Skip to content

Latest commit

 

History

History
139 lines (103 loc) · 7.46 KB

README.md

File metadata and controls

139 lines (103 loc) · 7.46 KB
Raw

Desktop GUI Setup for simulator environments

Contents

Nautilus GUI Desktop

1. Follow the Nautilus setup tutorial to install the kubernetes command line and get access to the AIEA lab namespace

https://docs.nationalresearchplatform.org/userdocs/start/quickstart/

2. Once you have installed kubernetes command and can view the AIEA lab's namespace pvc create the following:

  • A persistent volume claim (PVC) with seaweed-nvme for your main storage. 64Gb should suffice if you're just trying to run CARLA without machine learning. Note that is is important that you choose seaweed-nvme for your storage because it has a better high read/write performance compared to ceph-rock.
  • A persistent volume claim for your pod's cache. I got by with 1Gb, for reference.

3. Once the PVCs are created you need to ask a turn-shared secret from Nautilus Support

  • Once you're done setting up your pvcs head on to the nautilus support channel and ask them for a turn secret. Someone from their team will ping and share the key to you. Keep that key with you, you will require it at a later stage when creating your secrets.
  • The turn-shared secret is required if you're setting up a server instead of port forwarding your pod everytime. For more details you can check out this page - https://docs.nationalresearchplatform.org/userdocs/running/gui-desktop/#secret-generation

4. Nautilus' GUI desktop runs in your web browser to port-forward the GUI screen. So you will need to create two secrets for the yaml file below.

Kubernetes secrets are basically Python dictionaries.

  1. kubectl create secret generic [my-pass] --from-literal=[my-password]=[YOUR_PASSWORD] --from-literal=[turn-secret]=[TURN_SHARED_SECRET_FROM_NAUTILUS]
{
    my-pass: 
        my-password: 
            YOUR_PASSWORD: ****
        turn-secret:
            TURN_SHARED_SECRET_FROM_NAUTILUS: **********************
}
  1. kubectl create secret generic [turn-shared-secret] --from-literal=[turn-shared]=[MY_TURN_SHARED_SECRET]
{
    turn-shared-secret: 
        turn-shared: 
            MY_TURN_SHARED_SECRET: ********
}


Give your secrets a unique name so there are no conflicts. Consider using your ucsc username.
Update your values appropriately at places where its enclosed in square brackets (Don't use square brackets in your command)
The first secret generates your authentication to the GUI Desktop and the second secret will be your turn-share server authentication details, if you're planning to set that up. Check the docs on Step 3 to know how to do it

5. Edit nautilus-files/pod-version-xgl.yaml

  • Change the name in line 4
  • Change line 28, 29 to your secret name and key (my-pass and my-password in the command provided)
  • Change line 43, 47 to your secret name and key (my-pass and turn-secret in the command provided)
  • Change line 50 to a username of your choice (This would be your turn-server username)
  • Change line 54, 55 to your secret name and key (turn-shared-secret and turn-shared in the command provided)
  • Change line 93 to your cache volume
  • Change line 96 to your storage volume

6. Run the pod

kubectl create -f [your-xgl-filename].yaml

7. Check if pod is running

  • Check pod status kubectl get pods
  • If you face any issues that occur regarding your pod, check and debug using this command kubectl describe pod [pod-name]
    If you see your pod is running, then you've completed the hard part!

8. Port-forward the pod

Run - kubectl port-forward pod/[pod-name] 8080:8080
You should see an output that looks like this: Forwarding from 127.0.0.1:8080 -> 8080
Copy and paste 127.0.0.1:8080 into your web browser

Make sure to run this command in your local terminal where you have access to a web browser

9. Localhost page

  • The page should prompt you to enter the username which is "user" and password from line 50 of your secret from the pod's yaml file.

10. Run any program that requires graphics!

Once you have gained proficiency in using a GUI pod on Nautilus, you may ask Coen Adler ctadler@ucsc.edu or Shreedhar Jangam sjangam@ucsc.edu for permission to run a Deployment version

CARLA Installation Instructions

Disclaimer

When installing Carla use the persistent volume path cd ../../mnt/persistent/ from the terminal, to have your data present whenever you port-forward the pod. If you don't do that, then you may lose your data every time you stop your connection. Note that is it not recommended to install Carla directly on the PCV - install Carla in your home directory for best performance.

Bash Script Way

1. Run the bash script, install_carla.sh

You might need to run chmod +x install_carla.sh first if it is not executable. After, run ./install_carla.sh.

Manual Way

1. Install Miniconda

2. Install CARLA

3. Python packages

  • Once CARLA is installed or everything is unzipped install the recommended Python packages by running
cd PythonAPI\examples
python3 -m pip install -r requirements.txt

It might show an error saying scipy or scikitlearn is not installed. If so, run pip install scipy or pip install scikit-learn

4. Running CARLA

./CarlaUE4.sh -carla-rpc-port=4000 # port num is up to you

You might get an error saying "Cannot find a compatible Vulkan device...". Sometimes vulkan is not the default GPU driver in a Nautilus GUI pod with multiple GPUs. You can get around this by adding the flag

./CarlaUE4.sh -graphicsadapter=2

add the flag -prefernvidia if carla is not running on the GPU.

python generate_traffic.py -p 4000 --tm-port 4050

5. Witness the power of cloud computing!

Common Errors that you might face

  • If you're facing "Transport endpoint is not connected" when your running cd command, try recreating your volumes or unmount them and run the pod again.
  • If you login and are stuck on a black screen, open the side menu and reduce the video bitrate (you can increase this after the GUI shows properly).
  • Sometimes the GUI will freeze all of a sudden. You can fix this by exiting the port-forward, ssh-ing into the pod, run htop, then kill the program selkies-gstreamer in white text using the most CPU. Exit the ssh pod and start a new port-forward.
  • CARLA, by default, uses port 8000 for any Traffic Manager related function calls. However, the Nautilus GUI is using port 8000. Thus, you will have to specify a new Traffic Manager Port.
  • You can request certain GPUs like in any pod in the yaml file.
  • For more information regarding this pod use this link - https://docs.nationalresearchplatform.org/userdocs/running/gui-desktop/