The mesos/docker provider uses docker-compose to launch Kubernetes as a Mesos framework, running in docker with its dependencies (etcd & mesos).
The cluster consists of several docker containers linked together by docker-managed hostnames:
|docker-grand-ambassador||Proxy to allow circular hostname linking in docker|
|etcd||etcd||Key/Value store used by Mesos|
|Mesos Master||mesosmaster1||REST endpoint for interacting with Mesos|
|Mesos Slave (x2)||mesosslave1, mesosslave2||Mesos agents that offer resources and run framework executors (e.g. Kubernetes Kublets)|
|Kubernetes API Server||apiserver||REST endpoint for interacting with Kubernetes|
|Kubernetes Controller Manager||controller|
|Kubernetes Scheduler||scheduler||Schedules container deployment by accepting Mesos offers|
It’s possible to install all of the above via Homebrew on a Mac.
Some steps print instructions for configuring or launching. Make sure each is properly set up before continuing to the next step.
brew install git brew install caskroom/cask/brew-cask brew cask install virtualbox brew install docker brew install docker-machine brew install docker-compose
Most of the above are available via apt and yum, but depending on your distribution, you may have to install via other means to get the latest versions.
It is recommended to use Ubuntu, simply because it best supports AUFS, used by docker to mount volumes. Alternate file systems may not fully support docker-in-docker.
In order to build Kubernetes, the current user must be in a docker group with sudo privileges. See the docker docs for instructions.
If on a Mac using docker-machine, the following steps will make the docker IPs (in the virtualbox VM) reachable from the host machine (Mac).
docker-machine create --driver virtualbox kube-dev eval "$(docker-machine env kube-dev)"
Set the VM’s host-only network to “promiscuous mode”:
docker-machine stop kube-dev
VBoxManage modifyvm kube-dev --nicpromisc2 allow-all
docker-machine start kube-dev
This allows the VM to accept packets that were sent to a different IP.
Since the host-only network routes traffic between VMs and the host, other VMs will also be able to access the docker IPs, if they have the following route.
Route traffic to docker through the docker-machine IP:
sudo route -n add -net 172.17.0.0 $(docker-machine ip kube-dev)
Since the docker-machine IP can change when the VM is restarted, this route may need to be updated over time. To delete the route later: `sudo route delete 172.17.0.0`
git clone https://github.com/kubernetes/kubernetes
By default, that will get you the bleeding edge of master branch. You may want a release branch instead, if you have trouble with master.
You’ll need to build kubectl (CLI) for your local architecture and operating system and the rest of the server binaries for linux/amd64.
Building a new release covers both cases:
For developers, it may be faster to build locally.
[Optional] Build docker images
The following docker images are built as part of
./cluster/kube-up.sh, but it may make sense to build them manually the first time because it may take a while.
Test image includes all the dependencies required for running e2e tests.
In the future, this image may be available to download. It doesn’t contain anything specific to the current release, except its build dependencies.
Kubernetes-Mesos image includes the compiled linux binaries.
This image needs to be built every time you recompile the server binaries.
[Optional] Configure Mesos resources
By default, the mesos-slaves are configured to offer a fixed amount of resources (cpus, memory, disk, ports).
If you want to customize these values, update the
MESOS_RESOURCES environment variables in
If you delete the
MESOS_RESOURCES environment variables, the resource amounts will be auto-detected based on the host resources, which will over-provision by > 2x.
If the configured resources are not available on the host, you may want to increase the resources available to Docker Engine. You may have to increase you VM disk, memory, or cpu allocation. See the Docker Machine docs for details (Virtualbox)
This tells cluster scripts to use the code within
If you manually built all the above docker images, you can skip that step during kube-up:
After deploying the cluster,
~/.kube/config will be created or updated to configure kubectl to target the new cluster.
To learn more about Pods, Volumes, Labels, Services, and Replication Controllers, start with the Kubernetes Walkthrough.
To skip to a more advanced example, see the Guestbook Example
kube-up for the mesos/docker provider will automatically deploy KubeDNS and KubeUI addons as pods/services.
Check their status with:
./cluster/kubectl.sh get pods --namespace=kube-system
The web-based Kubernetes UI is accessible in a browser through the API Server proxy:
By default, basic-auth is configured with user
admin and password
The IP of the API Server can be found using
Warning: e2e tests can take a long time to run. You may not want to run them immediately if you’re just getting started.
While your cluster is up, you can run the end-to-end tests:
- Increase the logging verbosity:
- Run only a subset of the tests (regex matching):
To build, deploy, test, and destroy, all in one command (plus unit & integration tests):
When compiling from source, it’s simplest to use the
./cluster/kubectl.sh script, which detects your platform &
architecture and proxies commands to the appropriate
./cluster/kubectl.sh get pods
Kill all docker containers
docker ps -q -a | xargs docker rm -f
Clean up unused docker volumes
docker run -v /var/run/docker.sock:/var/run/docker.sock -v /var/lib/docker:/var/lib/docker --rm martin/docker-cleanup-volumes
The steps above tell you how to build in a container, for minimal local dependencies. But if you have Go and Make installed you can build locally much faster:
However, if you’re not on linux, you’ll still need to compile the linux/amd64 server binaries:
KUBERNETES_CONTRIB=mesos build/run.sh hack/build-go.sh
The above two steps should be significantly faster than cross-compiling a whole new release for every supported platform (which is what
KUBERNETES_CONTRIB=mesos- enables building of the contrib/mesos binaries
hack/build-go.sh- builds the Go binaries for the current architecture (linux/amd64 when in a docker container)
make- delegates to
build/run.sh- executes a command in the build container
build/release.sh- cross compiles Kubernetes for all supported architectures and operating systems (slow)
|IaaS Provider||Config. Mgmt||OS||Networking||Docs||Conforms||Support Level|
|Mesos/Docker||custom||Ubuntu||Docker||docs||Community (Kubernetes-Mesos Authors)|
For support level information on all solutions, see the Table of solutions chart.