How to run a Sahara cluster on bare metal servers

How to run a Sahara cluster on bare metal servers

Hadoop clusters are designed to store and analyze extremely large amounts of unstructured data in distributed computing environments. Sahara enables you to boot Hadoop clusters in both virtual and bare metal environments. When Booting Hadoop clusters with Sahara on bare metal servers, you benefit from the bare metal performance with self-service resource provisioning.

  1. Create a new OpenStack environment using Devstack as described in the Devstack Guide
  2. Install Ironic as described in the Ironic Installation Guide
  3. Install Sahara as described in the Sahara Installation Guide
  4. Build the Sahara image and prepare it for uploading to Glance:
$ qemu-img convert -O raw image-converted.qcow image-converted-from-qcow2.raw
  • Mount the raw image to the system.
  • chroot to the mounted directory and remove the installed grub.
  • Build grub2 from sources and install to /usr/sbin.
  • In /etc/sysconfig/selinux, disable selinux SELINUX=disabled
  • In the configuration file, set onboot=yes and BOOTPROTO=dhcp for every interface.
  • Add the configuration files for all interfaces in the /etc/sysconfig/network-scripts directory.
  1. Upload the Sahara disk image to Glance, and register it in the Sahara Image Registry. Referencing its separate kernel and initramfs images.
  2. Configure the bare metal network for the Sahara cluster nodes:
    • Add bare metal servers to your environment manually referencing their IPMI addresses (Ironic does not detect servers), for Ironic to manage the servers power and network. For example:
$ ironic node-create -d pxe_ipmitool \
$ -i ipmi_address=$IP_ADDRESS \
$ -i ipmi_username=$USERNAME \
$ -i ipmi_password=$PASSWORD \
$ -i pxe_deploy_kernel=$ \
$ -i pxe_deploy_ramdisk=$

$ ironic port-create -n $NODE_ID -a "$MAC_eth1"
  • Add the hardware information:
$ ironic node-update $NODE_ID add properties/cpus=$CPU \
$ properties/memory_mb=$RAM properties/local_gb=$ROOT_GB \
$ properties/cpu_arch='x86_64'
  1. Add a special flavor for the bare metal instances with an arch meta parameter to match the virtual architecture of the server’s CPU with the metal one. For example:
$ nova flavor-create baremetal auto $RAM $DISK_GB $CPU
$ nova flavor-key baremetal set cpu_arch=x86_64


The vCPU ad vRAM parameters (x86_64 in the example) will not be applied because the operating system has access to the real CPU cores and RAM. Only the root disk parameter is applied, and Ironic will resize the root disk partition. Ironic supports only a flat network topology for the bare metal provisioning, you must use Neutron to configure it.

  1. Launch your Sahara cluster on Ironic from the cluster template:
    • Log in to Horizon.
    • Go to Data Processing > Node Group Templates.
      • Find the templates that belong to the plugin you would like to use
      • Update those templates to use ‘bare metal’ flavor instead of the default one
    • Go to Data Processing > Cluster Templates.
    • Click Launch Cluster.
    • On the Launch Cluster dialog:
      • Specify the bare metal network for cluster nodes

The cluster provisioning time is slower compared to the cluster provisioning of the same size that runs on VMs. Ironic does real hardware reports which is time consuming, and the whole root disk is filled from /dev/zero for security reasons.

Known limitations:

  • Security groups are not applied.
  • When booting a nova instance with a bare metal flavor, the user can not provide a pre-created neutron port to nova boot command. LP1544195
  • Nodes are not isolated by projects.
  • VM to Bare Metal network routing is not allowed.
  • The user has to specify the count of ironic nodes before Devstack deploys an OpenStack.
  • The user cannot use the same image for several ironic node types. For example, if there are 3 ironic node types, the user has to create 3 images and 3 flavors.
  • Multiple interfaces on a single node are not supported. Devstack configures only one interface.
Creative Commons Attribution 3.0 License

Except where otherwise noted, this document is licensed under Creative Commons Attribution 3.0 License. See all OpenStack Legal Documents.