VELDA

Appearance

Running Workloads in Velda ​

This guide explains how to run workloads in Velda, including running on different resource pools, reattaching to sessions, and connecting to open ports.

Getting Started ​

To run a workload on a different resource pool or in an ephemeral session, use:

sh
vrun -P [pool-name] command [args]

Example:

sh
vrun -P cpu-large make -j 16

Note: If the pool is not specified, the command will run on the shell pool by default, which typically provides a single CPU instance.

Running on the Same VM ​

If you need to run a workload on the same physical VM (e.g., for debugging or system monitoring), you can assign a custom service name and reattach later using the -s flag.

Start your workload in one terminal:

sh
vrun -s workload --pool my-pool my-workload

Reattach a new shell session in another terminal:

sh
vrun -s workload ps aux
# This should show the process for 'my-workload'.

Connecting to Open Ports ​

Velda provides several options for connecting to open ports from other sessions.

1. Use Session Name as Hostname ​

The simplest way is to use the service name directly as the hostname to access your workload. This method works when connecting from a session on the same instance.

sh
vrun -s web python -m http.server 8000 &
curl web:8000

2. Port-Forward While Running Workload ​

This method allows you to forward a port while launching your workload, making it accessible via localhost:

sh
vrun -p 8000:8000 python -m http.server 8000 &
curl localhost:8000

3. Port-Forward for Browser or Remote Access ​

This approach enables access from your local machine to services running in Velda:

Start your workload:

sh
vrun -s web python -m http.server 8000

From your local machine (requires Velda CLI):

sh
velda port-forward -s web --port 8000 -l :8000

After running this command, you can access the service at http://localhost:8000 in your local browser.

4. Use Proxy (HTTP/HTTPS Only) ​

Most Velda deployments include a proxy server, allowing you to use a special hostname to access your service.

If your workload is running as:

sh
vrun -s web python -m http.server 8000

And your Velda service is hosted at velda.example.com, you can access it at:

http://8000-web-[instance-name].i.velda.example.com

Note: This URL requires sign-in for security, so access it from your browser.

5. Use VS Code (or Your IDE) Port Forwarding ​

Alternatively, you can use your IDE's built-in port-forwarding feature. Refer to your specific IDE's documentation for details.

Common Examples ​

Here are some common usage patterns for running workloads in Velda:

Run a CPU-intensive compilation job:

sh
vrun -P cpu-large make -j 16

Launch a long-running Jupyter notebook server:

sh
vrun -s jupyter -p 8888:8888 jupyter notebook --no-browser

Run a GPU-accelerated training job:

sh
vrun -P gpu-t4 python train.py --epochs 100