**How severe does this issue affect your experience of using Ray?
High: It blocks me to complete my task.
I copy some code from the example-full.yaml:
available_node_types:
ray.head.default:
# The resources provided by this node type.
resources: {“CPU”: 2}
# Provider-specific config, e.g. instance type.
node_config:
azure_arm_parameters:
vmSize: Standard_D2s_v3
# List images Find and use marketplace purchase plan information using the CLI - Azure Virtual Machines | Microsoft Learn
imagePublisher: microsoft-dsvm
imageOffer: ubuntu-1804
imageSku: 1804-gen2
imageVersion: latest
How can I change these parameters correctly if I want to use another image, for example I want to use ubuntu 22.04.1? The yaml file I think corresponds to azure’s arm module, but I can’t find the relationship file, which leaves me with the example configuration to generate the virtual host cluster. I really need a vmss configured with a special environment, and I hope the environment of each vm is the same. How can I achieve this need with the least trouble? Please ask someone to help me to answer it.
If all of my cluster VMS are running windows server, I find myself only choosing to create ray clusters manually instead of using the example-full.yaml file.
you can replace the imagePublisher, imageOffer, imageSku and imageVersion with any vm image available on azure, see the website listed above those variables for instructions on how to find images.
I don’t believe there is an Ubuntu 22.04 image available currently (latest I see from micrsofot-dsvm is 20.04)
Blockquote
az vm image list -l eastus --publisher microsoft-dsvm --all --output table
You can use other images, but probably if you want to customize your workers it will be easier to do that with a docker image, since ray is by default run inside docker on these vms. Look in the docker section of the yaml. You’ll need to build and publish the docker image so you can download it during deployment.
Alternatively, you can just modify the setup_commands sections to run commands on each vm at deployment.