ZeroGPU Space stuck in APP_STARTING after Git push — Factory reboot doesn't help

Hi Hugging Face team,

My ZeroGPU Gradio Space was working perfectly until recently, both locally and on Hugging Face.

After encountering some issues and pushing updated code via Git, the Space became stuck indefinitely in APP_STARTING.

Space URL: Seb Full Avatar - a Hugging Face Space by seb2oo

Current symptoms:

  • Runtime status remains APP_STARTING indefinitely.

  • The build appears to complete, but the application never starts.

  • Container logs only display:

    ===== Application Startup at 2026-10-27 00:19:23 =====
    
  • None of my Python startup logs appear, even though I have print statements at the very beginning of my application.

  • Restarting the Space does not resolve the issue.

  • Factory reboot also fails to resolve the issue.

  • The application was previously running correctly on the same ZeroGPU hardware.

Runtime information:

  • SDK: Gradio

  • Hardware: zero-a10g

  • Runtime stage: APP_STARTING

  • Persistent bucket mounted at /data

I found a previous discussion describing almost exactly the same symptoms:

Since the application never reaches its first Python startup logs, I suspect an issue with the Hugging Face runtime, container scheduling, or deployment infrastructure rather than the application itself.

Could someone from the Hugging Face team please investigate the runtime state of my Space and check whether it needs to be reset on the infrastructure side?

Thank you very much for your help!

Your Space’s runtime already settles this: the API at huggingface.co/api/spaces/seb2oo/seb_full_avatar/runtime reports stage RUNTIME_ERROR with “Scheduling failure: unable to schedule” and hardware requested zero-a10g but current empty, so no container was ever allocated and your code never ran — which is exactly why no Python logs appear and why restarts and factory reboots cannot help. This needs a scheduling reset on the Hugging Face side, and quoting that stage, error, and hardware pair gives the team exactly what they need to act; re-checking the same endpoint will show it worked when current is no longer empty.

Thanks for investigating!

This strongly suggests that the issue is related to the Hugging Face scheduling infrastructure rather than my application code.

The runtime API reports:

  • Stage: RUNTIME_ERROR

  • Error: Scheduling failure: unable to schedule

  • Requested hardware: zero-a10g

  • Current hardware: empty / unallocated

Could someone from the Hugging Face team please reset the scheduling/runtime state of my Space?

Space: Seb Full Avatar - a Hugging Face Space by seb2oo

Thanks again for your help!

Hmm…? @hysts

I encountered the same situation.

My Space was working normally before. Today I restarted it several times without success, and the status it shows is exactly the same as the OP. Runtime API also shows Scheduling failure: unable to schedule with empty hardware.current.

We’re aware of the issue, and our infra team is working on it. Thanks for your patience.

The scheduling failure detail is the key clue here. Since the runtime reports no hardware allocated, and no Python startup logs appear, this does not look like a problem in the app code itself, which is why repeated restarts and a factory reboot would not fix it. While the infra team works on it, it may help to note whether other ZeroGPU Spaces show the same stage and error, and to try again once the scheduling issue is confirmed fixed rather than changing code that was previously working.

It’s fixed now. Thanks!