os.replace vs. tempfile.NamedTemporaryFile for idempotent file writes

Ensuring Write Atomicity During Retries

When implementing retry logic for file-based data persistence in Python 3.11+, the goal is to prevent partial writes or duplicated data if a process is interrupted and restarted. Two common patterns involve the use of os.replace() and tempfile.NamedTemporaryFile.

Using os.replace() provides an atomic replacement of the target file on POSIX and Windows systems. However, integrating this with tempfile.NamedTemporaryFile(delete=False) introduces a requirement for manual cleanup of the temporary file if the operation fails before the final replacement occurs.

The trade-off centers on the risk of leaving orphaned temporary files versus the risk of non-atomic writes on specific filesystems where os.replace() might not be fully guaranteed.

• Does os.replace() maintain atomicity across network-attached storage (NAS) environments?
• Is there a documented method to ensure NamedTemporaryFile cleanup without risking the deletion of a successfully replaced target file?