Running multiple VMs that each run an ONOS instance is one way of running a multi-instance ONOS deployment. It is however not practical to do on my resource-constrained laptop. Using Linux Containers is a great alternative that achieves the same thing but uses way less CPU and memory. I also didn't care much for Docker's approach: Docker is very convenient to distribute ONOS as a self-contained application, but I had trouble integrating it into my development cycle which requires fast code/deploy/test cycles.
I'm using VirtualBox as virtualization environment, but I see no reason why this shouldn't work on VMWare or other hypervisors.
The first step is to complete the first three steps of the ONOS from Scratch tutorial. So only do the VM preparation, install required software, and setup the build environment. Don't bother creating a cell definition, nor packaging and deploying ONOS. This document has new instructions for those steps.
This VM will eventually host the ONOS instance containers. In the following, we'll first create a single container that is fully configured, and then clone the original as many times as needed.
Creating Your First Container
Installing LXC is as simple as running
You should run
lxc-checkconfig to determine your system properly supports this technology.
Next we'll create a new container with a clean Ubuntu install. This command will download all kinds of dependencies so it might take a while to complete.
You can verify the container is now available on your system. Note the container is currently stopped.
Go ahead and start the container. The -d flag instructs LXC to daemonize the container, so we'll stay in our shell while the container runs in the background.
Take a look at the output of
lxc-ls again: note the container is now started and, if all has gone well, has received an IP address.
Containers are very convenient since we can automatically obtain their IPs. This is what my cell definition looks like; you want to put this in a file in
onos/tools/test/cells/. You definitely want to verify if the
ONOS_NIC variable corresponds to your system settings, and also
OCN which points to the machine where Mininet is running. Feel free to customize the
ONOS_FEATURES to your liking. Finally, you have to reload the cell if you create additional containers, so the new IPs get picked up.
Customizing Your Container
You want to do three things: enable paswordless login, enable passwordless sudo, and install Java 8.
First, push your public key to the container (make sure you have reloaded your cell definition):
Then ssh into the container as user ubuntu:
sudo visudo and add the following line to the end of the file:
Installing Java is done as follows:
Finally, log out of the container by pressing Ctrl-d or typing
We're now ready to make as many clones of the original container as you want ONOS instances. Repeat the following steps as many times as you'd like, giving each new container a unique name.
Be sure to reload your cell configuration, which will automatically create and assign new $OCx variables.
From now on, it becomes very easy to develop and test code in a multi-instance environment. The process is as follows:
- Write some code
- Compile ONOS by running
- Package ONOS by running
- Install and run the latest package:
onos-group install -f $OC1 $OC2 $OC3etc.
- Go back to step 1
Linux Bridge Configuration
I ran into the weird problem that the ONOS instances weren't discovering each other. This process is driven by Hazelcast and relies on IP multicast. It turned out that LXC uses Linux bridge for connectivity among the containers, and the bridge by default does not forward multicast traffic! This is easily solved by checking the Linux bridge documentation; here's an excerpt
IGMP snooping support is not yet included in bridge-utils or iproute2, but it can be easily controlled through sysfs interface. For brN, the settings can be found under /sys/devices/virtual/net/brN/bridge.
This option allows the user to disable IGMP snooping completely. It also allows the user to reenable snooping when it has been automatically disabled due to hash collisions. If the collisions have not been resolved however the system will refuse to reenable snooping.
This allows the user to forcibly enable/disable ports as having multicast routers attached. A port with a multicast router will receive all multicast traffic.
The value 0 disables it completely. The default is 1 which lets the system automatically detect the presence of routers (currently this is limited to picking up queries), and 2 means that the ports will always receive all multicast traffic.
Note: this setting can be enabled/disable on a per-port basis, also through sysfs interface (e.g. if eth0 is some bridge's active port, then you can adjust /sys/...../eth0/brport/multicast_router)
You may want to add this command to your startup scripts so your don't lose these settings when you reboot.
Start Containers on Boot
If you would like your containers to be automatically started on boot, you'll need to add the following line to
/var/lib/lxc/NAME/config, where NAME is your container's name. By the way, on my system the
/var/lib/lxc directory is only accessible by root.
lxc-ls --fancy to verify your containers are in fact started on boot.
If you want to access the ONOS GUI from a browse on the host machine, you need to set up port forwarding on the VM. The following command forwards traffic on port 8181 to the first ONOS container (