feat(metadata): skip models on offline roots during bulk fetch (#1108)

Bulk metadata refresh picked candidates purely from cache state, so the
entries the scoped scan deliberately keeps for unreachable drives were
fetched anyway — burning CivitAI rate limit on sidecar/preview writes
that cannot land.

- Filter the fetch queue by root reachability (live check via
  describe_model_roots), covering both startup-offline and mid-session
  drive-off; probing failures degrade to no filtering
- Report offline_skipped / offline_roots in the WS terminal payloads,
  the HTTP result and the completion message
- Show a "Drive Offline" stat card in the fetch summary modal when
  any model was skipped this way

locales/de.json

@@ -2130,6 +2130,7 @@
             "statSuccess": "Erfolgreich",
             "statFailed": "Fehlgeschlagen",
             "statSkipped": "Übersprungen",
+            "statOffline": "[TODO: Translate] Drive Offline",
             "statTotal": "Gesamt geprüft",
             "statDuration": "Dauer",
             "successMessage": "Alle {count} {type}s erfolgreich aktualisiert!",

locales/en.json

@@ -2130,6 +2130,7 @@
       "statSuccess": "Success",
       "statFailed": "Failed",
       "statSkipped": "Skipped",
+      "statOffline": "Drive Offline",
       "statTotal": "Total Scanned",
       "statDuration": "Duration",
       "successMessage": "All {count} {type}s updated successfully!",

locales/es.json

@@ -2130,6 +2130,7 @@
             "statSuccess": "Éxito",
             "statFailed": "Fallido",
             "statSkipped": "Omitido",
+            "statOffline": "[TODO: Translate] Drive Offline",
             "statTotal": "Total escaneado",
             "statDuration": "Duración",
             "successMessage": "¡Todos los {count} {type}s actualizados correctamente!",

locales/fr.json

@@ -2130,6 +2130,7 @@
             "statSuccess": "Réussi",
             "statFailed": "Échoué",
             "statSkipped": "Ignoré",
+            "statOffline": "[TODO: Translate] Drive Offline",
             "statTotal": "Total scanné",
             "statDuration": "Durée",
             "successMessage": "Tous les {count} {type}s mis à jour avec succès !",

locales/he.json

@@ -2130,6 +2130,7 @@
             "statSuccess": "הצלחה",
             "statFailed": "נכשל",
             "statSkipped": "דולג",
+            "statOffline": "[TODO: Translate] Drive Offline",
             "statTotal": "סה\"כ נסרק",
             "statDuration": "משך",
             "successMessage": "כל {count} {type}s עודכנו בהצלחה!",

locales/ja.json

@@ -2130,6 +2130,7 @@
             "statSuccess": "成功",
             "statFailed": "失敗",
             "statSkipped": "スキップ",
+            "statOffline": "[TODO: Translate] Drive Offline",
             "statTotal": "スキャン合計",
             "statDuration": "所要時間",
             "successMessage": "すべての{count}件の{type}を正常に更新しました",

locales/ko.json

@@ -2130,6 +2130,7 @@
             "statSuccess": "성공",
             "statFailed": "실패",
             "statSkipped": "건너뜀",
+            "statOffline": "[TODO: Translate] Drive Offline",
             "statTotal": "총 스캔",
             "statDuration": "소요 시간",
             "successMessage": "모든 {count}개 {type}이(가) 성공적으로 업데이트되었습니다",

locales/ru.json

@@ -2130,6 +2130,7 @@
             "statSuccess": "Успешно",
             "statFailed": "Ошибка",
             "statSkipped": "Пропущено",
+            "statOffline": "[TODO: Translate] Drive Offline",
             "statTotal": "Всего проверено",
             "statDuration": "Длительность",
             "successMessage": "Все {count} {type}s успешно обновлены",

locales/zh-CN.json

@@ -2130,6 +2130,7 @@
             "statSuccess": "成功",
             "statFailed": "失败",
             "statSkipped": "已跳过",
+            "statOffline": "[TODO: Translate] Drive Offline",
             "statTotal": "总计扫描",
             "statDuration": "耗时",
             "successMessage": "全部 {count} 个 {type} 更新成功！",

locales/zh-TW.json

@@ -2130,6 +2130,7 @@
             "statSuccess": "成功",
             "statFailed": "失敗",
             "statSkipped": "已跳過",
+            "statOffline": "[TODO: Translate] Drive Offline",
             "statTotal": "總計掃描",
             "statDuration": "耗時",
             "successMessage": "全部 {count} 個 {type} 更新成功！",

py/services/use_cases/bulk_metadata_refresh_use_case.py

@@ -7,6 +7,7 @@ import time
 from typing import Any, Dict, List, Optional, Protocol, Sequence
 
 from ..metadata_sync_service import MetadataSyncService
+from ..model_scanner import _normalized_root_prefix, _path_matches_prefix
 from ..model_sources import has_external_source
 from ...utils.metadata_manager import MetadataManager
 
@@ -46,7 +47,7 @@ class BulkMetadataRefreshUseCase:
 
         enable_metadata_archive_db = self._settings.get("enable_metadata_archive_db", False)
         skip_paths = self._settings.get("metadata_refresh_skip_paths", [])
-        to_process: Sequence[Dict[str, Any]] = [
+        eligible: Sequence[Dict[str, Any]] = [
             model
             for model in cache.raw_data
             if not model.get("skip_metadata_refresh", False)
@@ -68,6 +69,26 @@ class BulkMetadataRefreshUseCase:
             )
         ]
 
+        # Models on a drive that cannot be read right now stay in the cache
+        # (scoped scan keeps them), but fetching their metadata would only
+        # spend CivitAI rate limit on sidecar/preview writes that cannot land.
+        offline_roots = self._offline_root_details()
+        if offline_roots:
+            to_process = [
+                model
+                for model in eligible
+                if not self._is_under_offline_root(model.get("file_path", ""), offline_roots)
+            ]
+        else:
+            to_process = list(eligible)
+        offline_skipped = len(eligible) - len(to_process)
+        if offline_skipped:
+            self._logger.info(
+                "Bulk metadata refresh: %d model(s) skipped, drive offline: %s",
+                offline_skipped,
+                ", ".join(root["label"] for root in offline_roots),
+            )
+
         total_to_process = len(to_process)
         initial_skipped = total_models - total_to_process  # models excluded from fetch queue
         processed = 0
@@ -90,14 +111,21 @@ class BulkMetadataRefreshUseCase:
                 "success": success,
                 "failure_count": len(failures),
                 "skipped_count": skipped_count,
+                "offline_skipped": offline_skipped,
                 "handled": handled_count,
                 "elapsed_seconds": int(time.monotonic() - start_time),
             }
             # Only include full failure details in terminal emits (completed,
             # cancelled, rate_limited) to avoid serializing the list on every
             # per-model progress update.
-            if failures and status in ("completed", "cancelled", "rate_limited"):
+            if status in ("completed", "cancelled", "rate_limited"):
-                payload["failures"] = failures
+                if failures:
+                    payload["failures"] = failures
+                if offline_roots:
+                    payload["offline_roots"] = [
+                        {"path": root["path"], "label": root["label"]}
+                        for root in offline_roots
+                    ]
             payload.update(extra)
             await progress_callback.on_progress(payload)
 
@@ -110,7 +138,7 @@ class BulkMetadataRefreshUseCase:
             if self._service.scanner.is_cancelled():
                 self._logger.info("Bulk metadata refresh cancelled by user")
                 await emit("cancelled", processed=processed, success=success)
-                return {"success": False, "message": "Operation cancelled", "processed": processed, "updated": success, "total": total_models, "failures": failures, "failure_count": len(failures), "skipped_count": skipped_count, "elapsed_seconds": int(time.monotonic() - start_time)}
+                return {"success": False, "message": "Operation cancelled", "processed": processed, "updated": success, "total": total_models, "failures": failures, "failure_count": len(failures), "skipped_count": skipped_count, "offline_skipped": offline_skipped, "elapsed_seconds": int(time.monotonic() - start_time)}
             try:
                 original_name = model.get("model_name")
 
@@ -202,6 +230,7 @@ class BulkMetadataRefreshUseCase:
                         "failures": failures,
                         "failure_count": len(failures),
                         "skipped_count": skipped_count,
+                        "offline_skipped": offline_skipped,
                         "elapsed_seconds": int(time.monotonic() - start_time),
                     }
 
@@ -237,8 +266,58 @@ class BulkMetadataRefreshUseCase:
             "Successfully updated "
             f"{success} of {processed} processed {self._service.model_type}s (total: {total_models})"
         )
+        if offline_skipped:
+            message += (
+                f"; {offline_skipped} skipped, drive offline ("
+                + ", ".join(root["label"] for root in offline_roots)
+                + ")"
+            )
 
-        return {"success": True, "message": message, "processed": processed, "updated": success, "total": total_models, "failures": failures, "failure_count": len(failures), "skipped_count": skipped_count, "elapsed_seconds": int(time.monotonic() - start_time)}
+        return {"success": True, "message": message, "processed": processed, "updated": success, "total": total_models, "failures": failures, "failure_count": len(failures), "skipped_count": skipped_count, "offline_skipped": offline_skipped, "elapsed_seconds": int(time.monotonic() - start_time)}
+
+    def _offline_root_details(self) -> List[Dict[str, str]]:
+        """Configured roots whose directory cannot be read right now.
+
+        Uses the scanner's root description (live ``os.path.exists`` per root),
+        so both a drive switched off at startup and one unplugged mid-session
+        are covered. Root probing must never break a fetch: any failure yields
+        an empty list, which disables the offline filter.
+        """
+        describe = getattr(self._service.scanner, "describe_model_roots", None)
+        if describe is None:
+            return []
+        try:
+            details = describe()
+        except Exception as exc:  # pragma: no cover - defensive logging
+            self._logger.debug("Failed to describe model roots: %s", exc)
+            return []
+
+        offline: List[Dict[str, str]] = []
+        for detail in details:
+            if not isinstance(detail, dict) or detail.get("reachable", True):
+                continue
+            path = detail.get("path") or ""
+            if not path:
+                continue
+            offline.append(
+                {
+                    "path": path,
+                    "label": detail.get("label") or path,
+                    "prefix": _normalized_root_prefix(path),
+                }
+            )
+        return offline
+
+    @staticmethod
+    def _is_under_offline_root(
+        file_path: str, offline_roots: Sequence[Dict[str, str]]
+    ) -> bool:
+        """Match a cached business path against offline root prefixes."""
+        if not file_path:
+            return False
+        return any(
+            _path_matches_prefix(file_path, root["prefix"]) for root in offline_roots
+        )
 
     @staticmethod
     def _is_in_skip_path(folder: str, skip_paths: List[str]) -> bool:

static/css/components/metadata-refresh-result.css

@@ -55,6 +55,10 @@
     border-left-color: var(--color-warning);
 }
 
+.stat-card-offline {
+    border-left-color: var(--text-secondary);
+}
+
 .stat-card-total {
     border-left-color: var(--lora-border);
 }

static/js/api/baseModelApi.js

@@ -898,7 +898,9 @@ export class BaseModelApiClient {
     }
 
     _showFailureDetailsModal(data) {
-        const { failures = [], success, processed, total, failure_count, skipped_count, elapsed_seconds } = data;
+        const { failures = [], success, processed, total, failure_count, skipped_count, elapsed_seconds, offline_skipped = 0, offline_roots = [] } = data;
+
+        const offlineRootNames = offline_roots.map((root) => root.label || root.path || String(root)).join(', ');
 
         // Build failure list HTML
         const failureRows = failures.map((f, i) =>
@@ -935,6 +937,13 @@ export class BaseModelApiClient {
                                 <span class="stat-card-value">${skipped_count}</span>
                             </div>
                         </div>
+                        ${offline_skipped > 0 ? `
+                        <div class="stat-card stat-card-offline" title="${this._escapeHtml(offlineRootNames)}">
+                            <div class="stat-card-body">
+                                <span class="stat-card-label">${translate('modals.metadataFetchSummary.statOffline', {}, 'Drive Offline')}</span>
+                                <span class="stat-card-value">${offline_skipped}</span>
+                            </div>
+                        </div>` : ''}
                         <div class="stat-card stat-card-total">
                             <div class="stat-card-body">
                                 <span class="stat-card-label">${translate('modals.metadataFetchSummary.statTotal', {}, 'Total Scanned')}</span>

tests/services/use_cases/test_bulk_metadata_refresh_use_case.py

@@ -402,3 +402,138 @@ async def test_model_without_hash_skipped(use_case, mock_service, mock_metadata_
 
     assert result["processed"] == 1
     assert result["updated"] == 0
+
+
+def _offline_scanner_mock(scanner):
+    """Point the scanner mock at one offline and one online root."""
+    scanner.describe_model_roots = MagicMock(
+        return_value=[
+            {"path": "/mnt/offline_loras", "label": "offline_loras", "reachable": False, "available": False, "models": 1},
+            {"path": "/models", "label": "models", "reachable": True, "available": True, "models": 1},
+        ]
+    )
+
+
+@pytest.mark.asyncio
+@patch.object(metadata_manager.MetadataManager, "hydrate_model_data")
+async def test_fetch_skips_models_on_offline_root(mock_hydrate, use_case, mock_service, mock_metadata_sync):
+    """Models under an unreachable root are excluded from the fetch queue."""
+    mock_hydrate.return_value = None
+    _offline_scanner_mock(mock_service.scanner)
+
+    offline_model = {
+        "file_path": "/mnt/offline_loras/model.safetensors",
+        "sha256": "hash_offline",
+        "model_name": "Offline Model",
+        "civitai": {},
+        "from_civitai": False,
+        "civitai_deleted": False,
+    }
+    online_model = {
+        "file_path": "/models/model.safetensors",
+        "sha256": "hash_online",
+        "model_name": "Online Model",
+        "civitai": {},
+        "from_civitai": False,
+        "civitai_deleted": False,
+    }
+
+    cache = SimpleNamespace(raw_data=[offline_model, online_model], resort=AsyncMock())
+    mock_service.scanner.get_cached_data.return_value = cache
+
+    reporter = MockProgressReporter()
+    result = await use_case.execute(progress_callback=reporter)
+
+    # Only the online model is fetched
+    mock_metadata_sync.fetch_and_update_model.assert_called_once()
+    call_args = mock_metadata_sync.fetch_and_update_model.call_args[1]
+    assert call_args["file_path"] == "/models/model.safetensors"
+
+    assert result["success"] is True
+    assert result["processed"] == 1
+    assert result["offline_skipped"] == 1
+    # The offline model still counts as excluded from the queue
+    assert result["skipped_count"] == 1
+
+    completed_calls = [c for c in reporter.progress_calls if c["status"] == "completed"]
+    assert completed_calls[0]["offline_skipped"] == 1
+    assert completed_calls[0]["offline_roots"] == [
+        {"path": "/mnt/offline_loras", "label": "offline_loras"}
+    ]
+
+
+@pytest.mark.asyncio
+@patch.object(metadata_manager.MetadataManager, "hydrate_model_data")
+async def test_fetch_offline_filter_only_catches_offline_prefix(mock_hydrate, use_case, mock_service, mock_metadata_sync):
+    """A sibling path that merely shares a prefix string is not excluded."""
+    mock_hydrate.return_value = None
+    _offline_scanner_mock(mock_service.scanner)
+
+    # "/mnt/offline_loras_backup/..." must NOT match the "/mnt/offline_loras/" prefix
+    sibling_model = {
+        "file_path": "/mnt/offline_loras_backup/model.safetensors",
+        "sha256": "hash_sibling",
+        "model_name": "Sibling Model",
+        "civitai": {},
+        "from_civitai": False,
+        "civitai_deleted": False,
+    }
+
+    cache = SimpleNamespace(raw_data=[sibling_model], resort=AsyncMock())
+    mock_service.scanner.get_cached_data.return_value = cache
+
+    result = await use_case.execute()
+
+    mock_metadata_sync.fetch_and_update_model.assert_called_once()
+    assert result["offline_skipped"] == 0
+
+
+@pytest.mark.asyncio
+@patch.object(metadata_manager.MetadataManager, "hydrate_model_data")
+async def test_fetch_proceeds_unfiltered_without_root_description(mock_hydrate, use_case, mock_service, mock_metadata_sync):
+    """Scanners without describe_model_roots keep the previous behaviour."""
+    mock_hydrate.return_value = None
+    del mock_service.scanner.describe_model_roots
+
+    model = {
+        "file_path": "/mnt/offline_loras/model.safetensors",
+        "sha256": "hash123",
+        "model_name": "Test Model",
+        "civitai": {},
+        "from_civitai": False,
+        "civitai_deleted": False,
+    }
+
+    cache = SimpleNamespace(raw_data=[model], resort=AsyncMock())
+    mock_service.scanner.get_cached_data.return_value = cache
+
+    result = await use_case.execute()
+
+    mock_metadata_sync.fetch_and_update_model.assert_called_once()
+    assert result["offline_skipped"] == 0
+
+
+@pytest.mark.asyncio
+@patch.object(metadata_manager.MetadataManager, "hydrate_model_data")
+async def test_fetch_proceeds_unfiltered_when_root_description_fails(mock_hydrate, use_case, mock_service, mock_metadata_sync):
+    """A failing describe_model_roots must not break the fetch."""
+    mock_hydrate.return_value = None
+    mock_service.scanner.describe_model_roots = MagicMock(side_effect=RuntimeError("boom"))
+
+    model = {
+        "file_path": "/models/model.safetensors",
+        "sha256": "hash123",
+        "model_name": "Test Model",
+        "civitai": {},
+        "from_civitai": False,
+        "civitai_deleted": False,
+    }
+
+    cache = SimpleNamespace(raw_data=[model], resort=AsyncMock())
+    mock_service.scanner.get_cached_data.return_value = cache
+
+    result = await use_case.execute()
+
+    mock_metadata_sync.fetch_and_update_model.assert_called_once()
+    assert result["success"] is True
+    assert result["offline_skipped"] == 0