Troubleshooting

If you are running into errors that are not included in this page or in the rest of the documentation, please email hpc@umass.edu or join the Unity User Community Slack to chat with the Unity team and fellow Unity users.

Connect to Unity

When I connect over SSH I get a message saying permission denied (public key)

Ensure the following:

• You provided your private key while connecting. Use ssh -i <private_key_location> <user>@unity.rc.umass.edu
• You are assigned to at least one PI group. We require at least one PI to endorse your account before you can use Unity. Request to join a PI on the My PIs page if needed.
• You have added a public key to your Unity account via the Account Settings page. For more information about public and private keys, see our guide on SSH Connection.
• Your login shell is valid. In Account Settings, try setting it to “/bin/bash” or “/bin/zsh”.
• If you are a PI, do not use your PI group name as your login username (your login username should not start with pi_).

I am a Windows user using PowerShell SSH and am getting an error message when I try to connect to Unity via SSH.

When connecting, if you see Corrupted MAC on input, it is due to a known bug in Microsoft’s SSH library. This bug causes problems connecting to newer OpenSSH installs from Microsoft’s SSH client included in PowerShell.

To work around this, Windows users using PowerShell SSH must add MACs hmac-sha2-512-etm@openssh.com to their ~/.ssh/config file, or use the flag option in your SSH command so that it reads ssh -m hmac-sha2-512-etm@openssh.com <rest of the cmd>.

Job Management and File Storage

I got an error that said “disk quota exceeded.” What should I do?

Move or delete some files to reduce your storage usage. For more information, see our documentation on disk quota management.

When I try to queue a job, I get denied for MaxCpuPerAccount.

Resource limits are set per lab, with 1000 concurrent CPUs and 64 concurrent GPUs shared across your entire PI group. If you encounter this error, adjust your usage or discuss resource allocations with your PI or the Unity team.

I have jobs in the queue, but I need to prioritize a specific job now. Can I do this without canceling the queued jobs?

You can de-prioritize your other queued jobs with scontrol update jobid=.... nice=100.

If it’s a short job, you can try the Quality of Service (QOS) to “jump the line” with a higher priority for a small, short job. (This does not always mean no wait, just a shorter wait, and depends on resource availability.) See the QOS announcement for details.

I have too many jobs to run, and slurm won’t let me submit them all. I get an error for MaxJobCount or MaxJobSubmit. How can I work around this?

Slurm has limits that are set to avoid either too many jobs being submitted at once or too many jobs running at once. To work around this, submit jobs as an array job instead.