Before you start
Open a GPU instance you own. Some provider log paths can be temporarily unavailable while provisioning or after stopping.
Steps
- Open the instance’s Container logs section and enable live following where available.
- Watch startup messages for image initialization, model downloads, ready URLs, and the application port.
- Use pause or scrollback to read earlier output without losing your place. Copy only a short redacted excerpt when asking for help.
- Match a failure to its cause: insufficient disk, insufficient VRAM, missing credentials, image pull error, or an invalid startup command.
- Correct the relevant configuration and restart only when needed. A restart interrupts running jobs; do not repeatedly restart a long model download.
Check the result
The log shows a ready application and the Connections endpoint responds. Normal application silence does not necessarily mean log streaming failed.
Troubleshooting
Check the instance status first if no logs appear. Custom applications must write useful output to stdout or stderr to appear in container logs; application log files may need to be read inside the container.