VM-based routers integration#

Containerlab focuses on containers, but many routing products ship only in virtual machine packaging. Leaving containerlab users without the ability to create topologies with both containerized and VM-based routing systems would have been a shame.

Keeping this requirement in mind from the very beginning, we added bridge/ovs-bridge kind that allows bridging your containerized topology with other resources available via a bridged network. For example, a VM based router:

With this approach, you could bridge VM-based routing systems by attaching interfaces to the bridge you define in your topology. However, it doesn't allow users to define the VM-based nodes in the same topology file. With vrnetlab integration, containerlab is now capable of launching topologies with VM-based routers defined in the same topology file.

Vrnetlab#

Vrnetlab packages a regular VM inside a container and makes it runnable as if it was a container image.

To make this work, vrnetlab provides a set of scripts that build the container image out of a user-provided VM disk. This integration enables containerlab to build topologies that consist both of native containerized NOSes and VMs:

Compatibility matrix#

To make vrnetlab images to work with container-based networking in containerlab, we needed to fork vrnetlab project and implement the necessary improvements. VM-based routers that you intend to run with containerlab should be built with hellt/vrnetlab project, and not with the upstream vrnetlab/vrnetlab.

Containerlab depends on hellt/vrnetlab project, and sometimes features added in containerlab must be implemented in vrnetlab (and vice-versa). This leads to a cross-dependency between these projects. The following table provides a link between the version combinations for the recent versions:

• When new VM-based platform support is added to hellt/vrnetlab, it is usually accompanied by a new containerlab version. In this case the table row will have both containerlab and vrnetlab versions.
• When vrnetlab adds new features that don't require containerlab changes, the table will state only the vrnetlab version.
• When containerlab adds new features that don't require vrnetlab changes, the table will state only the containerlab version.

It is worth noting, that you can use the latest containerlab version with a given vrnetlab version, even if the table doesn't list the latest containerlab version.

Building vrnetlab images#

To build a vrnetlab image compatible with containerlab, users first need to ensure that the versions of both projects follow compatibility matrix.

1. Clone hellt/vrnetlab and checkout to a version compatible with containerlab release:

git clone https://github.com/hellt/vrnetlab && cd vrnetlab

1. Enter the directory for the image of interest

cd sros

1. Follow the build instructions from the README.md file in the image directory

Supported VM products#

The images that work with containerlab will appear in the supported list as we implement the necessary integration.

Tuning qemu parameters#

When vrnetlab starts a VM inside the container it uses qemu command to define the VM parameters such as disk drives, cpu type, memory, etc. Containerlab allows users to tune some of these parameters by setting the environment variables in the topology file. The values from these variables will override defaults set by vrnetlab for this particular VM image.

The following env vars are supported:

• QEMU_SMP - sets the number of vCPU cores and their configuration. Use this when the default number of vCPUs is not enough or excessive.
• QEMU_MEMORY - sets the amount of memory allocated to the VM in MB. Use this when you want to alter the amount of allocated memory for the VM. Note, that some kinds have a different way to set CPU/MEM parameters, which is explained in the kind's documentation.
• QEMU_CPU - sets the default CPU model/type for the node. Use this when the default cpu type is not suitable for your host or you want to experiment with others.
• QEMU_ADDITIONAL_ARGS - allows users to pass additional qemu arguments to the VM. These arguments will be appended to the list of the existing arguments. Use this when you need to pass some specific qemu arguments to the VM overriding the defaults set by vrnetlab.

Datapath connectivity#

By hosting a VM inside a container, we made it easy to run VM-based routers in a containerized environment. However, how would you connect container's interfaces to the VM's tap interfaces in a transparent way?

To solve this challenge containerlab uses tc backend⁴, which mirrors the traffic to and from container interfaces to the appropriate VM interfaces. A huge bonus of tc is that there are not bridges inbetween, and we have a clear channel that supports transparent passage of any frames, like LACP, for example.

Management interface#

By default, vrnetlab uses the qemu user mode networking to connect the VM's (guest) management interface to the host. In this mode, the host can reach the guests management interface using a predefined IP address, which in vrnetlab's case is 10.0.0.15.

In the VM's CLI you will note the management interface to be configured with the said address:

csr-r1#show ip int br
Interface              IP-Address      OK? Method Status                Protocol
GigabitEthernet1       10.0.0.15       YES manual up                    up
GigabitEthernet2       unassigned      YES unset  administratively down down

The downside of this method of exposing the management interface to the host is that ALL VM nodes will have the same IP address assigned to their management interface. This poses two problems:

1. What if you require the management address to have a specific IP that matches your production IP plan?
2. How to integrate the nodes with external management systems that often glean the management interface IP address to identify the node against the inventory?

Starting with hellt/vrnetlab v0.21.0 a new datapath mode has been added to the vrnetlab called Transparent Management. When this mode is enabled, the IP address that Containerlab assigns to the management interface will be configured in the Network OS configuration.

To enabled this mode, set the CLAB_MGMT_PASSTHROUGH to true in your topology file:

topology:
  defaults:
    env:
      CLAB_MGMT_PASSTHROUGH: "true"

This will result in all your VM-based nodes that have been adapted to use this mode (see vrnetlab/#286 for the supported NOSes) to be configured with the IP addresses that you see in the containerlab table upon deploy.

Networking#

Vrnetlab-based container images expose their management interface on the eth0 interface. Further eth interfaces, that is, eth1 and up, are considered data interfaces and are connected to the VM using tc connection mode. Data plane interfaces are connected to the VM preserving both the order of and discontinuity between eth data-plane interfaces. Internally, the vrnetlab launch script achieves this by mapping the eth interfaces to virtualised NICs at the corresponding PCI bus addresses, filling out gaps with dummy network interfaces.

For example, a vrnetlab node with endpoints eth2, eth3 and eth5 would have these devices mapped to PCI bus addresses 2, 3 and 5 respectively, while addresses 1 and 4 would be allocated an unconnected (dummy) virtualised NIC.

For convenience and easier adaptation of configurations and lab diagrams to Containerlab topologies, vrnetlab-based nodes also support interface aliasing. Interface aliasing allows for the use of the same interface naming conventions in containerlab topologies as in the NOS, as long as interface aliasing is implemented for the NOS' kind. Note that not all NOS' implementations have support for interface aliases at the moment. For information about the supported interface naming conventions for each NOS, check out their specific Kinds page.

Boot order#

A simultaneous boot of many qemu nodes may stress the underlying system, which sometimes renders in a boot loop or system halt. If the container host doesn't have enough capacity to bear the simultaneous boot of many qemu nodes, it is still possible to successfully run them by scheduling their boot time.

Starting with v0.51.0 users may define a "staged" boot process by defining the stages and wait-for dependencies between the VM-based nodes.

Consider the following example where the first SR OS nodes will boot immediately, whereas the second node will wait till the first node is reached the healthy stage:

name: boot-order
topology:
  nodes:
    sr1:
      kind: nokia_sros
      image: nokia_sros:latest
    sr2:
      kind: nokia_sros
      image: nokia_sros:latest
      stages:
        create:
          wait-for:
            - node: sr1
              stage: healthy

Boot delay#

A predecessor of the Boot Order is the boot delay that can be set with BOOT_DELAY environment variable that the supported VM-based nodes will respect.

Consider the following example where the first SR OS nodes will boot immediately, whereas the second node will sleep for 30 seconds and then start the boot process:

name: boot-delay
topology:
  nodes:
    sr1:
      kind: nokia_sros
      image: nokia_sros:21.2.R1
      license: license-sros21.txt
    sr2:
      kind: nokia_sros
      image: nokia_sros:21.2.R1
      license: license-sros21.txt
      env:
        # boot delay in seconds
        BOOT_DELAY: 30

This method is not as flexible as the Boot Order, since you rely on the fixed delay, and it doesn't allow for the dynamic boot order based on the node health.

Memory optimization#

Typically a lab consists of a few types of VMs which are spawned and interconnected with each other. Consider a lab consisting of 5 interconnected routers; one router uses VM image X, and four routers use VM image Y.

Effectively we run just two types of VMs in that lab, and thus we can implement a memory deduplication technique that drastically reduces the memory footprint of a lab. In Linux, this can be achieved with technologies like KSM (via ksmtuned). Install KSM package on your distribution and enable it to save memory.

Find some examples below (or contribute a new one)

sudo apt-get update -y
sudo apt-get install -y ksmtuned

sudo systemctl status ksm.service
sudo systemctl restart ksm.service
sudo echo 1 > /sys/kernel/mm/ksm/run

grep . /sys/kernel/mm/ksm/*