utils

Utilities for NF developer setup/etc

Creating a Developer environment with vagrant

1. Clone this repository, MoonGen, and NetBricks into the same parent directory.

   $ for repo in utils MoonGen NetBricks; do \
       git clone [email protected]:williamofockham/${repo}.git; \
     done

2. Install Vagrant and VirtualBox.
3. Install the vagrant-disksize (required) and vagrant-vbguest (recommended) vagrant-reload (required) plugins:

   $ vagrant plugin install vagrant-disksize vagrant-vbguest vagrant-reload

4. Boot the VM:

   $ cd utils
   $ vagrant up

The above steps will prepare your virtual machine with all of the appropriate DPDK settings (multiple secondary NICs, install kernel modules, enable huge pages, bind the extra interfaces to DPDK drivers) and the sandbox Docker image.

If you have MoonGen and NetBricks cloned as described in step 1 above, those repositories will be shared into the VM at /MoonGen and /NetBricks respectively.

Design of our Docker images

ubuntu/xenial                        (upstream)
+- williamofockham/dpdk              (container-friendy DPDK build)
   +- williamofockham/sandbox        (base image, copies in MoonGen)
   /  +- williamofockham/netbricks   (NFV framework)
+- williamofockham/moongen           (traffic/packet generator)

Most times you will want to run the williamofockham/netbricks container, which includes all of the other tools. This is started by default in the Vagrant setup.

The goal of our structure here is to avoid requiring rebuilding each of the frameworks every time. We use multi-staged image builds based on ubuntu/xenial, which is the same as the host OS in the Vagrant setup above. The williamofockham/dpdk image is built in this repository from the contents of the dpdk directory, and the williamofockham/sandbox image is built from the Dockerfile.sandbox in the root directory. The other images are built from their respective repositories.

Because DPDK has a lot of requirements from the host OS, many files and directories are mounted into the container, which also runs in "privileged" mode. The following mounts are required (and handled in most docker.mk files in our repositories):

# Kernel modules and headers
/lib/modules
/usr/src

# Access to the host PCI bus
/sys/bus/pci/drivers

# Access to huge pages
/sys/kernel/mm/hugepages

# Access to NUMA configuration
/sys/devices/system/node

# Enumerating kernel modules
/sbin/modinfo
/bin/kmod
/sbin/lsmod

# Device nodes
/dev

# Huge-pages filesystem
/mnt/huge

# Sharing DPDK runtime configuration or unix sockets
/var/run

[DEPRECATED] Creating Docker host with corectl (Mac)

NOTE: We are no longer using corectl because getting proper DPDK compatibility is too difficult.

corectl is a tool for running CoreOS virtual machines using OS/X's built-in hypervisor toolkit. To get started, be sure you have Homebrew installed and then run these commands inside the sandbox:

$ brew update
$ make -f corectl.mk vm

This will, in order:

1. Install corectl via homebrew.
2. Start up the daemon that manages CoreOS virtual machines, corectld. At this step you will be asked for your password.
3. Download the latest alpha channel CoreOS image.
4. Create and format a persistent disk image in the current directory.
5. Boot the williamofockham VM with the persistent disk attached and configure Docker for local use with IPv6 enabled.

Once the VM is booted, you can interact with its Docker daemon with the following:

$ export DOCKER_HOST=tcp://$(corectl q -i williamofockham):2375

or

$ eval `make -f corectl.mk env`

and then

$ docker ps
CONTAINER ID        IMAGE               COMMAND             CREATED             STATUS              PORTS               NAMES

There should be no containers listed. You are now ready to use Docker for local development.

Tweaking the CoreOS configuration

There are a number of environment variables you may change when invoking make -f corectl.mk to change the functionality of the VM:

• COREOS_CHANNEL: (default: alpha) You can choose beta or stable update channels for CoreOS.
• CORECTL_VOLUME_SIZE: (default: 40GiB) The size of the QCOW2 image that will be created to store the Docker host's container images and state.
• CORECTL_VM_NAME: (default: williamofockham) The name of the CoreOS VM to be created.
• CORECTL_CPUS: (default: 2) The number CPU cores to allocate to the virtual machine.
• CORECTL_RAM: (default: 4096) The amount of RAM to allocate to the virtual machine, in megabytes.
• CORECTL_VOLUME: (default: williamofockham-docker.img.qcow2) The filename of the QCOW2 image that will be created to store the Docker host's container images and state.
• CORECTL_INIT: (default: corectl/cloud-init.yaml) A cloud-init file used to configure the virtual machine on startup.
• CORECTL_KERNEL_FLAGS: (default: hugepages=1024) Boot arguments to the Linux kernel in the virtual machine, which you can use to enable, disable, or configure certain operating system features.

SSH'ing and Destroying the vm

If you need to connect to the VM, for example to get the configured IPv6 address, use corectl ssh williamofockham.

To shutdown and destroy the VM and its persistent storage, run make -f corectl.mk destroy.

Caveats

• corectl has not been updated recently and so it has not taken into account the latest signature key for validating CoreOS images. corectl.mk circumvents this by fetching the image manually into the appropriate directories and passes the -o (offline) flag when running VMs to bypass the built-in fetching behavior.