A version of Kubernetes using Apache Hadoop YARN as the scheduler. Integrating Kubernetes with YARN lets users run Docker containers packaged as pods (using Kubernetes) and YARN applications (using YARN), while ensuring common resource management across these (PaaS and data) workloads.
Kubernetes-YARN is currently in the protoype/alpha phase
This integration is under development. Please expect bugs and significant changes as we work towards making things more stable and adding additional features.
Kubernetes and Kubernetes-YARN are written in Go. Currently, vagrant and ansible based setup mechanims are supported. The instructions below are for creating a vagrant based cluster. For ansible instructions, see here.
Please ensure you have boot2docker, Go (at least 1.3), Vagrant (at least 1.6), VirtualBox (at least 4.3.x) and git installed. Run boot2docker to bring up a VM with a running docker daemon (this is used for building release binaries for Kubernetes).
$ $(boot2docker shellinit) #sets up docker env vars $ echo $GOPATH /home/user/goproj $ mkdir -p $GOPATH/src/github.com/hortonworks/ $ cd $GOPATH/src/github.com/hortonworks/ $ git clone firstname.lastname@example.org:hortonworks/kubernetes-yarn.git $ cd kubernetes-yarn $ build/release.sh #builds kubernetes release binaries $ hack/build-go.sh #builds kubernetes client binaries $ cluster/kube-up.sh #brings up kubernetes cluster
Following these steps will bring up a multi-VM cluster (1 master and 3 minions, by default) running Kubernetes and YARN. Please note that, depending on your local hardware and available bandwidth, bringing the cluster up could take a while to complete.
By default, the kubernetes master is assigned the IP 10.245.1.2. The YARN resource manager runs on the name host. Once the vagrant cluster is running, the YARN dashboard accessible at http://10.245.1.2:8088/
The HDFS dashboard is accessible at http://10.245.1.2:50070/
Interacting with the Kubernetes-YARN cluster
Creating pods/running Docker containters
For instructions on creating pods, running containers and other interactions with the cluster, please see Kubernetes' vagrant instructions here
Running a test map-reduce job
In order to run a test map-reduce job, log into the cluster (ensure that you are in the
kubernetes-yarn directory) and run the included test script.
$ vagrant ssh master [vagrant@kubernetes-master ~]$ cd hadoop/install/hadoop-2.6.0-SNAPSHOT/ [vagrant@kubernetes-master hadoop-2.6.0-SNAPSHOT]$ sudo su #you need to be root to run this test script [root@kubernetes-master hadoop-2.6.0-SNAPSHOT]# ./test-pi-yarn.sh