vSphere Integrated Containers 0.4 – Inspecting VCH and ContainerVM

Last week, I had an interesting conversation with my friend Michael on vSphere Integrated Containers (VIC) in it’s current version 0.4. We discussed some of the key concepts and how they relate to other container implementations out there. I decided to summarize the key observations with a little more detail here as I expect this information to be interesting for operations teams once they start running VIC.
Please note: this is based on the currently available Open Source VIC project in version 0.4 running on vSphere 6.0 in my homelab.
For simplicity reasons, I decided to go with a “standalone ESXi” installation of my Virtual Container Host (VCH) in this example.

First, I created a new container host called “VCH001” on my ESXi host from my PhotonOS based worker VM:
root@photonbox [ /workspace/vic ]# ./vic-machine-linux create --bridge-network 'VCH Bridge' --external-network 'VM Network' --image-datastore mydatastore --target 'root@esxi01.think-v.com' --name VCH001

The output of this command shows the necessary details:
INFO[2016-08-06T19:51:56Z] Please enter ESX or vCenter password:
INFO[2016-08-06T19:52:00Z] ### Installing VCH ####
INFO[2016-08-06T19:52:00Z] Generating certificate/key pair - private key in ./VCH001-key.pem
INFO[2016-08-06T19:52:02Z] Validating supplied configuration
INFO[2016-08-06T19:52:05Z] Firewall status: DISABLED on /ha-datacenter/host/esxi01.think-v.com/esxi01.think-v.com
INFO[2016-08-06T19:52:05Z] Firewall configuration OK on hosts:
INFO[2016-08-06T19:52:05Z] /ha-datacenter/host/esxi01.think-v.com/esxi01.think-v.com
INFO[2016-08-06T19:52:05Z] License check OK
INFO[2016-08-06T19:52:05Z] DRS check SKIPPED - target is standalone host
INFO[2016-08-06T19:52:07Z] Creating Resource Pool VCH001
INFO[2016-08-06T19:52:07Z] Creating appliance on target
INFO[2016-08-06T19:52:07Z] Network role client is sharing NIC with external
INFO[2016-08-06T19:52:07Z] Network role management is sharing NIC with external
INFO[2016-08-06T19:52:09Z] Uploading images for container
INFO[2016-08-06T19:52:09Z] bootstrap.iso
INFO[2016-08-06T19:52:09Z] appliance.iso
INFO[2016-08-06T19:52:22Z] Waiting for IP information
INFO[2016-08-06T19:52:42Z] Waiting for major appliance components to launch
INFO[2016-08-06T19:52:44Z] Initialization of appliance successful
INFO[2016-08-06T19:52:44Z] Log server:
INFO[2016-08-06T19:52:44Z] https://VCH_IP:2378
INFO[2016-08-06T19:52:44Z] DOCKER_HOST=VCH_IP:2376
INFO[2016-08-06T19:52:44Z] Connect to docker:
INFO[2016-08-06T19:52:44Z] docker -H VCH_IP:2376 --tls info
INFO[2016-08-06T19:52:44Z] Installer completed successfully

More details about the inner workings can be found in the VIC 0.4 blogposts by Cormac that are also listed in the link section below. In this post I’d like to focus more on the topic of state information and how this is handled in VIC 0.4.

First of all, it is important to understand the difference between VCHs in VIC in comparison to other (in this case linux-based) container solutions. While each container in a N:1 model (containers:linux) has its private namespace, the underlying shared kernel provides the container control plane to look into containers and perform process-related actions (start, stop, …). Runtime environment and control plane are directly coupled.

In VIC, the runtime/execution environment of the container is a so called containerVM (based on PhotonOS) which is decoupled from it’s “control plane”, the Virtual Container Host itself. This creates a new layer of abstraction where communication flow but also state information needs to be captured and made available.

To establish a secure communications path between these two components, VIC also introduces the concept of a Tether to connect into the actual containerVM. This concept is part of the Port Layer Abstractions that allows VIC to be extensible. More details are described on the VIC Container Abstractions documentation page.

Let me share a summary of how the VCH and containerVMs actually look like on the infrastructure – and where information on state is actually stored. At first, let me go into the VMX file of the VCH. As expected, there are two vNICs attached:

ethernet0.virtualDev = "vmxnet3"
ethernet0.networkName = "VM Network"
ethernet0.pciSlotNumber = "192"
ethernet0.uptCompatibility = "TRUE"
ethernet0.present = "TRUE"
ethernet1.virtualDev = "vmxnet3"
ethernet1.networkName = "VCH Bridge"
ethernet1.pciSlotNumber = "224"
ethernet1.uptCompatibility = "TRUE"
ethernet1.present = "TRUE"

Here, we also find the boot disk that got transferred with the deployment of the VCH:

ide0:0.deviceType = "cdrom-image"
ide0:0.fileName = "appliance.iso"
ide0:0.present = "TRUE"

The general approach for storing state information is described in the Configuration persistence mechanism overview documentation. According to this, VIC actually makes use of the vSphere extraConfig and guestinfo mechanisms to store relevant information. But where do extraConfig and guestinfo actually reside? In a normal vSphere VM, this information is stored in the VMX file of the VM (and remember, a container in VIC actually is a VM – the containerVM).

Starting a simple “hello-world” container should trigger the whole workflow that also creates a new VM. But let’s go through it step by step:

root@photonbox [ /workspace/vic ]# docker -H VCH_IP:2376 --tls run -it hello-world
Unable to find image 'hello-world:latest' locally
latest: Pulling from library/hello-world
a3ed95caeb02: Pull complete
c04b14da8d14: Pull complete
Digest: sha256:548e9719abe62684ac7f01eea38cb5b0cf467cfe67c58b83fe87ba96674a4cdd
Status: Downloaded newer image for library/hello-world:latest

Looking at the recently executed containers from my worker VM, we can see the following reference:

root@photonbox [ /workspace/vic ]# docker -H VCH_IP:2376 --tls ps -a
2cf7f483bf6e hello-world "/hello" Less than a second ago Stopped jolly_panini

So our container ran as ID 2cf7f483bf6e. How does that containerVM actually look on our standalone ESXi host and even more interestingly, where does the information about the container (from docker ps -a) come from?

First of all, there is a newly created VM named 2cf7f483bf6e7f32daa53f51ca388d5fb153f78d3a74d313318099086638ad58 – just as expected. Looking at the VMX file, we’ll find a lot of session information that we already found in docker ps -a:

guestinfo./common/name = "jolly_panini"
guestinfo./sessions|2cf7f483bf6e7f32daa53f51ca388d5fb153f78d3a74d313318099086638ad58/common/name = "jolly_panini"
guestinfo./sessions|2cf7f483bf6e7f32daa53f51ca388d5fb153f78d3a74d313318099086638ad58/cmd/Path = "/hello"
guestinfo./repo = "hello-world"

The containerVM mounts the “bootstrap.iso” from the VCH001’s VM folder (that also got deployed via the vic-machine installer):

ide0:0.deviceType = "cdrom-image"
ide0:0.fileName = "/vmfs/volumes/d37f7a1b-0ab13c48/VCH001/bootstrap.iso"
ide0:0.present = "TRUE"

The containerVM also has a serial connection to the VCH (explanation):

serial0.allowGuestConnectionControl = "FALSE"
serial0.fileType = "network"
serial0.fileName = "tcp://VCH_IP:8080"
serial0.network.endPoint = "client"
serial0.yieldOnMsrRead = "TRUE"
serial0.present = "TRUE"
serial0.hardwareFlowControl = "TRUE"

The containerVM’s network adapter is connected on the “VCH Bridge” portgroup and therefore only talks to the VCH. This is where the container traffic is flowing, management and control plane traffic is going via serial0.

ethernet0.virtualDev = "vmxnet3"
ethernet0.networkName = "VCH Bridge"
ethernet0.pciSlotNumber = "192"
ethernet0.uptCompatibility = "TRUE"
ethernet0.present = "TRUE"

The containerVM also has it’s own harddisk (attached VMDK):

scsi0.virtualDev = "pvscsi"
scsi0.present = "TRUE"
scsi0:0.deviceType = "scsi-hardDisk"
scsi0:0.fileName = "2cf7f483bf6e7f32daa53f51ca388d5fb153f78d3a74d313318099086638ad58.vmdk"
scsi0:0.present = "TRUE"

To delete the VCH and the containerVMs, vic-machine-linux is called with the “delete” option:

root@photonbox [ /workspace/vic ]# ./vic-machine-linux delete --target esxi01.think-v.com --user root --name VCH001
INFO[0000] Please enter ESX or vCenter password:
INFO[2016-08-06T20:50:24Z] ### Removing VCH ####
INFO[2016-08-06T20:50:28Z] Removing VMs
INFO[2016-08-06T20:50:33Z] Removing images
INFO[2016-08-06T20:50:34Z] Removing volumes
INFO[2016-08-06T20:50:36Z] Removing appliance VM network devices
INFO[2016-08-06T20:50:38Z] Bridge network was not created during VCH deployment, leaving it there
INFO[2016-08-06T20:50:40Z] Removing Resource Pool VCH001
INFO[2016-08-06T20:50:40Z] Completed successfully


In summary, all container state information is kept close to the containerVM, stored in the VMX file. VCH and containerVM use the ISO-files that are tranferred during the vic-machine install process. VIC also introduces a new level of abstraction between control plane and execution environment that allows VIC to be extensible for future usecases.


Additional links around VIC 0.4 – most of them by Cormac:

Leave a Comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.