OpenKilda is a Web-Scale Software-Defined Networking controller. OpenKilda is capable of manage traffic on tens of thousands of switches simultaneously, control millions of flows, and provide sub-second network telemetry. OpenKilda provides a variety of features, such as:
- manual and automatic management of L2 services: point-to-point flows, Y-flows, etc.;
- advanced features for some types of services: mirroring, data gathering, path pinning, etc.;
- path computation engine: calculating paths based on different cost functions;
- resilient mechanisms: quick reaction on changes on hardware level, paths recalculations, diverse flow groups to ensure back-up routes;
- parallelism: flexible distributed computation, easy-to-change configuration;
- visualized telemetry and reporting;
- REST API and GUI to configure and manage OpenKilda capabilities;
- and more!
You can find OpenKIlda high level design here.
Since the release 1.126.2 (December 2022), single-table mode for switches became deprecated. New features will be designed only for multi-table mode. Current features will support single table mode till July 1, 2023. After that, single table mode support might be removed from any feature.
OpenKilda is an open source project that welcomes contributions from developers. To contribute, it is mandatory to follow our contribution workflow.
The build process described below requires to install a number of packages. It is recommended that you build OpenKilda on a virtual machine.
The following packages are required for building OpenKilda controller:
- Gradle 7.0+
- Maven 3.3.9+
- JDK 8+
- Python 3.8+
- Docker 19.03.3+
- Docker Compose 1.20.0+
- GNU Make 4.1+
- Open vSwitch 2.9+
We do not recommend upgrading pip and install docker-compose using the methods described below, bypassing the packer managers. Instead, please read the documentation for installing the pip and the docker-compose.
For running a virtual environment (that is a Docker instance with the Open vSwitch service) it is required to have Linux kernel 4.18+ for OVS meters support. The following commands will install necessary dependencies on Ubuntu 18.04: Add Python's PPA repository
sudo add-apt-repository -y ppa:deadsnakes/ppaInstall required packages:
sudo apt update && sudo apt-get install -y \
maven \
openjdk-8-jdk \
python \
python3.8 \
python3-pip \
virtualenv \
make \
tox \
rsync \
openvswitch-switch \
linux-generic-hwe-18.04Install required components for Python 3.6:
sudo pip3 install setuptools-rust==1.1.2 setuptools==46.4.0 --upgradeUpgrade pip3:
sudo pip3 install pip --upgrade(Optional) To avoid version conflict install python3-pip with the official script. To do it, you need to download script & run script:
wget -P /tmp/ https://bootstrap.pypa.io/get-pip.py \
&& sudo python3.8 /tmp/get-pip.pyAfter pip upgrade install Docker compose:
sudo pip3 install docker-composeThe following commands will install necessary dependencies on Ubuntu 20.04:
sudo apt update && sudo apt install -y \
maven \
make \
openjdk-8-jdk \
openvswitch-switch \
python3-pip \
tox \
rsyncUpgrade pip3:
sudo pip3 install pip --upgrade(Optional) To avoid version conflict install python3-pip with the official script. To do it, you need to download script & run script:
wget -P /tmp/ https://bootstrap.pypa.io/get-pip.py \
&& sudo python3.8 /tmp/get-pip.pyAfter pip upgrade install Docker compose:
sudo pip3 install docker-composeYou can either install Gradle, or use Gradle wrapper:
- Option 1: Use Gradle wrapper. The OpenKilda repository contains an instance of Gradle Wrapper which can be used straight from here without further installation.
- Option 2: Install Gradle 7.0 or later versions - https://gradle.org/install/
Note that your build user needs to be a member of the docker group for the build to work. Do that by adding the user to /etc/groups, logging out, and logging in back again.
sudo apt-get install -y ca-certificates curl gnupg lsb-release \
&& sudo mkdir -p /etc/apt/keyrings \
&& curl -fsSL https://download.docker.com/linux/ubuntu/gpg | sudo gpg --dearmor -o /etc/apt/keyrings/docker.gpg \
&& echo "deb [arch=$(dpkg --print-architecture) signed-by=/etc/apt/keyrings/docker.gpg] https://download.docker.com/linux/ubuntu $(lsb_release -cs) stable" \
| sudo tee /etc/apt/sources.list.d/docker.list > /dev/null \
&& sudo apt-get update && sudo apt-get install -y \
docker-ce \
docker-ce-cli \
containerd.io \
docker-compose-plugin \
&& sudo usermod -aG docker $USER
# re-login or reboot to apply the usermod commandYou also need to increase the maven RAM limit at least up to 1G.
export MAVEN_OPTS="-Xmx1g -XX:MaxPermSize=128m"From the base directory, execute the following command:
make build-stableNote that additional Ubuntu packages will be installed as a part of the build process.
From the base directory, execute the following command:
make cleanNB: To run OpenKilda, you should have built it already (see the previous section). This is particularly important because docker-compose will expect that some specific containers already exist.
From the base directory, execute the following command:
make up-stablemake test-topologyBlue-green mode is an implementation of zero downtime feature. In this mode, you have two versions of OpenKilda running: the old one (blue) and the new one (green). Then, at some point in time switch blue to green.
First of all you need to build two sets of images.
To build blue version of OpenKilda you need to run:
make build-stableTo build a green version of OpenKilda you need to run:
make build-latestThese two commands build images with tags stable and latest.
These tags will be used to run OpenKilda in blue mode (from stable images)
or in green mode (for latest images).
There are 3 new commands to run OpenKilda in blue-green mode:
The following command runs OpenKilda in blue mode from stable images. It starts all common components like Zookeeper, database, Kafka, etc.
make up-stableThe next command starts the green version of OpenKilda from the latest images.
Common components wouldn't be restarted (we started them using the previous command).
Floodlight 1 wouldn't be restarted; Floodlight 1 will stay on blue mode.
Only Floodlight 2 will be restarted.
make up-greenThe next command is used to test rollbacks. It runs stable components in blue mode.
The difference with make up-stable is that this command wouldn't start common components
(like Zookeeper, Kafka, etc) and Floodlight 2 (it stays in green mode).
make up-blueAn important aspect of troubleshooting errors and problems in your code is to avoid them in the first place. It's not always that easy, so we should have reliable tools that help in resolving problems. Adding any diagnostic code may be helpful, but there are more convenient ways. Just a few configuration changes will allow us to use a debugging toolkit. As an example, let's take the northbound component. This is a simple REST application providing the interface for interaction with switches, links, flows, features, health-check controllers, and so on. The first thing that we need to do is to add
"-agentlib:jdwp=transport=dt_socket,address=50505,suspend=n,server=y"
to the CMD block in docker/northbound/Dockerfile, where 50505 is any arbitrary port we’ll be using to connect a debugger.
The final file will be similar to the following:
ARG base_image=kilda/base-ubuntu
FROM ${base_image}
ADD BUILD/northbound/libs/northbound.jar /app/
WORKDIR /app
CMD ["java", "-XX:+PrintFlagsFinal", "-XX:+UnlockExperimentalVMOptions", "-XX:+UseCGroupMemoryLimitForHeap", "-agentlib:jdwp=transport=dt_socket,address=50505,suspend=n,server=y", "-jar", "northbound.jar"]Since debugging is done over the network, that also means we need to expose that port in Docker. For that purpose we need
to add "50505:50505" to the northbound the ports block in docker-compose.yml:
northbound:
container_name: openkilda-northbound
build:
context: docker
dockerfile: northbound/Dockerfile
ports:
- '8088:8080'
- '50505:50505'
...
After making these changes, we need to configure a remote debugging in a debugger. For example, in IntelliJ IDEA: navigate to Edit Configurations -> Remote
and set up the debug port as 50505. For more details, please see the documentation of a particular debugger application.
Next, we just run docker-compose up. If everything above was done correctly you must see:
"java -agentlib:jdwp=transport=dt_socket,address=50505,suspend=n,server=y -jar northbound.jar"
in the command column for the open-kilda_northbound. To see the status of the northbound container, we can use the command docker ps -a --no-trunc | grep northbound.
Also, we can check the log record in open-kilda_northbound logs. If the debugger in this container is ready, the following entry will be present in the log:
Listening for transport dt_socket at address: 50505
Now we can run the debugger. The console log in the debugger should contain the following message:
Connected to the target VM, address: 'localhost:50505', transport: 'socket'
To check how debugging works we need to:
- set up a breakpoint,
- make a call to execute some functionality.
In some cases, we must have an approach for debugging a deploy process for a couple (or more) components that interact with each other. Let's
suppose both of them work under docker and some component doesn't belong to us and provided as a library. The typical case:
WorkflowManager (further WFM) and Storm. The approach that is going to be used is almost the same as for northbound but there are
nuances.
First of all, we need to check which version of Storm is used in Open Kilda Controller. For that open docker/storm/Dockerfile
and find the version of Storm. In our case, the Storm version is 1.1.0. To be able to debug Storm we have to clone
the sources from the GitHub repo https://github.com/apache/storm.git and switch to the release 1.1.0.
git checkout -b 1.1.0 e40d213. Information about releases can be found here https://github.com/apache/storm/releases/
Then, go to docker/wfm/Dockerfile and add ENV STORM_JAR_JVM_OPTS "-agentlib:jdwp=transport=dt_socket,server=y,suspend=n,address=50506"
The final file will look similar to the following:
ARG base_image=kilda/storm:latest
FROM ${base_image}
ENV STORM_JAR_JVM_OPTS "-agentlib:jdwp=transport=dt_socket,server=y,suspend=n,address=50506"
...Next, it only remains to add the port 50506 in the WFM container section in the dockerfile as in the example below:
wfm:
container_name: wfm
build:
context: docker
dockerfile: wfm/Dockerfile
...
ports:
- "50506:50506"
After executing docker-compose up you should see the following log record Listening for transport dt_socket at address: 50506
in the WFM logs.
Now, after we configure the remote debugger to connect to the port 50506, we'll be able to debug both components: WFM and Storm.
For example, in IntelliJ IDEA: navigate to Edit Configurations -> Remote and set up the debug port as 50506.
For more details, please see the documentation of a particular debugger application.
In order to debug a topology, for example, NetworkTopology:
create (or open if already exists) NetworkTopology.main application debug configuration and add --local to Program arguments,
execute docker-compose up and run in the debug mode NetworkTopology.main.
Please refer to the Testing section on our Wiki.
Start with the following:
make unitFrom there, you can go to specific projects to build / develop / unit test.
Just follow the make unit trail. Most projects have a gradle build or maven target.
Look in the docker/hacks/usecase directory. You'll find several makefiles that will assist
with the development and testing of that use case.
As an example, you can execute the following command for more information on the network discovery use case:
make -f docker/hacks/usecase/network.disco.make helpor
cd docker/hacks/usecase
make -f network.disco.make helpVirtualBox and Vagrant are popular options for creating VMs. A VM may be your best bet for doing development with OpenKilda. There are a set of files in the source tree that will facilitate this.
- NB1: Ensure you have VirtualBox and Vagrant installed and on the path
- NB2: It is recommended to clone OpenKilda again, to ensure there are no any write permission issues between the guest VM and the host.
Steps:
- From the root directory, look at the Vagrantfile; feel free to change its parameters.
vagrant up: create the VM; it'll be running after this step.vagrant ssh: this will log you into the vm.ssh-keygen -t rsa -C "your_email@example.com": you'll use this for GitHub.- Add the ~/.ssh/id-rsa.pub key to your GitHub account so that you can clone OpenKilda.
- Clone and Build
# NB: Instead of putting it in vm-dev, you can use /vagrant/vm-dev
# This has the added benefit that the code will appear outside of the VM
# because /vagrant is a shared directory with the same directory on the host where the Vagrantfile is located.
git clone git@github.com:<your_github_account>/open-kilda.git vm-dev
cd vm-dev
git checkout mvp1rc
make build-base
docker-compose build
make unit
make up-test-mode
We have confd for managing config/properties files from templates. Confd configs, templates and variable file stored in confd/ folder.
confd/conf.d/*.toml: files with the description on how to process templates (src. path, dst.path.... etc).
confd/templates/*/*.tmpl: templates in the go-template format.
confd/vars/main.yaml: a file with all variables to be substituted in templates.
- Create and place the template file to
confd/templates/directory. - Create and place the template description in
confd/conf.d/directory. - Change vars in
confd/main.yamlif needed. - Execute:
make update-props-dryrunfor verifying that the templates can be processed. - Execute:
make update-propsfor applying the templates.
Let's suppose, you have an OrientDB server, and you want to use it instead of the dockerized OrientDB.
You can add OrientDB endpoints to confd/vars/main.yaml and create properties template for services which use OrientDB:
confd/conf.d/base-storm-topology.topologies.toml:
[template]
src = "base-storm-topology/topology.properties.tmpl"
dest = "src-java/base-topology/base-storm-topology/src/release/resources/topology.properties"
keys = [ "/" ]
mode = "0644"confd/vars/main.yaml:
kilda_orientdb_hosts: "odb1.pendev,odb2.pendev,odb3.pendev"
kilda_orientdb_hosts_single: "odb1.pendev"
kilda_orientdb_user: "kilda"
kilda_orientdb_password: "kilda"confd/templates/base-storm-topology/topology.properties.tmpl
...
{{if not (exists "/single_orientdb")}}
orientdb.url=remote:{{ getv "/kilda_orientdb_hosts" }}/{{ getv "/kilda_orientdb_database" }}
{{else}}
orientdb.url=remote:{{ getv "/kilda_orientdb_hosts_single" }}/{{ getv "/kilda_orientdb_database" }}
{{end}}
orientdb.user = {{ getv "/kilda_orientdb_user" }}
orientdb.password = {{ getv "/kilda_orientdb_password" }}
...
In this example, we will generate the file src-java/base-topology/base-storm-topology/src/release/resources/topology.properties
from the template confd/templates/base-storm-topology/topology.properties.tmpl