SUSE Now Offers Non-Disruptive Upgrades for OpenStack
SUSE has just made it a lot easier to upgrade the company’s OpenStack distribution, SUSE OpenStack Cloud 6 (SOC 6).
“If enterprise customers want to move to a new version of OpenStack they don’t have to replace and rebuild; they can now do a normal upgrade from an older version of OpenStack cloud to a newer version,” said SUSE CEO Nils Brauckmann. “What it means is that they can easily move with OpenStack innovation.”
One of the big sticking points from further enterprise adoption of OpenStack has been the upgrade process, which is onerous. With this release, “SUSE is the first and only provider to offer an uninterrupted upgrade path to new versions, starting with SUSE OpenStack Cloud 6,” Brauckmann said.
This release is based on OpenStack Liberty and built on SUSE Linux Enterprise (SLE) Server 12 SP1, the latest version of the top SLE. What it means is that SUSE can provide support for the entire stack — including the host Linux and guest operating systems.
SUSE has been working on this release for awhile, releasing a beta last year. The company has been through some belt-tightening, though after being purchased by Micro Focus in 2014, SUSE has been regaining strength; the company is now investing heavily in innovation and it reflected in products like OpenStack, Brauckmann said.
With this release, SUSE brings that non-disruptive upgrade capabilities for future OpenStack releases. The company is also aiming for a more business-oriented release cycle for easier migration processes and less disruption to production environments.
“Now that customers are using OpenStack for more and more production workloads, meeting enterprise SLAs [service level agreements] becomes an important consideration,” wrote Pete Chadwick, SUSE senior product manager for cloud infrastructure.
SUSE routinely prioritizes building out deployment frameworks to maximize uptime and minimize disruption during migration and upgrade cycles, Chadwick explained. The company has built high availability extensions and live kernel patching capabilities into its Linux distribution, SUSE Linux Enterprise Server.
SUSE OpenStack Cloud 6 leverages these key high availability features and provides the architecture to support non-disruptive upgrades for future releases, Chadwick said.
But are there real challenges for upgrading in OpenStack? What about that extremely popular video from the OpenStack Foundation, showing engineers upgrading an OpenStack system in 10 minutes. True, all updates are not same, depending on the organization, infrastructure and other factors.
The upgrade in the video showed a process that was designed for a single customer environment, pointed out Chadwick and Mark Smith, Global Product and Solutions Manager at SUSE, when I spoke to them about the existing challenges of OpenStack upgrade.
Contrary to single customer OpenStack environment SUSE needs to define a process that can be used by a wider variety of customer deployment scenarios that includes OpenStack, partner drivers and the underlying OS and hypervisor explained the SUSE managers.
They also pointed out that the video was not truly non-disruptive; there was (minimal) downtime for the OpenStack control plane.
“OpenStack allows a lot of flexibility in how it is deployed, configured and utilized. There are many configuration assumptions that can be made in a pre-defined scenario that would simply not apply in real world installations.
There is a big difference between a process that can update a single, well defined OpenStack install and solve the upgrade problem for our customers,” Smith said.
In addition to that, the 10-minute upgrade video didn’t include Neutron, OpenStack networking component. When there is VM network traffic running through Neutron, shutting Neutron is the same as shutting down the data plane. In such cases, both the OpenStack control and data plane would need to be taken offline to upgrade an OpenStack instance.
“Since most customers are using Neutron, a non-disruptive upgrade has to eliminate downtime even for the control plane,” Smith said.
True, it could be argued that being unable to start or stop a VM for 10 minutes is not a big problem. However, any downtime is likely to increase with the size of the cloud. If that time were to increase significantly, to say, 30 minutes, that is likely to be an issue for most customers, the SUSE managers noted.
And even with a 10-minute outage on the control plane, there is a risk that if something were to change, i.e. a server goes down, the absence of scheduler communication would result in either immediate issues, Smith noted. As a result, workloads don’t get migrated, or additional problems may surface after the control plane is restarted, such as NOVA scheduling workloads to a server that is offline.
Service downtime can be eliminated through good design, but taking nodes (virtual or physical) out of the control cluster and bringing them back gracefully is challenging. And such challenges can grow to be even more complex as workloads begin to take advantage of advanced orchestration features such as Heat and Ceilometer.
A more complex upgrade scenario also needs to go beyond just the OpenStack layer. It needs to include upgrading the deployment framework so that users can continue using the deployment tools for further changes to their setups. There is also an additional challenge with the data plane upgrading the operating system and hypervisor on the compute nodes brings new challenges in ensuring the ongoing integrity of the data plane.
These are some of the areas that SUSE is addressing with its upgrade process, according to Smith and Chadwick:
• Control plane: Upgrade services without disrupting the operation of VMs or preventing loss of control plane state.
• Data plane (or compute cluster): Live migration of VMs to evacuate compute nodes for full upgrades of the entire compute stack including OS, hypervisor and OpenStack components.
• Database: Schema update and data migration.
• Deployment framework: The deployment framework contains the state of the physical cluster including where services are deployed. This needs to be upgraded to include new OpenStack services. Similar to the migration of the OpenStack databases, this upgrade must be done with no loss of state. In SUSE OpenStack Cloud 6 comes with changes to the framework that enables an upgrade to the framework without disrupting the cloud. Once the framework has been upgraded, it then automates the upgrade process of the cloud.
The purpose of the deployment framework, Chadwick and Smith explain, is to provide a structured process for installing OpenStack using a set of predefined scripts and descriptor files that executes a full deployment through a well-defined and repeatable process. Default options are assumed to minimize the options though the package still provides the flexibility to address a wide range of scenarios.
What Else is New?
SUSE OpenStack Cloud 6 now supports IBM z/VM for compute nodes. SUSE said that this is a direct response to customer requests to be able to incorporate their mainframe into their OpenStack cloud. In terms of hypervisors, SUSE OpenStack Cloud already supports KVM, Xen, Microsoft Hyper-V, VMware vCenter and now IBM z/VM.
SUSE has support for Linux containers since the very first release of SUSE OpenStack cloud and with this release, they are adding deployment capabilities for Docker containers and workloads.
The company has also introduced a new web-based installation framework and role-based management for an improved user experience. It also supports Shared File-as-a-Service within the cloud, called Manila.
“We are excited about the release of SUSE OpenStack Cloud 6 with Manila support,” Jeff O’Neal, senior director of the Cloud Solutions Group at NetApp, said. “As the founder of the Manila project and a charter member of the OpenStack Foundation, we have contributed deep expertise and technology that will enable shared volumes between public and private clouds for our Data ONTAP customers, a feature many enterprises have been looking for to accelerate their cloud adoption.”
SUSE OpenStack 6 is available immediately.
Feature image via Pixabay.
