View Only

PowerVC 2.0.2 – Control plane scale and resiliency

By Shyama Venugopal posted Thu September 30, 2021 07:05 AM


PowerVC 2.0.2 is released in Sept 2021 with all new high availability and scale architecture. This blog addresses what has changed and what has not changed with the new PowerVC version. 

PowerVC, built on OpenStack, was an “All-in-one deployment” up until version 2.0.1 – one node where all the components were installed. This deployment model did not meet the scale and HA requirements of large data centers.

PowerVC 2.0.2 opens up more possibilities allowing management of higher number of hosts and adding resiliency by moving to a multinode architecture.

New components

PowerVC can be installed as a single node instance or a 3-node instance starting 2.0.2 release. This version brings in additional clustering components.

  • Virtual IP
  • Pacemaker
  • Corosync
  • Haproxy
  • Galera cluster
  • Rabbitmq cluster

These components help in monitoring of all PowerVC & OpenStack services and load balancing the incoming requests. The data plane consisting of mariadb and rabbitmq are setup in a clustered fashion. All communication is done over the virtual IP. With this architecture, at any point of time if one of the nodes of PowerVC is unavailable the requests are routed to the remaining 2 nodes.

If PowerVC is setup as an one-node instance all the clustering components are still installed. User can later choose to add 2 more nodes to this one-node cluster and make it a multinode setup.  The Virtual IP makes it possible to seamlessly scale up the number of nodes and hence is mandatory for PowerVC install.

Clustering PowerVC and OpenStack services

The PowerVC and OpenStack services run in a clustered fashion behind the haproxy load balancer. The services are added to pacemaker as resources and monitored by Pacemaker and Corosync. Most of the PowerVC services run in an Active-Active fashion while few of them are configured Active-Passive.

Listed below are services running as Active-Passive

  • powervc-bumblebee
  • openstack-nova-ibm-ego-ha-service
  • openstack-nova-ibm-ego-resource-optimization
  • powervc-cinder-ibm-health
  • powervc-neutron-ibm-health
  • powervc-nova-ibm-health
  • openstack-ceilometer-notification
  • openstack-ceilometer-polling
  • panko-api
  • ttv-validation-api
  • openstack-cinder-volume

For the services running in Active-Passive, in case of failure on the Active node the service will be started on another available node. This move will be done automatically by Pacemaker.

New way to manage the PowerVC cluster

Since the new architecture brings in additional nodes and more components this also warrants new ways to manage the cluster. The powervc-opsmgr CLI provides commands to create inventory, install, uninstall, scale-up, upgrade of PowerVC nodes. Internally, the powervc-opsmgr uses Ansible for deployment.

What has not changed

With all the new components what happens to the orchestrators or automation tools built on PowerVC? Will they still work? The answer is Yes. There is NO change in the GUI or API end points. All the endpoints will now be accessed using the virtual IP. No other impact to the existing functionality.

Watch this space for more detailed blogs on the new powervc-opsmgr CLI and related functionality.

For PowerVC 2.0.2 installation procedure, read this blog. Please reach out to us if you have any queries or comments. 

Knowledge center link: PowerVC
Don't forget to follow us on LinkedInTwitterYouTube, and Facebook.

Blog Author:
Shyama Venugopal