build-vm-image-template.yml

steps:
- script: |
    curl -sL https://aka.ms/InstallAzureCLIDeb | sudo bash
  displayName: Install azcli

# Please follow the tutorial of [image builder](https://docs.microsoft.com/en-us/azure/virtual-machines/image-builder-overview)
# to set up a managed identity, and,
# 1. Assign the role following the instruction.
# 2. Assign contributor role of the resource group to the identity.
# 3. Add the identity to VMSS.
- script: |
    az login --identity --allow-no-subscriptions --username $(identity_id)
  displayName: Login to Azure

# Make sure all these are registered.
# If not, might need az provider register -n xxx
# Need subscription-write access.

- script: |
    set -e
    az provider show -n Microsoft.VirtualMachineImages -o json
    az provider show -n Microsoft.KeyVault -o json
    az provider show -n Microsoft.Compute -o json
    az provider show -n Microsoft.Storage -o json
    az provider show -n Microsoft.Network -o json
  displayName: Register features

# Need to create an image gallerybefore this.
# Only need to create once.
# az sig create --resource-group <resource_group> --gallery-name <sig_name>

# Add a image definition (also only once).
# az sig image-definition create -g <resource_group> \
#   --gallery-name <sig_name> \
#   --gallery-image-definition <image_def>
#
# For example,
# az sig image-definition create -g nni --gallery-name nniImageGallery \
#   --gallery-image-definition nniLinuxImage \
#   --publisher NNI \
#   --offer ubuntu \
#   --sku 20_04-nni \
#   --os-type Linux \
#   --hyper-v-generation V2

- script: |
    set -e
    set -x
    az image list -g $(resource_group)
    if az image list -g $(resource_group) --query [].'name' | grep -q $(managed_image_name); then
      az image delete -n $(managed_image_name) -g $(resource_group)
    fi
  displayName: List existing images (and delete)

- script: |
    set -e
    curl -fsSL https://apt.releases.hashicorp.com/gpg | sudo apt-key add -
    sudo apt-add-repository "deb [arch=amd64] https://apt.releases.hashicorp.com $(lsb_release -cs) main"
    sudo apt-get update && sudo apt-get install packer
  displayName: Install packer

- script: |
    set -e
    cd test/vso_tools/build_vm
    export IP_ADDRESS=$(curl -s ifconfig.me)
    export VERSION=$(date "+%Y").$(date "+%m%d").$(date "+%H%M%S")
    export CONFIG_PATH=$(packer_config).json
    sed -i -e "s/<client_id>/$(identity_id)/g" $CONFIG_PATH
    sed -i -e "s/<subscription_id>/$(subscription_id)/g" $CONFIG_PATH
    sed -i -e "s/<managed_image_name>/$(managed_image_name)/g" $CONFIG_PATH
    sed -i -e "s/<resource_group>/$(resource_group)/g" $CONFIG_PATH
    sed -i -e "s/<network_security_group>/$(network_security_group)/g" $CONFIG_PATH
    sed -i -e "s/<gallery_name>/$(gallery_name)/g" $CONFIG_PATH
    sed -i -e "s/<image_name>/$(image_definition_name)/g" $CONFIG_PATH
    sed -i -e "s/<image_version>/${VERSION}/g" $CONFIG_PATH
    sed -i -e "s/<ip_address>/${IP_ADDRESS}/g" $CONFIG_PATH
    cat $CONFIG_PATH
    echo "##vso[task.logissue type=warning]During packer build, please avoid cancelling this task. Otherwise, created resources might need manual cleanup."
  displayName: Prepare configuration

# Microsoft has a security group for VM created under their subscriptions, that,
# based on my observations (though I had no clearance to see it myself):
# 1. A low priority deny all that denies all unintended incoming traffic.
# 2. A medium-high priority denial for all traffic coming from small ports (lower than 8000 probably).
# 3. A high priority allowance for traffics from Microsoft-internal IPs.
#
# We can only insert new rules below medium. Therefore,
# 1. For Linux, we change the ssh port to 10022. This is done at provisioning by injecting user / custom data.
# 2. For Windows, they can't execute the user data script: https://stackoverflow.com/questions/62888359/custom-data-with-azure-windows-vm-run-powersell-script
#    We can't use custom script extensions either because it's not supported in packer.
#    We also can't use shell-local provisioner to invoke command, because when the VM is ready, packer always try to connect to WinRM.
#    The workaround here is to use a monitor to detect the machine ready signal and change its WinRM port.
- script: |
    cd test/vso_tools/build_vm
    python3 packer_build_windows.py
  displayName: (Windows) Packer build
  condition: and(succeeded(), contains(variables['packer_config'], 'windows'))

- script: |
    cd test/vso_tools/build_vm
    PACKER_LOG=1 packer build $(packer_config).json
  displayName: (Linux) Packer build
  condition: and(succeeded(), contains(variables['packer_config'], 'linux'))

# TODO: Should delete the managed image after build is done.
# Image gallery alone is enough. Keeping it for now for debugging purposes.

# To deploy the image on VMSS, run this in Cloud Shell:
# az vmss update --resource-group nni --name nni-windows-it \
#   --set virtualMachineProfile.storageProfile.imageReference.id=/subscriptions/{subscriptionId}/resourceGroups/nni/providers/Microsoft.Compute/galleries/nniImageGallery/images/nniWindowsImage/versions/Latest
#
# Probably need to enlarge the disk size, in case it's too small:
# az vmss update -n nni-it -g nni --set virtualMachineProfile.storageProfile.osDisk.diskSizeGb=50
#
# No need to update the image every time, because it's already set to latest.