Skip to main content

Connect and log in

Log in​

Your SSH key must be on the cluster's People tab.

  1. Open the cluster's page in the console.

  2. Copy the SSH command from the Overview tab.

  3. Run the command on your computer.

    ssh user1@203.0.113.10

If several people use the cluster, the console shows ssh <login-name>@<public-IP>. Replace <login-name> with your own login name.

If your key is not the default key on your computer, add -i:

ssh -i ~/.ssh/my_key user1@203.0.113.10

You land in your home folder, /home/jl_fs/<login-name>. In commands, ~ means your home folder. For example, for user1, ~/results is /home/jl_fs/user1/results. The welcome text at login has lines about updates. Ignore them.

What the login VM is for​

Use the login VM to edit files, submit jobs and read results. Run all real work in jobs.

The login VM does not have these tools: gcc, make, pip, mpicc and nvidia-smi. Also, python3 -m venv fails there. Some messages tell you to ask your administrator to install a tool. Ignore those messages, and run the command in a job.

Command 'nvidia-smi' not found ...
Ask your administrator to install one of them.

Compute nodes have gcc, make, mpicc and mpirun. They do not have pip, and python3 -m venv fails there too. Use uv. Refer to Software environments.

sudo is on every VM, but it asks for a password that you do not have. It is for admins, who do not need a password. docker also refuses you, with permission denied while trying to connect to the docker API. Use containers through srun. Refer to Containers.

All VMs use UTC. So --begin=16:00 means 16:00 UTC.

Copy files​

Run these commands on your computer. Add -i <key file> if your key is not the default.

To copy a file to your home folder:

scp train.py user1@203.0.113.10:~/

To copy a file back to your computer:

scp user1@203.0.113.10:~/results/model.pt .

rsync and sftp also work.

To download a large file to the shared filesystem, write in large pieces. This form is much faster than a plain download:

wget -O - <url> | dd of=/home/jl_fs/user1/data.tar bs=4M iflag=fullblock

Keep work alive after you disconnect​

A foreground srun ends when your connection drops. sbatch jobs do not. Use sbatch for long work.

tmux on the login VM keeps your session alive when your connection drops. A login VM restart still ends tmux and all srun commands that it holds. Refer to Restart a compute node or the login VM.

Connect to a compute node​

From the login VM, use the compute node's name:

ssh node-1

A user can do this only with a job on that compute node. Refer to SSH to compute nodes.

Use the name, not the IP address. An SSH connection to a private IP address fails with Host key verification failed.

A new compute node's name can take a short time to reach the login VM. Until then, you get Could not resolve hostname.

SSH rules for the login VM​

A security group controls who can reach the login VM's public IP address. VMs in the same VPC can still reach it on the private network. Jarvislabs creates this group for your cluster. Its name starts with your cluster's name and -login, and it starts with one inbound rule: SSH from anywhere.

You can change this group on the Security Groups page. For example, you can allow SSH from your office only. You can also add a rule for another port. Refer to Security Groups for the steps.

Caution:

A narrow SSH rule also applies to you. If your IP address changes, you cannot log in until you change the rule.

You cannot delete the group. If a change leaves no inbound rule, the console refuses it with this message:

This security group protects the login VM of your Slurm cluster 'team-a': keep at least one inbound rule. A group with no inbound rules leaves the VM open to all traffic.

Failed logins​

Three failed logins from one IP address in 10 minutes ban that IP address for 1 hour. Repeat bans last longer, and a further rule can ban the IP address for 1 week. The ban applies to all people behind that IP address, for example a whole office.

If you get a ban, wait for it to end or connect from another IP address. To prevent a ban, check your login name and key before you try again.

If SSH warns that the host changed​

A new login VM can get the public IP address of an old one, with a new host key. SSH then shows REMOTE HOST IDENTIFICATION HAS CHANGED.

Warning:

Do this step only if you created the cluster again. If you did not, do not connect. Contact support.

  1. Run this command on your computer, with the login VM's IP address.

    ssh-keygen -R 203.0.113.10

  2. Connect again.

A login VM restart keeps the host key, so this warning does not come after a restart.