist_pending = True
+                self.bump_cache_version()
+        else:
+            cache.rebuild_version_index()
 
-        await cache.resort()
+            await cache.resort()
 
-        if cache_modified:
-            await self._persist_current_cache()
-            self.bump_cache_version()
+            if cache_modified:
+                await self._persist_current_cache()
+                self.bump_cache_version()
 
         if metadata and cache_entry is not None:
             return cache_entry

py/services/recipe_scanner.py

@@ -4580,14 +4580,64 @@ class RecipeScanner:
 
         return syntax_parts
 
+    async def build_lora_hash_index(self) -> Dict[str, List[Dict[str, Any]]]:
+        """Build a one-shot lowercase-LoRA-hash → recipes index.
+
+        Scans the recipe cache exactly once (O(recipes × loras)) and returns
+        a mapping of lowercase lora ``hash`` to the list of recipe dicts
+        containing it. Bulk rename loops pass this index to
+        :meth:`update_lora_filename_by_hash` so per-file lookups are O(1)
+        instead of rescanning every recipe for each renamed LoRA.
+        """
+        cache = await self.get_cached_data()
+        index: Dict[str, List[Dict[str, Any]]] = {}
+        if not cache or not cache.raw_data:
+            return index
+
+        for recipe in cache.raw_data:
+            loras = recipe.get("loras", [])
+            if not isinstance(loras, list):
+                continue
+            for lora in loras:
+                if not isinstance(lora, dict):
+                    continue
+                hash_value = (lora.get("hash") or "").lower()
+                if hash_value:
+                    index.setdefault(hash_value, []).append(recipe)
+        return index
+
+    async def finalize_bulk_filename_updates(self) -> None:
+        """Run once after a bulk rename session that deferred maintenance.
+
+        Refreshes folder metadata and schedules a single re-sort. Filename-only
+        renames never change recipe folders, so the deferred refresh is
+        redundant but cheap; skipping it per file is what makes bulk renames
+        O(1)-per-file.
+        """
+        if self._cache is None:
+            return
+        self._schedule_resort()
+
     async def update_lora_filename_by_hash(
-        self, hash_value: str, new_file_name: str
+        self,
+        hash_value: str,
+        new_file_name: str,
+        *,
+        hash_index: Optional[Dict[str, List[Dict[str, Any]]]] = None,
+        defer_maintenance: bool = False,
     ) -> Tuple[int, int]:
         """Update file_name in all recipes that contain a LoRA with the specified hash.
 
         Args:
             hash_value: The SHA256 hash value of the LoRA
             new_file_name: The new file_name to set
+            hash_index: Optional prebuilt index from
+                :meth:`build_lora_hash_index`. When given, the O(recipes)
+                cache scan (and its folder-metadata walk) is skipped and the
+                affected recipes are looked up directly — the bulk rename path.
+            defer_maintenance: When True, skip the folder-metadata refresh and
+                resort scheduling. The caller MUST run
+                :meth:`finalize_bulk_filename_updates` exactly once afterwards.
 
         Returns:
             Tuple[int, int]: (number of recipes updated in files, number of recipes updated in cache)
@@ -4598,17 +4648,21 @@ class RecipeScanner:
         # Always use lowercase hash for consistency
         hash_value = hash_value.lower()
 
-        # Get cache
-        cache = await self.get_cached_data()
-        if not cache or not cache.raw_data:
-            return 0, 0
+        if hash_index is not None:
+            candidate_recipes = hash_index.get(hash_value, [])
+        else:
+            # Get cache
+            cache = await self.get_cached_data()
+            if not cache or not cache.raw_data:
+                return 0, 0
+            candidate_recipes = cache.raw_data
 
         file_updated_count = 0
         cache_updated_count = 0
 
-        # Find recipes that need updating from the cache
+        # Find recipes that need updating
         recipes_to_update = []
-        for recipe in cache.raw_data:
+        for recipe in candidate_recipes:
             loras = recipe.get("loras", [])
             if not isinstance(loras, list):
                 continue
@@ -4654,7 +4708,9 @@ class RecipeScanner:
         # We don't necessarily need to resort because LoRA file_name isn't a sort key,
         # but we might want to schedule a resort if we're paranoid or if searching relies on sorted state.
         # Given it's a rename of a dependency, search results might change if searching by LoRA name.
-        self._schedule_resort()
+        # Bulk callers defer this to a single finalize_bulk_filename_updates() call.
+        if not defer_maintenance:
+            self._schedule_resort()
 
         return file_updated_count, cache_updated_count
 

py/services/use_cases/filename_template_use_case.py

@@ -33,7 +33,9 @@ class FilenameTemplateUseCase:
     An empty template restores the recorded original filename instead of
     rendering a template. Shares the auto-organize lock (and its in-progress
     error) so a bulk rename never runs concurrently with an auto-organize
-    operation.
+    operation. The whole loop runs inside a bulk rename session so cache
+    persist/resort and recipe maintenance happen once at the end instead of
+    per renamed file.
     """
 
     def __init__(
@@ -106,23 +108,24 @@ class FilenameTemplateUseCase:
 
             await self._emit_progress(progress_callback, result, "started")
 
-            for index in range(0, result.total, AUTO_ORGANIZE_BATCH_SIZE):
-                if self._scanner.is_cancelled():
-                    logger.info(
-                        "Filename template apply cancelled for %s", self._model_type
-                    )
-                    break
-
-                batch = models[index : index + AUTO_ORGANIZE_BATCH_SIZE]
-                for model in batch:
+            async with self._lifecycle_service.bulk_rename_session() as bulk_context:
+                for index in range(0, result.total, AUTO_ORGANIZE_BATCH_SIZE):
                     if self._scanner.is_cancelled():
+                        logger.info(
+                            "Filename template apply cancelled for %s", self._model_type
+                        )
                         break
-                    await self._process_model(model, template, result)
-                    result.processed += 1
 
-                await self._emit_progress(progress_callback, result, "processing")
-                # Yield between batches so the server stays responsive.
-                await asyncio.sleep(0.1)
+                    batch = models[index : index + AUTO_ORGANIZE_BATCH_SIZE]
+                    for model in batch:
+                        if self._scanner.is_cancelled():
+                            break
+                        await self._process_model(model, template, result, bulk_context)
+                        result.processed += 1
+
+                    await self._emit_progress(progress_callback, result, "processing")
+                    # Yield between batches so the server stays responsive.
+                    await asyncio.sleep(0.1)
 
             if self._scanner.is_cancelled():
                 result.status = "cancelled"
@@ -150,6 +153,7 @@ class FilenameTemplateUseCase:
         model: Dict[str, Any],
         template: str,
         result: AutoOrganizeResult,
+        bulk_context: Any = None,
     ) -> None:
         model_name = model.get("model_name", "Unknown")
         try:
@@ -177,7 +181,7 @@ class FilenameTemplateUseCase:
                 return
 
             await self._lifecycle_service.rename_model(
-                file_path=file_path, new_file_name=new_stem
+                file_path=file_path, new_file_name=new_stem, bulk_context=bulk_context
             )
             result.success_count += 1
 

tests/conftest.py

@@ -3,9 +3,10 @@ import importlib.util
 import inspect
 import sys
 import types
+from contextlib import asynccontextmanager
 from dataclasses import dataclass, field
 from pathlib import Path
-from typing import Any, Dict, List, Optional, Sequence
+from typing import Any, AsyncIterator, Dict, List, Optional, Sequence
 from unittest import mock
 
 import pytest
@@ -177,6 +178,10 @@ class MockScanner:
     def reset_cancellation(self) -> None:
         self._cancelled = False
 
+    @asynccontextmanager
+    async def defer_cache_persist(self) -> AsyncIterator[None]:
+        yield None
+
     async def get_cached_data(self, force_refresh: bool = False):
         return self._cache
 

tests/services/test_bulk_rename_apply.py

@@ -0,0 +1,639 @@
+"""Performance-regression tests for the bulk filename-template rename path.
+
+The bulk "Apply to Library" rename must be O(1)-per-file: the heavyweight
+cache persist/resort chain and the recipe maintenance run exactly once per
+bulk operation (even on cancellation or mid-loop errors), while single-shot
+renames keep their immediate per-call behavior.
+"""
+
+from __future__ import annotations
+
+import asyncio
+import json
+from contextlib import asynccontextmanager
+from pathlib import Path
+from typing import Any, AsyncIterator, Dict, List, Optional
+
+import pytest
+
+from py.services import model_scanner as model_scanner_module
+from py.services.model_cache import ModelCache
+from py.services.model_lifecycle_service import ModelLifecycleService
+from py.services.model_scanner import ModelScanner
+from py.services.settings_manager import get_settings_manager
+from py.services.use_cases.filename_template_use_case import FilenameTemplateUseCase
+from py.utils.models import BaseModelMetadata
+
+
+class BulkDummyScanner(ModelScanner):
+    """Minimal concrete scanner for cache-behavior tests."""
+
+    def __init__(self, root: Path):
+        self._root = str(root)
+        super().__init__(
+            model_type="dummy",
+            model_class=BaseModelMetadata,
+            file_extensions={".txt"},
+        )
+
+    def get_model_roots(self) -> List[str]:
+        return [self._root]
+
+
+@pytest.fixture(autouse=True)
+def _reset_model_scanner_singletons():
+    ModelScanner._instances.clear()
+    ModelScanner._locks.clear()
+    yield
+    ModelScanner._instances.clear()
+    ModelScanner._locks.clear()
+
+
+@pytest.fixture(autouse=True)
+def _disable_persistent_cache_env(monkeypatch: pytest.MonkeyPatch):
+    monkeypatch.setenv("LORA_MANAGER_DISABLE_PERSISTENT_CACHE", "1")
+
+
+@pytest.fixture(autouse=True)
+def _stub_register_service(monkeypatch: pytest.MonkeyPatch):
+    async def noop(*_args: Any, **_kwargs: Any) -> None:
+        return None
+
+    monkeypatch.setattr(
+        model_scanner_module.ServiceRegistry, "register_service", noop
+    )
+
+
+def _metadata(stem: str, path: str, sha256: str, civitai_id: int = 1) -> Dict[str, Any]:
+    return {
+        "file_name": stem,
+        "file_path": path,
+        "model_name": stem,
+        "sha256": sha256,
+        "folder": "",
+        "tags": [],
+        "size": 1,
+        "modified": 1.0,
+        "civitai": {"id": civitai_id, "modelId": 2},
+    }
+
+
+class _SpiedCache:
+    """Scanner cache with call counters for resort/persist/sync."""
+
+    def __init__(self, scanner: BulkDummyScanner, entries: List[Dict[str, Any]]):
+        self.cache = ModelCache(raw_data=[dict(e) for e in entries], folders=[])
+        scanner._cache = self.cache
+        self.resort_calls = 0
+        self.save_calls: List[bool] = []
+        self.sync_calls = 0
+        self._original_resort = self.cache.resort
+
+    async def resort(self) -> None:
+        self.resort_calls += 1
+        await self._original_resort()
+
+    async def fake_save(self, scan_result: Any, *, force: bool = False) -> None:
+        self.save_calls.append(force)
+
+    async def fake_sync(self, raw_data: Any, *, source: str) -> None:
+        self.sync_calls += 1
+
+
+async def _make_spied_scanner(
+    tmp_path: Path, monkeypatch: pytest.MonkeyPatch, entries: List[Dict[str, Any]]
+) -> tuple[BulkDummyScanner, _SpiedCache]:
+    scanner = BulkDummyScanner(tmp_path)
+    spied = _SpiedCache(scanner, entries)
+    # Truthy stand-in so _persist_current_cache() does not early-return.
+    scanner._persistent_cache = object()
+    monkeypatch.setattr(scanner, "_save_persistent_cache", spied.fake_save)
+    monkeypatch.setattr(scanner, "_sync_download_history", spied.fake_sync)
+    # Flush the resort task scheduled by ModelCache.__post_init__ before
+    # installing the counting wrapper.
+    await asyncio.sleep(0)
+    await asyncio.sleep(0)
+    monkeypatch.setattr(spied.cache, "resort", spied.resort)
+    return scanner, spied
+
+
+# ---------------------------------------------------------------------------
+# ModelScanner deferred persist
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.asyncio
+async def test_update_single_model_cache_persists_immediately_by_default(
+    tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+):
+    path_a = (tmp_path / "a.txt").as_posix()
+    scanner, spied = await _make_spied_scanner(
+        tmp_path, monkeypatch, [_metadata("a", path_a, "hash-a")]
+    )
+
+    await scanner.update_single_model_cache(
+        path_a, path_a, _metadata("a", path_a, "hash-a")
+    )
+
+    assert spied.save_calls == [False]
+    assert spied.sync_calls == 1
+    assert spied.resort_calls == 1
+
+
+@pytest.mark.asyncio
+async def test_deferred_cache_persist_persists_once_for_many_updates(
+    tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+):
+    entries = []
+    for index, stem in enumerate(("a", "b", "c")):
+        path = (tmp_path / f"{stem}.txt").as_posix()
+        entries.append(_metadata(stem, path, f"hash-{stem}", civitai_id=index + 1))
+    scanner, spied = await _make_spied_scanner(tmp_path, monkeypatch, entries)
+
+    async with scanner.defer_cache_persist():
+        for index, stem in enumerate(("a", "b", "c")):
+            old_path = (tmp_path / f"{stem}.txt").as_posix()
+            new_path = (tmp_path / f"{stem}-renamed.txt").as_posix()
+            await scanner.update_single_model_cache(
+                old_path,
+                new_path,
+                _metadata(
+                    f"{stem}-renamed", new_path, f"hash-{stem}", civitai_id=index + 1
+                ),
+            )
+        # Nothing heavy may run mid-loop.
+        assert spied.save_calls == []
+        assert spied.sync_calls == 0
+        assert spied.resort_calls == 0
+
+    # Exactly one heavyweight finalize for the whole bulk operation.
+    assert spied.save_calls == [True]
+    assert spied.sync_calls == 1
+    assert spied.resort_calls == 1
+
+    # In-memory state is correct, including the incremental version index.
+    cache = await scanner.get_cached_data()
+    cached_paths = {item["file_path"] for item in cache.raw_data}
+    for stem in ("a", "b", "c"):
+        assert (tmp_path / f"{stem}.txt").as_posix() not in cached_paths
+        assert (tmp_path / f"{stem}-renamed.txt").as_posix() in cached_paths
+    assert cache.version_index[1]["file_path"].endswith("a-renamed.txt")
+    assert scanner._hash_index.get_path("hash-a").endswith("a-renamed.txt")
+
+
+@pytest.mark.asyncio
+async def test_deferred_cache_persist_finalizes_despite_cancellation(
+    tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+):
+    path_a = (tmp_path / "a.txt").as_posix()
+    scanner, spied = await _make_spied_scanner(
+        tmp_path, monkeypatch, [_metadata("a", path_a, "hash-a")]
+    )
+
+    scanner.cancel_task()
+    async with scanner.defer_cache_persist():
+        new_path = (tmp_path / "a-renamed.txt").as_posix()
+        await scanner.update_single_model_cache(
+            path_a, new_path, _metadata("a-renamed", new_path, "hash-a")
+        )
+
+    # Files are already renamed on disk, so the persist must be forced even
+    # though the cancellation flag is set.
+    assert spied.save_calls == [True]
+    assert spied.sync_calls == 1
+    assert spied.resort_calls == 1
+
+
+@pytest.mark.asyncio
+async def test_deferred_cache_persist_finalizes_despite_mid_loop_error(
+    tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+):
+    path_a = (tmp_path / "a.txt").as_posix()
+    scanner, spied = await _make_spied_scanner(
+        tmp_path, monkeypatch, [_metadata("a", path_a, "hash-a")]
+    )
+
+    with pytest.raises(RuntimeError, match="boom"):
+        async with scanner.defer_cache_persist():
+            new_path = (tmp_path / "a-renamed.txt").as_posix()
+            await scanner.update_single_model_cache(
+                path_a, new_path, _metadata("a-renamed", new_path, "hash-a")
+            )
+            raise RuntimeError("boom")
+
+    assert spied.save_calls == [True]
+    assert spied.sync_calls == 1
+
+
+@pytest.mark.asyncio
+async def test_deferred_cache_persist_nested_contexts_finalize_once(
+    tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+):
+    path_a = (tmp_path / "a.txt").as_posix()
+    scanner, spied = await _make_spied_scanner(
+        tmp_path, monkeypatch, [_metadata("a", path_a, "hash-a")]
+    )
+
+    async with scanner.defer_cache_persist():
+        async with scanner.defer_cache_persist():
+            new_path = (tmp_path / "a-renamed.txt").as_posix()
+            await scanner.update_single_model_cache(
+                path_a, new_path, _metadata("a-renamed", new_path, "hash-a")
+            )
+
+    assert spied.save_calls == [True]
+    assert spied.resort_calls == 1
+
+
+@pytest.mark.asyncio
+async def test_deferred_cache_persist_without_updates_persists_nothing(
+    tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+):
+    scanner, spied = await _make_spied_scanner(tmp_path, monkeypatch, [])
+
+    async with scanner.defer_cache_persist():
+        pass
+
+    assert spied.save_calls == []
+    assert spied.resort_calls == 0
+
+
+# ---------------------------------------------------------------------------
+# ModelLifecycleService bulk rename session
+# ---------------------------------------------------------------------------
+
+
+class _SessionScanner:
+    model_type = "lora"
+
+    def __init__(self, root: Path):
+        self._root = str(root)
+        self.cache_updates: List[tuple[str, str]] = []
+        self.defer_enters = 0
+        self.defer_exits = 0
+
+    def get_model_roots(self) -> List[str]:
+        return [self._root]
+
+    @asynccontextmanager
+    async def defer_cache_persist(self) -> AsyncIterator[None]:
+        self.defer_enters += 1
+        try:
+            yield
+        finally:
+            self.defer_exits += 1
+
+    async def update_single_model_cache(
+        self, old_path: str, new_path: str, metadata: Dict[str, Any]
+    ) -> bool:
+        self.cache_updates.append((old_path, new_path))
+        return True
+
+
+class _PassthroughMetadataManager:
+    def __init__(self) -> None:
+        self.saved: List[str] = []
+
+    async def save_metadata(self, path: str, metadata: Dict[str, Any]) -> bool:
+        self.saved.append(path)
+        return True
+
+
+class RecordingRecipeScanner:
+    """Records bulk-mode recipe calls and applies filename changes."""
+
+    def __init__(self) -> None:
+        self.index_builds = 0
+        self.updates: List[Dict[str, Any]] = []
+        self.finalizes = 0
+        self._recipes: Dict[str, Dict[str, Any]] = {}
+
+    def add_recipe(self, recipe_id: str, lora_hash: str, file_name: str) -> None:
+        self._recipes[recipe_id] = {
+            "id": recipe_id,
+            "loras": [{"hash": lora_hash, "file_name": file_name}],
+        }
+
+    async def build_lora_hash_index(self) -> Dict[str, List[Dict[str, Any]]]:
+        self.index_builds += 1
+        index: Dict[str, List[Dict[str, Any]]] = {}
+        for recipe in self._recipes.values():
+            for lora in recipe["loras"]:
+                index.setdefault(lora["hash"].lower(), []).append(recipe)
+        return index
+
+    async def update_lora_filename_by_hash(
+        self,
+        hash_value: str,
+        new_file_name: str,
+        *,
+        hash_index: Optional[Dict[str, List[Dict[str, Any]]]] = None,
+        defer_maintenance: bool = False,
+    ) -> tuple[int, int]:
+        self.updates.append(
+            {
+                "hash_value": hash_value,
+                "new_file_name": new_file_name,
+                "hash_index": hash_index,
+                "defer_maintenance": defer_maintenance,
+            }
+        )
+        matched = 0
+        for recipe in (self._recipes.values() if hash_index is None else hash_index.get(hash_value.lower(), [])):
+            for lora in recipe["loras"]:
+                if lora["hash"].lower() == hash_value.lower():
+                    lora["file_name"] = new_file_name
+                    matched += 1
+        return (matched, matched)
+
+    async def finalize_bulk_filename_updates(self) -> None:
+        self.finalizes += 1
+
+
+def _write_model_with_sidecar(
+    root: Path, stem: str, sha256: str, model_name: Optional[str] = None
+) -> str:
+    model_path = root / f"{stem}.safetensors"
+    model_path.write_bytes(b"model")
+    (root / f"{stem}.metadata.json").write_text(
+        json.dumps(
+            {
+                "file_name": stem,
+                "file_path": model_path.as_posix(),
+                "model_name": model_name or stem,
+                "sha256": sha256,
+            }
+        )
+    )
+    return model_path.as_posix()
+
+
+def _make_lifecycle_service(
+    scanner: _SessionScanner, recipe_scanner: RecordingRecipeScanner
+) -> ModelLifecycleService:
+    async def metadata_loader(path: str) -> Dict[str, Any]:
+        with open(path, "r", encoding="utf-8") as handle:
+            return json.load(handle)
+
+    async def recipe_scanner_factory() -> RecordingRecipeScanner:
+        return recipe_scanner
+
+    return ModelLifecycleService(
+        scanner=scanner,  # pyright: ignore[reportArgumentType]
+        metadata_manager=_PassthroughMetadataManager(),  # pyright: ignore[reportArgumentType]
+        metadata_loader=metadata_loader,
+        recipe_scanner_factory=recipe_scanner_factory,
+    )
+
+
+@pytest.mark.asyncio
+async def test_bulk_rename_session_defers_and_finalizes_once(tmp_path: Path):
+    scanner = _SessionScanner(tmp_path)
+    recipe_scanner = RecordingRecipeScanner()
+    recipe_scanner.add_recipe("r1", "aa" * 32, "old_a")
+    recipe_scanner.add_recipe("r2", "bb" * 32, "old_b")
+    service = _make_lifecycle_service(scanner, recipe_scanner)
+
+    path_a = _write_model_with_sidecar(tmp_path, "model-a", "aa" * 32)
+    path_b = _write_model_with_sidecar(tmp_path, "model-b", "bb" * 32)
+
+    async with service.bulk_rename_session() as bulk_context:
+        await service.rename_model(
+            file_path=path_a, new_file_name="renamed-a", bulk_context=bulk_context
+        )
+        await service.rename_model(
+            file_path=path_b, new_file_name="renamed-b", bulk_context=bulk_context
+        )
+        # No finalize may run before the session ends.
+        assert recipe_scanner.finalizes == 0
+        assert scanner.defer_enters == 1
+        assert scanner.defer_exits == 0
+
+    assert scanner.defer_exits == 1
+    # Hash index built at most once for the whole session.
+    assert recipe_scanner.index_builds == 1
+    assert recipe_scanner.finalizes == 1
+    assert len(recipe_scanner.updates) == 2
+    for update in recipe_scanner.updates:
+        assert update["hash_index"] is not None
+        assert update["defer_maintenance"] is True
+
+    # Recipes were re-pointed.
+    assert recipe_scanner._recipes["r1"]["loras"][0]["file_name"] == "renamed-a"
+    assert recipe_scanner._recipes["r2"]["loras"][0]["file_name"] == "renamed-b"
+
+
+@pytest.mark.asyncio
+async def test_bulk_rename_session_finalizes_recipes_despite_error(tmp_path: Path):
+    scanner = _SessionScanner(tmp_path)
+    recipe_scanner = RecordingRecipeScanner()
+    recipe_scanner.add_recipe("r1", "aa" * 32, "old_a")
+    service = _make_lifecycle_service(scanner, recipe_scanner)
+
+    path_a = _write_model_with_sidecar(tmp_path, "model-a", "aa" * 32)
+
+    with pytest.raises(RuntimeError, match="mid-loop"):
+        async with service.bulk_rename_session() as bulk_context:
+            await service.rename_model(
+                file_path=path_a, new_file_name="renamed-a", bulk_context=bulk_context
+            )
+            raise RuntimeError("mid-loop")
+
+    assert recipe_scanner.finalizes == 1
+    assert scanner.defer_exits == 1
+
+
+@pytest.mark.asyncio
+async def test_bulk_rename_session_skips_recipe_finalize_when_untouched(tmp_path: Path):
+    scanner = _SessionScanner(tmp_path)
+    recipe_scanner = RecordingRecipeScanner()
+    recipe_scanner.add_recipe("r1", "cc" * 32, "old_c")
+    service = _make_lifecycle_service(scanner, recipe_scanner)
+
+    # Model whose hash matches no recipe: the lookup runs but no recipe is
+    # touched, so no recipe maintenance is needed at finalize.
+    path_a = _write_model_with_sidecar(tmp_path, "model-a", "aa" * 32)
+
+    async with service.bulk_rename_session() as bulk_context:
+        await service.rename_model(
+            file_path=path_a, new_file_name="renamed-a", bulk_context=bulk_context
+        )
+
+    assert len(recipe_scanner.updates) == 1
+    assert recipe_scanner.updates[0]["defer_maintenance"] is True
+    assert recipe_scanner.finalizes == 0
+    assert recipe_scanner._recipes["r1"]["loras"][0]["file_name"] == "old_c"
+
+
+@pytest.mark.asyncio
+async def test_single_rename_keeps_immediate_recipe_behavior(tmp_path: Path):
+    scanner = _SessionScanner(tmp_path)
+    recipe_scanner = RecordingRecipeScanner()
+    recipe_scanner.add_recipe("r1", "aa" * 32, "old_a")
+    service = _make_lifecycle_service(scanner, recipe_scanner)
+
+    path_a = _write_model_with_sidecar(tmp_path, "model-a", "aa" * 32)
+
+    await service.rename_model(file_path=path_a, new_file_name="renamed-a")
+
+    assert len(recipe_scanner.updates) == 1
+    update = recipe_scanner.updates[0]
+    assert update["hash_index"] is None
+    assert update["defer_maintenance"] is False
+    assert recipe_scanner.finalizes == 0
+    assert recipe_scanner._recipes["r1"]["loras"][0]["file_name"] == "renamed-a"
+
+
+# ---------------------------------------------------------------------------
+# Use case end-to-end: exactly one heavyweight persist for the whole loop
+# ---------------------------------------------------------------------------
+
+
+class _UseCaseScanner(ModelScanner):
+    def __init__(self, root: Path):
+        self._root = str(root)
+        super().__init__(
+            model_type="lora",
+            model_class=BaseModelMetadata,
+            file_extensions={".safetensors"},
+        )
+
+    def get_model_roots(self) -> List[str]:
+        return [self._root]
+
+
+class _UseCaseLockProvider:
+    def __init__(self) -> None:
+        self._lock = asyncio.Lock()
+
+    def is_auto_organize_running(self) -> bool:
+        return False
+
+    async def get_auto_organize_lock(self) -> asyncio.Lock:
+        return self._lock
+
+
+def _set_filename_template(template: str, model_type: str = "lora") -> None:
+    manager = get_settings_manager()
+    templates = dict(manager.settings.get("download_filename_templates") or {})
+    templates[model_type] = template
+    manager.settings["download_filename_templates"] = templates
+
+
+@pytest.mark.asyncio
+async def test_filename_template_bulk_apply_persists_and_repoints_once(
+    tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+):
+    _set_filename_template("{model_name}")
+
+    entries = []
+    for index, (stem, model_name, sha) in enumerate(
+        (("model-a", "alpha", "aa" * 32), ("model-b", "beta", "bb" * 32))
+    ):
+        path = _write_model_with_sidecar(tmp_path, stem, sha, model_name=model_name)
+        entries.append(_metadata(model_name, path, sha, civitai_id=index + 1))
+
+    scanner = _UseCaseScanner(tmp_path)
+    spied = _SpiedCache(scanner, entries)
+    scanner._persistent_cache = object()
+    monkeypatch.setattr(scanner, "_save_persistent_cache", spied.fake_save)
+    monkeypatch.setattr(scanner, "_sync_download_history", spied.fake_sync)
+    await asyncio.sleep(0)
+    await asyncio.sleep(0)
+    monkeypatch.setattr(spied.cache, "resort", spied.resort)
+
+    recipe_scanner = RecordingRecipeScanner()
+    recipe_scanner.add_recipe("r1", "aa" * 32, "alpha")
+    recipe_scanner.add_recipe("r2", "bb" * 32, "beta")
+
+    service = _make_lifecycle_service(  # type: ignore[arg-type]
+        scanner, recipe_scanner  # pyright: ignore[reportArgumentType]
+    )
+    use_case = FilenameTemplateUseCase(
+        scanner=scanner,
+        lifecycle_service=service,
+        lock_provider=_UseCaseLockProvider(),
+        model_type="lora",
+    )
+
+    result = await use_case.execute(progress_callback=None)
+
+    assert result.status == "success"
+    assert result.success_count == 2
+
+    # One heavyweight persist/resort for the entire bulk operation, not two.
+    assert spied.save_calls == [True]
+    assert spied.sync_calls == 1
+    assert spied.resort_calls == 1
+
+    # Recipes re-pointed via a single lazily built hash index, maintenance
+    # finalized once.
+    assert recipe_scanner.index_builds == 1
+    assert recipe_scanner.finalizes == 1
+    assert recipe_scanner._recipes["r1"]["loras"][0]["file_name"] == "alpha"
+    assert recipe_scanner._recipes["r2"]["loras"][0]["file_name"] == "beta"
+
+    # Cache reflects the new paths.
+    cache = await scanner.get_cached_data()
+    cached_paths = {item["file_path"] for item in cache.raw_data}
+    assert (tmp_path / "alpha.safetensors").as_posix() in cached_paths
+    assert (tmp_path / "beta.safetensors").as_posix() in cached_paths
+
+    # Sidecars moved alongside the model files.
+    assert (tmp_path / "alpha.metadata.json").exists()
+    assert (tmp_path / "beta.metadata.json").exists()
+
+
+@pytest.mark.asyncio
+async def test_filename_template_bulk_apply_finalizes_persist_on_cancellation(
+    tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+):
+    _set_filename_template("{model_name}")
+
+    entries = []
+    paths = []
+    for index, (stem, model_name, sha) in enumerate(
+        (("model-a", "alpha", "aa" * 32), ("model-b", "beta", "bb" * 32))
+    ):
+        path = _write_model_with_sidecar(tmp_path, stem, sha, model_name=model_name)
+        paths.append(path)
+        entries.append(_metadata(model_name, path, sha, civitai_id=index + 1))
+
+    scanner = _UseCaseScanner(tmp_path)
+    spied = _SpiedCache(scanner, entries)
+    scanner._persistent_cache = object()
+    monkeypatch.setattr(scanner, "_save_persistent_cache", spied.fake_save)
+    monkeypatch.setattr(scanner, "_sync_download_history", spied.fake_sync)
+    await asyncio.sleep(0)
+    await asyncio.sleep(0)
+    monkeypatch.setattr(spied.cache, "resort", spied.resort)
+
+    service = _make_lifecycle_service(
+        scanner,  # pyright: ignore[reportArgumentType]
+        RecordingRecipeScanner(),
+    )
+
+    original_rename = service.rename_model
+
+    async def cancelling_rename(**kwargs: Any) -> Dict[str, object]:
+        result = await original_rename(**kwargs)
+        scanner.cancel_task()
+        return result
+
+    monkeypatch.setattr(service, "rename_model", cancelling_rename)
+
+    use_case = FilenameTemplateUseCase(
+        scanner=scanner,
+        lifecycle_service=service,
+        lock_provider=_UseCaseLockProvider(),
+        model_type="lora",
+    )
+
+    result = await use_case.execute(progress_callback=None)
+
+    assert result.status == "cancelled"
+    # The one file renamed before cancellation is on disk; the cache must
+    # still be persisted exactly once.
+    assert spied.save_calls == [True]
+    assert spied.sync_calls == 1
+    assert spied.resort_calls == 1

tests/services/test_recipe_scanner.py

@@ -1903,6 +1903,115 @@ async def test_update_lora_filename_by_hash_updates_affected_recipes(
     assert cached1["loras"][0]["file_name"] == new_name
 
 
+@pytest.mark.asyncio
+async def test_update_lora_filename_by_hash_bulk_mode_skips_scan_and_defers_resort(
+    tmp_path: Path, recipe_scanner, monkeypatch: pytest.MonkeyPatch
+):
+    """Bulk rename path: prebuilt hash index, no per-call cache walk/resort."""
+    scanner, _ = recipe_scanner
+    recipes_dir = Path(config.loras_roots[0]) / "recipes"
+    recipes_dir.mkdir(parents=True, exist_ok=True)
+
+    recipe1_id = "recipe-bulk-1"
+    recipe1_path = recipes_dir / f"{recipe1_id}.recipe.json"
+    recipe1_data = {
+        "id": recipe1_id,
+        "file_path": str(tmp_path / "bulk1.png"),
+        "title": "Bulk 1",
+        "modified": 0.0,
+        "created_date": 0.0,
+        "loras": [{"file_name": "old_name", "hash": "hash1"}],
+    }
+    recipe1_path.write_text(json.dumps(recipe1_data))
+    await scanner.add_recipe(dict(recipe1_data))
+
+    # Build the index once (O(recipes)), as a bulk rename session does.
+    hash_index = await scanner.build_lora_hash_index()
+    assert "hash1" in hash_index
+
+    # Spies: the bulk call must not walk the recipe cache again, and must not
+    # schedule a resort per call.
+    get_cached_calls = 0
+    original_get_cached = scanner.get_cached_data
+
+    async def counting_get_cached_data(*args, **kwargs):
+        nonlocal get_cached_calls
+        get_cached_calls += 1
+        return await original_get_cached(*args, **kwargs)
+
+    monkeypatch.setattr(scanner, "get_cached_data", counting_get_cached_data)
+
+    resort_calls = 0
+    original_schedule = scanner._schedule_resort
+
+    def counting_schedule_resort(**kwargs):
+        nonlocal resort_calls
+        resort_calls += 1
+        original_schedule(**kwargs)
+
+    monkeypatch.setattr(scanner, "_schedule_resort", counting_schedule_resort)
+
+    file_count, cache_count = await scanner.update_lora_filename_by_hash(
+        "HASH1", "new_name", hash_index=hash_index, defer_maintenance=True
+    )
+
+    assert (file_count, cache_count) == (1, 1)
+    assert get_cached_calls == 0
+    assert resort_calls == 0
+
+    # The per-match recipe JSON rewrite must still happen.
+    persisted1 = json.loads(recipe1_path.read_text())
+    assert persisted1["loras"][0]["file_name"] == "new_name"
+    cached1 = next(r for r in hash_index["hash1"] if r["id"] == recipe1_id)
+    assert cached1["loras"][0]["file_name"] == "new_name"
+
+    # Deferred maintenance runs exactly once at finalize.
+    await scanner.finalize_bulk_filename_updates()
+    await asyncio.sleep(0)
+    assert resort_calls == 1
+
+
+@pytest.mark.asyncio
+async def test_update_lora_filename_by_hash_with_index_still_resorts_when_not_deferred(
+    tmp_path: Path, recipe_scanner, monkeypatch: pytest.MonkeyPatch
+):
+    """hash_index without defer_maintenance: O(1) lookup, immediate resort."""
+    scanner, _ = recipe_scanner
+    recipes_dir = Path(config.loras_roots[0]) / "recipes"
+    recipes_dir.mkdir(parents=True, exist_ok=True)
+
+    recipe1_id = "recipe-idx-1"
+    recipe1_path = recipes_dir / f"{recipe1_id}.recipe.json"
+    recipe1_data = {
+        "id": recipe1_id,
+        "file_path": str(tmp_path / "idx1.png"),
+        "title": "Index 1",
+        "modified": 0.0,
+        "created_date": 0.0,
+        "loras": [{"file_name": "old_name", "hash": "hash1"}],
+    }
+    recipe1_path.write_text(json.dumps(recipe1_data))
+    await scanner.add_recipe(dict(recipe1_data))
+
+    hash_index = await scanner.build_lora_hash_index()
+
+    resort_calls = 0
+    original_schedule = scanner._schedule_resort
+
+    def counting_schedule_resort(**kwargs):
+        nonlocal resort_calls
+        resort_calls += 1
+        original_schedule(**kwargs)
+
+    monkeypatch.setattr(scanner, "_schedule_resort", counting_schedule_resort)
+
+    file_count, cache_count = await scanner.update_lora_filename_by_hash(
+        "hash1", "new_name", hash_index=hash_index
+    )
+
+    assert (file_count, cache_count) == (1, 1)
+    assert resort_calls == 1
+
 @pytest.mark.asyncio
 async def test_get_paginated_data_filters_by_favorite(recipe_scanner):
     scanner, _ = recipe_scanner

tests/services/test_use_cases.py

@@ -1,8 +1,9 @@
 import asyncio
 import logging
+from contextlib import asynccontextmanager
 from dataclasses import dataclass
 from types import SimpleNamespace
-from typing import Any, Dict, List, Optional
+from typing import Any, AsyncIterator, Dict, List, Optional
 
 import pytest
 
@@ -547,7 +548,13 @@ class StubLifecycleService:
         self.cancel_on_rename = False
         self._scanner = scanner
 
-    async def rename_model(self, *, file_path: str, new_file_name: str) -> Dict[str, Any]:
+    @asynccontextmanager
+    async def bulk_rename_session(self) -> AsyncIterator[None]:
+        yield None
+
+    async def rename_model(
+        self, *, file_path: str, new_file_name: str, bulk_context: Any = None
+    ) -> Dict[str, Any]:
         if self.error is not None:
             raise self.error
         self.renames.append({"file_path": file_path, "new_file_name": new_file_name})