Skip to content

Troubleshooting

The page does not load, or times out

  • The Studio takes 1 to 5 minutes to start after the instance shows Running.
  • Check the address starts with http://, not https://, and ends in :8501.
  • The security group may not allow port 8501 from your current address. If you restricted it to your IP and your IP has changed, update the rule, or use a tunnel.
  • The instance needs a public IP address. An instance in a private subnet has none: reach it with a Session Manager tunnel, or over a VPN or load balancer you already have.
  • Stopping and starting an instance gives it a new public IP. Copy the new one from the console, or attach an Elastic IP to keep it fixed.

The instance ID is refused

Check you copied the whole ID, starting with i-. If the sign-in screen says Enter the password for this instance instead of asking for the instance ID, someone has set a different password. Connect to the instance and run sudo sushi-ml-studio reset-password to go back to the instance ID.

I forgot the password I set

Same fix: sudo sushi-ml-studio reset-password.

The page says "Sign-in unavailable"

The Studio could not find a password to check against. Either the password file cannot be read or is damaged, or no password is set and the instance ID could not be read from the instance metadata service, for example because it is turned off. Run sudo sushi-ml-studio set-password to choose a password, which fixes every one of these.

EC2 Instance Connect fails in the browser

Check that port 22 is open to AWS's Instance Connect addresses, that the Username is ubuntu, and that the instance has a public IP address. See Connect to your instance.

I was asked to sign in again after refreshing

Expected. A refresh starts a new session. It also clears any model you have not downloaded, and your browser asks before it happens.

My file was refused

The Studio says why in a sentence, for example that the file has no rows, only one column, or is larger than 50 MB. If an Excel file cannot be read, open it in Excel and save it again as .xlsx.

Training is slow

Untick the slower models on the Train page, where each card shows its speed, or turn off Search for better settings. Each training run uses at most two processor cores, whatever the instance size, so a larger instance type helps mainly by adding memory for very large files.

Still stuck?

See Support.