Skip to content

FF chevron

FAQ: Connection Issues

This page provides answers to some frequently asked questions relating to logging into or otherwise connecting to the UMD HPC clusters. Please reviewthese questions before asking HPC staff for assistance.

I cannot connect to the system. What is wrong?

Troubleshooting HPC Login Issues:

  • Verify Login Node: Ensure you're connecting to the correct login node. Specifically, for the Zaratan cluster, omitting "login" from the hostname will result in a failed connection.
  • Account Status: Login issues may arise from expired university affiliations, such as might occur due to graduation, end of appointment, or lapsed affiliate status (which requires annual renewal).
  • Contact Support: If the above doesn't resolve the issue, contact support with the following details:
    • Host being accessed.
    • Exact command or application settings.
    • Error messages.
    • Username (not password).
    • Time of failed login attempts (as precisely as possible).
    • IP address (obtainable from http://noc.net.umd.edu/cgi-bin/netmgr/whoami

I cannot transfer files. What gives?

  • Sensitivity to Output: scp and sftp are highly sensitive to extra output from your shell initialization scripts (like .bashrc, .profile).
  • Troubleshooting:
    • If you can ssh into the server, check for any unexpected errors or output.
    • Normal output should include only:
      • "Unauthorized access..." warning.
      • "Last login message."
      • "DISPLAY is..." message.
      • Any other output likely interferes with scp and sftp.
  • Solution: Edit your initialization scripts to remove the extra output.
  • Additional Help: Refer to the section on suppressing output from dot files for further guidance.

I get noVNC and/or connection errors when trying to launch an interactive application in the OnDemand portal

  • Browser Compatibility: VNC connections within the OnDemand portal, used for GUI displays, are browser-sensitive. Please provide your OS name and version and browser name and version when reporting issues to the HPC team.
  • Chrome on Windows Issues: Chrome on Windows often encounters problems. Recommendation: Windows users should use Firefox for reliable interactive application access in the OnDemand portal.
  • Interference from conda initialization code: If you have initialization code for conda running in your .bashrc or similar dot files, Recommendation: Keep your conda initialization code in a distinct file, separate from your dot files.

I am getting warnings about keys and fingerprints when I try to ssh. Should I be concerned?

  • SSH Security: SSH uses key verification to protect against security threats.
  • Two Key Warnings:
    • Cannot verify the key: This is often normal, especially on first login.
    • Key changed: This could indicate a security attack.
  • Verification:

How do I change my password?

  • UMD Directory ID: Zaratan uses your standard UMD campus directory ID and password.
  • Password Change:
  • University-Wide Change: Changing your password using these methods will change it across all UMD systems due to a shared authentication system.

I forgot my password. What can I do?

Since the Zaratan cluster uses your standard UMD campus directory ID and password, password resets are handled the same way as for this campus-wide password. Information for resetting this password can be found in the password reset knowledge base article.

Warning

The aforementioned procedures will reset your password on ALL university systems, as they are all part of one common authentication system.

When I try to login, I get an error like 'Resource temporarily unavailable'. What should I do?

The following text explains the "Resource temporarily unavailable" error encountered when logging into or running commands on a cluster's login nodes.

  • Cause:
    • The error indicates too many processes and/or threads running, exceeding the per-user limit imposed to preventauser's overloading the login node.
    • Stale processes from misbehaving programs, especially VS Code, are a common culprit.
    • Running resource-intensive commands directly on login nodes can also contribute to the problem.
  • Solutions: -For existing logins: - Use ps -vf to identify your processes and kill <PID> to terminate problematic processe. Replace <PID> with the process identifier given in the first (PID) column of the ps output. - Use pkill -u $USER to kill all user processes (logs you out). - Do these steps on all login nodes (e.g. login-1, login-2, and login-3).
    • For login failures:
  • Best Practices:
    • Avoid VS Code and similar programs that are known to leave stale processes around.
    • Run resource-intensive tasks as batch jobs or on interactive desktops via the OnDemand portal.
    • If you insist on using VS Code or similar misbehaving programs, frequently and regularly check all of the login nodes for stale processes they might have left behind and kill them. This will only provide temporary relief, so it must be repeated frequently and regularly

When I try to login, I get an error like 'shell request failed'. What should I do?

This error is another way in which the too many processes and/or threads issue can manifest itself. As with that FAQ, the issue is likely due to the use of vscode or similar applications --- we generally recommend not to use vscode.

Why was I advised not to use vscode on the HPC systems?

  • VS Code's Problem:
  • Cluster Policy:
    • The cluster's process limits are in place to prevent misuse, not due to cluster issues.
    • While not strictly prohibited, VS Code usage is strongly discouraged.
    • Users should rely on standard SSH and/or SCP for login and file transfer.
  • User Responsibility:
    • If users insist on using VS Code, they must regularly monitor and terminate stale processes.
    • Failure to do so will result in temporary lockouts and require system staff intervention.
    • System staff intervention for this issue is not considered an emergency.
  • Mitigation:

When I try to connect to one of the portal web sites (OnDemand, ColdFront, etc), I get connection errors

  • The various HPC portals all require you to be on the campus GlobalProtect VPN. By HPC portals, we include:
  • If you are already logged into VPN, please try refreshing your VPN connection. Sometimes, the VPN times out without visible indication, and refreshing will restore your VPN connection.
  • If you are trying to connect from a Mac system, please note that an update to the GlobalProtect VPN in August 2026 had an issue with handling IPv6 addresses in some cases. The KB article KB0020829 provides a simple workaround command. DIT managed Macs should have had this workaround applied automatically.