Make HF backups atomic

deploy/huggingface/backup_manager.py

@@ -10,10 +10,12 @@ from pathlib import Path, PurePosixPath
 from typing import Iterable, Mapping
 
 try:
-    from huggingface_hub import HfApi, hf_hub_download
+    from huggingface_hub import CommitOperationAdd, CommitOperationDelete, HfApi, hf_hub_download
     from huggingface_hub.errors import EntryNotFoundError
     from huggingface_hub.utils import HfHubHTTPError, get_token
 except ImportError:  # pragma: no cover - only exercised in runtime images without the dependency
+    CommitOperationAdd = None  # type: ignore[assignment]
+    CommitOperationDelete = None  # type: ignore[assignment]
     HfApi = None  # type: ignore[assignment]
     EntryNotFoundError = Exception  # type: ignore[assignment]
     HfHubHTTPError = Exception  # type: ignore[assignment]
@@ -93,7 +95,7 @@ def data_root_has_state(data_root: Path) -> bool:
 
 
 def _require_hub_support() -> None:
-    if HfApi is None:
+    if HfApi is None or CommitOperationAdd is None or CommitOperationDelete is None:
         raise RuntimeError("huggingface_hub is required for Hugging Face backup operations")
 
 
@@ -134,31 +136,27 @@ def backup_to_dataset(
 
         api = HfApi(token=resolved_token)
         commit_suffix = f" for {space_name}" if space_name else ""
-        api.upload_file(
-            path_or_fileobj=str(archive_path),
-            path_in_repo=f"backups/{archive_path.name}",
+        repo_paths = _iter_repo_paths(api, repo_id, resolved_token)
+        operations = [
+            CommitOperationAdd(
+                path_in_repo=f"backups/{archive_path.name}",
+                path_or_fileobj=str(archive_path),
+            ),
+            CommitOperationAdd(
+                path_in_repo="backups/latest.tar.gz",
+                path_or_fileobj=str(archive_path),
+            ),
+        ]
+        for old_path in select_backups_to_delete(repo_paths, keep=keep):
+            operations.append(CommitOperationDelete(path_in_repo=old_path))
+
+        api.create_commit(
             repo_id=repo_id,
+            operations=operations,
             repo_type="dataset",
             token=resolved_token,
-            commit_message=f"Add backup {archive_path.name}{commit_suffix}",
+            commit_message=f"Update backup {archive_path.name}{commit_suffix}",
         )
-        api.upload_file(
-            path_or_fileobj=str(archive_path),
-            path_in_repo="backups/latest.tar.gz",
-            repo_id=repo_id,
-            repo_type="dataset",
-            token=resolved_token,
-            commit_message=f"Update latest backup{commit_suffix}",
-        )
-
-        for old_path in select_backups_to_delete(_iter_repo_paths(api, repo_id, resolved_token), keep=keep):
-            api.delete_file(
-                path_in_repo=old_path,
-                repo_id=repo_id,
-                repo_type="dataset",
-                token=resolved_token,
-                commit_message=f"Prune old backup {PurePosixPath(old_path).name}",
-            )
 
         return archive_path.name