How to automate VM deployment from large USB keys using ESXi Kickstart?

During VMworld US, I had the opportunity to speak with several customers to learn about their VMware environment and some of the challenges they were facing. In some scenarios, I was able to offer a solution or a different way of solving the problem. For others, it was primarily feedback on how we can better improve some of our capabilities/features or specific feature requests they would like to see get added.

One interesting challenge that arose from a class of customers who manages hundreds of remote sites is the ability fully automate the provisioning of an ESXi host as well as set of Virtual Machines as part of the initial deployment. The provisioning is all done through Kickstart (unattended installation of ESXi) and usually from a USB device but it could also be from a custom ISO. One ask that kept coming up was the support for larger USB key support within ESXi so that it could be used to include additional payload.

As some of you may or may not know, ESXi can only access USB devices within the ESXi Shell formatted using the FAT16 filesystem which allows for a maximum file size of 2GB for each partition. However, this limitation is only for the ESXi Shell itself and for the size of the ESXi installation media, this is more than sufficient. If you wish to leverage larger USB keys which has increased significantly in recent years from 32GB, 64GB and even 128GB, you can directly pass that into any guest OS through the USB Arbitrator Service (enabled by default) and there you will be able to consume the entire capacity of the USB device. The challenge is how do you go about bootstrapping ESXi as well as the initial set of Virtual Machines with these limitations and completely automated using an ESXi Kickstart?

Over the years I have seen some really creative solutions to solving this problem and funny enough, right before VMworld I had several folks reach out asking similar questions. I decided to take a look and also build upon some earlier work done by a fellow VMware SE (Tim S) to come up with a completely automated solution that would scale to any size USB device and hopefully make it easy to extend if needed.

For this project, I used a 64GB USB key which I received from the folks over at Micron who I visited in the Solution Exchange during VMworld US (these guys are doing some really awesome stuff with VSAN and an All-Flash array, be sure to check them out).

Here is a diagram of the partition structure for the 64GB USB key which I will explain further:

The first partition is 2GB using a FAT16 filesystem and this is used to store the actual ESXi media along with an embedded ESXi Kickstart configuration file. You can easily reference a remote Kickstart if you wish, but for simplicity purposes and to support some of the requested use cases from customers, I have embedded it.

The second partition is also 2GB using a FAT16 filesystem and this is used to store a tiny VM which I am calling a "Service VM". This VM needs to be small enough to fit the partition and will be used to read the remainder capacity of the USB device which will be using a more capable filesystem type. I have decided to store a pre-configure vMA appliance which is tarred up to reduce the disk footprint.

The third and final partition will consume the remainder capacity of the USB device, in this case it would be 60GB and using a FAT32 partition which can support up to 2TB for a single volume. This is where additional Virtual Machines would be stored and accessed by the "Service VM".

As you can probably guess, the idea is to install ESXi as you normally would to a local disk or directly onto the USB device in which case an additional partition would be required. As part of the installation, the "Service VM" would be boot strapped as it would be visible within the ESXi Shell and registered and powered on during first bootup. A first boot script could then be included in the guestOS which can receive some details about the ESXi deployment which could be hard coded (not recommended) or dynamically discovered as I have implemented it. The USB device would then be passed directly to this "Service VM" to mount and then it would be able to deploy the remainder Virtual Machines which would be stored in this larger partition.

Here is the complete ESXi Kickstart which implements what has been discussed so far and I have also included a break down of the kickstart below:

vmaccepteula
install --firstdisk --overwritevmfs
rootpw vmware123
reboot

network --bootproto=static --ip=192.168.1.200 --netmask=255.255.255.0 --gateway=192.168.1.1 --hostname=mini.primp-industries.com --nameserver=192.168.1.1 --addvmportgroup=1

%post --interpreter=busybox

# stop USB Arbitrator service to access USB device in ESXi Shell
/etc/init.d/usbarbitrator stop

# copy service VM to local VMFS datastore
cp /vmfs/volumes/SERVICEVM/vMA.tar.gz /vmfs/volumes/datastore1
tar -zvxC /vmfs/volumes/datastore1 -f /vmfs/volumes/datastore1/vMA.tar.gz
rm -f /vmfs/volumes/datastore1/vMA.tar.gz

# add guestinfo property for ESXi IP Address for adv. VM deployment
ESXI_IP=$(localcli network ip interface ipv4 get | grep vmk0 | awk '{print $2}')
echo "guestinfo.esxi_ip = ${ESXI_IP}" >> /vmfs/volumes/datastore1/vMA/vMA.vmx

%firstboot --interpreter=busybox

# Ensure hostd is ready
while ! vim-cmd hostsvc/runtimeinfo; do
sleep 10
done

# enable & start SSH
vim-cmd hostsvc/enable_ssh
vim-cmd hostsvc/start_ssh

# enable & start ESXi Shell
vim-cmd hostsvc/enable_esx_shell
vim-cmd hostsvc/start_esx_shell

# Suppress ESXi Shell warning
esxcli system settings advanced set -o /UserVars/SuppressShellWarning -i 1

# rename datastore1
vim-cmd hostsvc/datastore/rename datastore1 mini-local-datastore-1

# Register VM
vim-cmd solo/registervm /vmfs/volumes/mini-local-datastore-1/vMA/vMA.vmx

# connect USB device via passthrough
USB_DEV_NAME="Alcor Micro Corp"
USB_DEV_BUSID=$(lsusb | grep "${USB_DEV_NAME}" | awk '{print $2}' | cut -c 2)
vim-cmd vmsvc/device.connusbdev 1 "path:${USB_DEV_BUSID}/0/1 version:2"

# power on VM
vim-cmd vmsvc/power.on 1

Line12 - Need to disable the USB Arbitrator Service so the USB device can be seen by ESXi since it is by default made ready to be exposed to a VM. The service will be automatically re-enabled after the installation of ESXi which will allow for the VM to connect to the USB device.

Line15-17 - Copy the "Service VM" from USB device to local VMFS datastore1. In the example, I have pre-configured the vMA appliance tarred up the VMX and its respective VMDK.

Line20-21 - Extract the ESXi IP Address and sets a custom guestInfo property so the "Service VM" knows where to deploy the additional VMs to

Line26-29 - This checks to ensure hostd is up and running before continuing on

Line45 - Register the "Service VM" within ESXi

Line49-50 - Identify the USB device ID which will be required to mount to the "Service VM". You will need to update USB_DEV_NAME based on the USB device you are using

Line51 - Connect the USB Device to "Service VM"

Line54 - Power on the "Service VM"

At this point, you should be able to access the USB device from within the "Service VM". We can easily verify this by running the following command:

sudo fdisk -l

As seen in the screenshot above, we can see our three partitions and third is the one with our FAT32 partition which contains a couple of Virtual Machines that I want to deploy. Of course, this partition can contain anything you wish to store, so the sky is the limit!

To mount the USB device and the specific partition, we will create a temporarily directory and issue the mount command by running these two commands:

sudo mkdir -p /mnt/USB;sudo mount /dev/sdb3 /mnt/USB

For my USB key, I have stored both the VCSA and NSX Manager OVA which can then be deployed using ovftool. The last part to be able to make this as seamless and automated as possible is to be able to identify the ESXi host information. If you recall earlier, we had set a custom guestInfo property within our "Service VM". This custom property can then be read by the guestOS leveraging VMware Tools and provides the IP Address to the guest. You can easily set other metadata information but to be able to deploy these additional OVA's, we would need to know the IP Address of the ESXi host and this makes it so you do not need to hard code anything (perhaps ESXi host credentials).

To retrieve this custom property, you will need to run the following command:

vmtoolsd --cmd "info-get guestinfo.esxi_ip"

With these last few guestOS commands, you will be able to create a firstboot script which will automatically mount the appropriate USB partition and deploy these additional Virtual Machines. This is just one of the many possibilities on how you can deploy additional VMs as part of your ESXi Kickstart deployment. Hopefully this solution provides a base in which you can easily customize based on your own requirements.