dehotot / hbase-standalone

a distributed hbase cluster that doesn't need hdfs

hbase-standalone

hbase cluster just needs a reliable shared filesystem between nodes. The shared filesystem doesn't need to be hdfs!

this deployment

This is a small-to-medium sized deployment of hbase without hadoop/hdfs.

It's a proof-of-concept to show that you can deploy a minimal hbase cluster like this.

Kubernetes configs in the k8s/ directory will run a cluster of 1 hbase master and 2 regionserver nodes.

As you can tell from some of the DNS namespaces it was originally intended to be the backend for an OpenTSDB db.

what, no hadoop/hdfs?

If you have a reliable cluster filesystem such as an NFS on Netapp, or Lustre, GPFS etc, you can create a hbase cluster without hadoop hdfs.

what is included

The no-hdfs hbase deployment is described using Docker images.

The images expect to find each other by DNS names.

You might need to change them for your own needs:

• hmaster-0.hmaster.opentsdb.svc.cluster.local
• regionserver-0.regionserver.opentsdb.svc.cluster.local
• regionserver-1.regionserver.opentsdb.svc.cluster.local

building

1. download hbase-1.2.6-bin.tar.gz (the most recent Stable version) from hbase into bin/
2. docker build -t 'dehotot/hbase-standalone:latest' .

Running in Docker

hbase cluster hmaster

docker run -it dehotot/hbase-standalone:latest /start-hmaster.sh

hbase cluster regionserver node

docker run -it dehotot/hbase-standalone:latest /start-regionserver.sh

Deploying in Kubernetes

You will need to build your own images from the dockerfiles provided.

Then you need:

• an opentsdb namespace
• a PersistentVolume and PersistentVolumeClaim called metrics

An example is provided in k8s/example-namespace.yaml

to deploy on Kubernetes

1. edit and apply example-namespace.yaml to set up the namespace and your NFS share
2. deploy the hmaster: kubectl apply -f k8s/hmaster.yaml
3. deploy the regionservers: kubectl apply -f k8s/regionservers.yaml
4. watch the logs on the hmaster to see the nodes join: kubectl logs -f hmaster0
5. if you make the hbase Service accessible (we use an externalIP on our bare-metal, you might use a LoadBalancer in the cloud) then you can view the cluster status web gui here.

scaling up

1. add the next logical worker to conf/regionservers
2. update the replicas value in k8s/regionservers.yaml
3. apply the new regionservers.yaml

scaling down

1. i've not had to do it yet
2. you will need to tell hbase to remove one of its nodes before you scale down in k8s

opentsdb namespace

This namespace is baked into the hbase cluster configuration, and therefore the Service entries.

If you want to use a different namespace, you need to change conf/regionservers, conf/hbase-site.xml and the namespace inside the k8s yamls.

zookeeper

Because Kubernetes gives reliablily to the node roles, and we use this cluster to backend OpenTSDB, only one Zookeeper is needed for us.

We use the Zookeeper that comes prebuilt with hbase.

intended purpose

The intended purpose is only to teach and demonstrate that you can run hbase without hdfs.

It's not intended for production. You need to evaluate the advice for your own purpose and make it usable for yourself.

No warranty is given or implied.

It is not advised to use it to run your local particle accelerator ;)

thanks

Thanks to https://github.com/HariSekhon @HariSekhon for the original inspiration and starting point.