Introduction to AzureVM

Hong Ooi

AzureVM is a package for interacting with virtual machines and virtual machine scalesets in Azure. You can deploy, start up, shut down, run scripts, deallocate and delete VMs and scalesets from the R command line. It uses the tools provided by the AzureRMR package to manage VM resources and templates.

Virtual machines

Here is a simple example. We create a VM using the default settings, run a shell command, resize the VM, and then delete it.


sub <- AzureRMR::get_azure_login()$

# calling create_vm() from a subscription object will create the VM in its own resource group
# default is an Ubuntu 18.04 VM, size Standard_DS3_v2, login via SSH key
# call sub$list_vm_sizes() to get the sizes available in your region
vm <- sub$create_vm("myubuntuvm", user_config("myname", "~/.ssh/"),

# some resources used by the VM

# run a shell script or command remotely (will be PowerShell on a Windows VM)
vm$run_script("echo hello world! > /tmp/hello.txt")

# ... and stop it

# ... and resize it

# ... and delete it (this can be done asynchronously for a VM in its own group)

AzureVM comes with a number of predefined configurations, for deploying commonly used VM images. For example, to create an Ubuntu DSVM accessible via SSH, JupyterHub and RStudio Server:

sub$create_vm("mydsvm", user_config("myname", "~/.ssh/"), config="ubuntu_dsvm",

And to create a Windows Server 2019 VM, accessible via RDP:

sub$create_vm("mywinvm", user_config("myname", password="Use-strong-passwords!"), config="windows_2019",

The available predefined configurations are ubuntu_18.04 (the default), ubuntu_16.04, ubuntu_dsvm, windows_2019, windows_2016, windows_dsvm, rhel_7.6, rhel_8, centos_7.5, centos_7.6, debian_8_backports and debian_9_backports. You can combine these with several other arguments to customise the VM deployment to your needs:

# Windows Server 2016, with a 500GB datadisk attached, not publicly accessible
sub$create_vm("mywinvm2", user_config("myname", password="Use-strong-passwords!"),
              size="Standard_DS4_v2", config="windows_2016", datadisks=500, ip=NULL,

# Ubuntu DSVM, GPU-enabled
sub$create_vm("mydsvm", user_config("myname", "~/.ssh/"), size="Standard_NC12",

# Red Hat VM, serving HTTP/HTTPS
sub$create_vm("myrhvm", user_config("myname", "~/.ssh/"), config="rhel_8",
              nsg=nsg_config(list(nsg_rule_allow_http, nsg_rule_allow_https)),

Full customisation is provided by the vm_config function, which also lets you specify the image to deploy, either from the marketplace or a disk. (The predefined configurations actually call vm_config, with the appropriate arguments for each specific config.)

## custom VM configuration: Windows 10 Pro 1903 with data disks
## this assumes you have a valid Win10 desktop license
user <- user_config("myname", password="Use-strong-passwords!")
image <- image_config(
datadisks <- list(
    datadisk_config(250, type="Premium_LRS"),
    datadisk_config(1000, type="Standard_LRS")
nsg <- nsg_config(
sub$create_vm("mywin10vm", user,

VM scalesets

The equivalent to create_vm for scalesets is the create_vm_scaleset method. By default, a new scaleset will come with a load balancer and autoscaler attached, but its instances will not be externally accessible.

# default is Ubuntu 18.04 scaleset, size Standard_DS1_v2
sub$create_vm_scaleset("myubuntuss", user_config("myname", "~/.ssh/"), instances=5,

Each predefined VM configuration has a corresponding scaleset configuration. To specify low-level scaleset options, use the scaleset_options function. Here are some sample scaleset deployments:

# Windows Server 2019
sub$create_vm_scaleset("mywinss", user_config("myname", password="Use-strong-passwords!"), instances=5,

# RHEL scaleset, serving HTTP/HTTPS
sub$create_vm_scaleset("myrhelss", user_config("myname", "~/.ssh/"), instances=5,
                        nsg=nsg_config(list(nsg_rule_allow_http, nsg_rule_allow_https)),

# Ubuntu DSVM, GPU-enabled, public instances, no load balancer or autoscaler
sub$create_vm_scaleset("mydsvmss", user_config("myname", "~/.ssh/"), instances=5,
                       size="Standard_NC6", config="ubuntu_dsvm_ss",
                       load_balancer=NULL, autoscaler=NULL,

# Large Debian scaleset (multiple placement groups), using spot VMs (low-priority)
# need to set the instance size to something that supports low-pri
sub$create_vm_scaleset("mylargess", user_config("myname", "~/.ssh/"), instances=10,
                       size="Standard_DS3_v2", config="debian_9_backports_ss",
                       options=scaleset_options(priority="spot", large_scaleset=TRUE),

Working with scaleset instances can be tedious if you have a large scaleset, since R can only connect to one instance at a time. To solve this problem, AzureVM can leverage the process pool functionality supplied by AzureRMR to connect in parallel with the scaleset, leading to significant speedups. The pool is created automatically the first time it is needed, and is deleted at the end of the session.

# this will create a pool of up to 10 processes that talk to the scaleset
mylargess$run_script("echo hello world! > /tmp/hello.txt")

You can control the size of the pool with the global azure_vm_minpoolsize and azure_vm_maxpoolsize options, which have default values 2 and 10 respectively. To turn off parallel connections, set options(azure_vm_maxpoolsize=0). Note that the pool size is unrelated to the scaleset size; it only controls how many instances can communicate with AzureVM simultaneously.

Sharing resources

You can also include an existing Azure resource in a deployment, by supplying an AzureRMR az_resource object as an argument in the create_vm or create_vm_scaleset call. For example, here we create a VM and a scaleset that share a single virtual network/subnet.

## VM and scaleset in the same resource group and virtual network
# first, create the resgroup
rg <- sub$create_resource_group("rgname", "australiaeast")

# create the master
rg$create_vm("mastervm", user_config("myname", "~/.ssh/"))

# get the vnet resource
vnet <- rg$get_resource(type="Microsoft.Network/virtualNetworks", name="mastervm-vnet")

# create the scaleset
# since the NSG is associated with the vnet, we don't need to create a new NSG either
rg$create_vm_scaleset("slavess", user_config("myname", "~/.ssh/"),
                      instances=5, vnet=vnet, nsg=NULL)