Urgent: Persistent storage (/data) suddenly unavailable — vote DB lost after restart (channelcorp/KO-TTS-Arena)

Hi Hugging Face team and community,

We’re experiencing a critical production incident on our Space where Persistent Storage (/data) suddenly became unavailable, and as a result our SQLite vote database appears to be lost.

This Space is a Korean TTS Arena where users vote and compare models, so the vote history is essentially community-built data accumulated over time. Losing it is extremely serious for us.

Space

What changed?

As far as we know: nothing on our side.

  • We did not modify Space settings, secrets, storage configuration, or deploy code changes intentionally around the incident.

  • The app started behaving as if persistent storage was disabled/detached.

What we see in logs

On startup, the Space prints:

  • “Persistent storage (/data) not available. Using ‘instance/’ (data may be lost on restart)”

  • Then it tries to restore the DB from an HF dataset backup, but that now fails with 401 Unauthorized / Repository Not Found.

  • After that it attempts to create a new DB and crashes with sqlite3.OperationalError: unable to open database file.

Log excerpt (full logs available):

===== Application Startup at 2026-02-04 03:55:05 =====
Running in a Hugging Face Space 🤗
⚠️ Warning: Persistent storage (/data) not available. Using 'instance/' (data may be lost on restart)

Database not found at instance/tts_arena.db, downloading from HF dataset (channelcorp/ko-tts-arena-db)...
Error downloading database from HF dataset: 401 Client Error. (Request ID: Root=1-6982c32a-2d6a493653b785c766f63865;76865d47-0b52-4f3d-a901-766738eceb48)
Repository Not Found for url: https://huggingface.co/datasets/channelcorp/ko-tts-arena-db/resolve/main/tts_arena.db
Invalid username or password. ⚠️
A new database will be created.
...
sqlite3.OperationalError: unable to open database file

What we tried

  • SSH into the Space and searched the filesystem, but we could not find the old DB anywhere.

  • Since /data is not mounted, anything previously stored there is currently inaccessible from the running instance.

Questions / help needed (urgent)

  1. Why would Persistent Storage (/data) suddenly become unavailable for a Space without any setting change on our side?

  2. Is there any way to restore/recover the previous persistent volume (snapshots, backups, internal recovery, etc.)?

  3. Is this potentially related to billing/permissions, a platform incident, or a region/infrastructure issue?

  4. Separately: why would our dataset backup download suddenly return 401 / repo not found?

    • Could secrets/tokens have been reset, or did access permissions change?

If any HF staff can help escalate this, we’d really appreciate it — this dataset represents real user activity and we’re trying to avoid losing that history permanently.

Thank you.

1 Like

Hmm… Perhaps related to this…? @meganariley