Version

7.4.2 (/app/src/__init__.py in container)

Deployment Method

Docker Compose

Database Type

PostgreSQL

Storage Setup

• Media on Docker volume using JuiceFS (S3 object storage backend)
• DEDUPLICATE_MEDIA=true
• VERIFY_MEDIA=true

Bug Description

VERIFY_MEDIA re-download repeatedly fails for a subset of media with this recurring pattern:

WARNING - Symlink failed, using direct path: [Errno 17] File exists: '../_shared/<name>' -> '/data/backups/media/<chat_id>/<name>'
ERROR   - Error downloading media: [Errno 2] No such file or directory: '/data/backups/media/_shared/<name>'
WARNING - Failed to re-download media for message <id>

This happens after backup completion during media verification/redownload.

What I observed on disk

For failed entries:

• Chat path exists as a dangling symlink:
  • /data/backups/media/-1002443412147/5956498299518196004_yCSXQMIrYQImig -> ../_shared/5956498299518196004_yCSXQMIrYQImig
• But _shared has only variant files with extensions/suffixes, e.g.:
  • 5956498299518196004_yCSXQMIrYQImig.mp4
  • 5956498299518196004_yCSXQMIrYQImig (1).mp4
  • 5956498299518196004_yCSXQMIrYQImig (2).mp4
• Exact _shared/<name> (without extension) is missing.

Current state in this environment:

• 33 dangling symlinks under media/-1002443412147
• All 33 have extension/suffixed candidates in _shared

Why this seems to happen (code-path analysis)

In src/telegram_backup.py:

1. _process_media checks if not os.path.exists(file_path) (around line 1342). For a dangling symlink this is True.
2. It attempts os.symlink(..., file_path) (line ~1362), which raises Errno 17 because the path entry already exists (even though dangling).
3. The fallback handles all OSError by doing shutil.move(shared_file_path, file_path) (line ~1368).
4. If Telethon saved to a different final filename (e.g. extension added / conflict suffix), shared_file_path no longer exists, causing Errno 2.

Also in _verify_and_redownload_media, cleanup uses os.path.exists(file_path) before os.remove (line ~584), so dangling symlinks are not removed before re-download.

Related issue

Likely related symptom family to #30, but this report includes a reproducible verification/redownload path and probable root cause.

Expected Behavior

VERIFY_MEDIA should repair missing/corrupt media without entering repeated Errno 17 + Errno 2 loops.

Suggested Fix (proposal)

1. Use os.path.lexists when deciding whether file_path already exists and before cleanup/removal in verify flow.
   • If lexists(file_path) and path is dangling symlink, unlink it before creating new symlink.
2. Treat EEXIST separately from symlink-unsupported errors:
   • for EEXIST, remove/recreate link if dangling;
   • do not immediately fall back to shutil.move.
3. Capture and use the actual return value of await self.client.download_media(...) as canonical downloaded path (Telethon may append extension or collision suffix).
4. Ensure DB file_name/file_path is updated to the canonical final filename/path actually written on disk.

If useful, I can open a PR with a concrete patch implementing the above.