View source for K-Scale Cluster

The K-Scale Labs clusters are shared cluster for robotics research. This page contains notes on how to access the clusters.

== Onboarding ==

To get onboarded, you should send us the public key that you want to use and maybe your preferred username.

=== Lambda Cluster ===

After being onboarded, you should receive the following information:

* Your user ID (for this example, we'll use <code>stompy</code>)
* The jumphost ID (for this example, we'll use <code>127.0.0.1</code>)
* The cluster ID (for this example, we'll use <code>127.0.0.2</code>)

To connect, you should be able to use the following command:

<syntaxhighlight lang="bash">
ssh -o ProxyCommand="ssh -i ~/.ssh/id_rsa -W %h:%p stompy@127.0.0.1" stompy@127.0.0.2 -i ~/.ssh/id_rsa
</syntaxhighlight>

Note that <code>~/.ssh/id_rsa</code> should point to your private key file.

Alternatively, you can add the following to your SSH config file, which should allow you to connect directly,
Use your favorite editor to open the ssh config file (normally located at <code>~/.ssh/config</code> for Ubuntu) and paste the text:

<syntaxhighlight lang="text">
Host jumphost
    User stompy
    Hostname 127.0.0.1
    IdentityFile ~/.ssh/id_rsa

Host cluster
    User stompy
    Hostname 127.0.0.2
    ProxyJump jumphost
    IdentityFile ~/.ssh/id_rsa
</syntaxhighlight>

After setting this up, you can use the command <code>ssh cluster</code> to directly connect.

You can also access via VS Code. Tutorial of using <code>ssh</code> in VS Code is [https://code.visualstudio.com/docs/remote/ssh-tutorial here].

Please inform us if you have any issues!

=== Notes ===

* You may need to restart <code>ssh</code> to get it working.
* You may be sharing your part of the cluster with other users. If so, it is a good idea to avoid using all the GPUs. If you're training models in PyTorch, you can do this using the <code>CUDA_VISIBLE_DEVICES</code> command.
* You should avoid storing data files and model checkpoints to your root directory. Instead, use the <code>/ephemeral</code> directory. Your home directory should come with a symlink to a subdirectory which you have write access to.

=== Andromeda Cluster ===

The Andromeda cluster is a different cluster which uses Slurm for job management. Authentication is different from the Lambda cluster - Ben will provide instructions directly.

Don't do anything computationally expensive on the main node or you will crash it for everyone. Instead, when you need to run some experiments, reserve a GPU (see below).

==== SLURM Commands ====

Show all currently running jobs:

<syntaxhighlight lang="bash">
squeue
</syntaxhighlight>

Show your own running jobs:

<syntaxhighlight lang="bash">
squeue --me
</syntaxhighlight>

Show the available partitions on the cluster:

<syntaxhighlight lang="bash">
sinfo
</syntaxhighlight>

You'll see something like this:

<syntaxhighlight lang="bash">
$ sinfo
PARTITION AVAIL  TIMELIMIT  NODES  STATE NODELIST
compute*     up   infinite      8   idle compute-permanent-node-[68,285,493,580,625-626,749,801]
</syntaxhighlight>

This means:

* There is one compute node type, called <code>compute</code>
* There are 8 nodes of that type, all currently in <code>idle</code> state
* The node names are things like <code>compute-permanent-node-68</code>

To launch a job, use [https://slurm.schedmd.com/srun.html srun] or [https://slurm.schedmd.com/sbatch.html sbatch].

* '''srun''' runs a command directly with the requested resources
* '''sbatch''' queues the job to run when resources become available

For example, suppose I have the following Shell script:

<syntaxhighlight lang="bash">
#!/bin/bash

echo "Hello, world!"

nvidia-smi
</syntaxhighlight>

I can use <code>srun</code> to run this script with the following result:

<syntaxhighlight lang="bash">
$ srun --gpus 8 ./test.sh
Hello, world!
Sat May 25 00:02:23 2024
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.54.15              Driver Version: 550.54.15      CUDA Version: 12.4     |
|-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|

... truncated
</syntaxhighlight>

Alternatively, I can queue the job using <code>sbatch</code>, which gives me the following result:

<syntaxhighlight lang="bash">
$ sbatch --gpus 16 test.sh
Submitted batch job 461
</syntaxhighlight>

We can specify <code>sbatch</code> options inside our shell script instead using the following syntax:

<syntaxhighlight lang="bash">
#!/bin/bash
#SBATCH --gpus 16

echo "Hello, world!"
</syntaxhighlight>

After launching the job, we can see it running using our original <code>squeue</code> command:

<syntaxhighlight lang="bash">
$ squeue --me
             JOBID PARTITION     NAME     USER ST       TIME  NODES NODELIST(REASON)
               461   compute  test.sh      ben  R       0:37      1 compute-permanent-node-285
</syntaxhighlight>

We can cancel an in-progress job by running <code>scancel</code>:

<syntaxhighlight lang="bash">
scancel 461
</syntaxhighlight>

[https://github.com/kscalelabs/mlfab/blob/master/mlfab/task/launchers/slurm.py#L262-L309 Here is a reference] <code>sbatch</code> script for launching machine learning jobs.

==== Reserving a GPU ====

Here is a script you can use for getting an interactive node through Slurm.

<syntaxhighlight lang="bash">
gpunode () {
    local job_id=$(squeue -u $USER -h -t R -o %i -n gpunode)
    if [[ -n $job_id ]]
    then
        echo "Attaching to job ID $job_id"
        srun --jobid=$job_id --partition=$SLURM_GPUNODE_PARTITION --gpus=$SLURM_GPUNODE_NUM_GPUS --cpus-per-gpu=$SLURM_GPUNODE_CPUS_PER_GPU --pty $SLURM_XPUNODE_SHELL
        return 0
    fi
    echo "Creating new job"
    srun --partition=$SLURM_GPUNODE_PARTITION --gpus=$SLURM_GPUNODE_NUM_GPUS --cpus-per-gpu=$SLURM_GPUNODE_CPUS_PER_GPU --interactive --job-name=gpunode --pty $SLURM_XPUNODE_SHELL
}
</syntaxhighlight>

Example env vars:
<syntaxhighlight lang="bash">
export SLURM_GPUNODE_PARTITION='compute'
export SLURM_GPUNODE_NUM_GPUS=1
export SLURM_GPUNODE_CPUS_PER_GPU=4
export SLURM_XPUNODE_SHELL='/bin/bash'
</syntaxhighlight>

Integrate the example script into your shell then run <code>gpunode</code>.

You can see partition options by running <code>sinfo</code>.

You might get an error like this: <code>groups: cannot find name for group ID 1506</code>. But things should still run fine. Check with <code>nvidia-smi</code>.

==== Useful Commands ====

Set a node state back to normal:

<syntaxhighlight lang="bash">
sudo scontrol update nodename='nodename' state=resume
</syntaxhighlight>

[[Category:K-Scale]]