It seems like it was only yesterday that the OpenStack community found itself gathering in Hong Kong to set the design goals for the Icehouse release. As we entered March development was still progressing at a fever pitch in the lead up to the feature freeze for the release but now the dust has started to settle and we are able to start getting a real feel for what OpenStack users and operators can look forward to in the Icehouse release.
Today I'll be giving a sneak peak to just some of the changes made in one of the two projects that made up the original OpenStack release and today is still one of the largest - showing no signs of the innovation slowing down - OpenStack Compute (Nova). OpenStack Compute is a cloud computing fabric controller, a central component of an Infrastructure as a Service (IaaS) system. It is responsible for managing the hypervisors on which virtual machine instances will ultimately run and managing the lifecycle of those virtual machine instances. This list is by no means exhaustive but highlights some key features and the rapid advances made by the contributors that make up the OpenStack community in a six month release cycle.

Libvirt/Kernel-based Virtual Machine (KVM) Driver Enhancements

The OpenStack User Survey results presented in Hong Kong indicated a whopping 62% of respondents are using the Libvirt/KVM hypervisor to power the compute services offered by their OpenStack clouds. The power of this combination of the Libvirt virtualization abstraction layer with the performance and security offered by the KVM hypervisor has been cemented in the datacenter and now extends to the elastic cloud. OpenStack contributors have continued in the Icehouse release to find new and innovative ways to expose the functionality provided by this combination of technologies for consumption by operators and users of elastic OpenStack clouds, delivering a number of tangible features in the abstraction layer provided by this compute driver:
  • It is now possible to add a Virtio RNG device to compute instances to provide increased entropy. Virtio RNG is a paravirtual random number generation device. It allows the compute node to provide entropy to the compute instances in order to fill their entropy pool. The default entropy device used on the host is /dev/random, however, use of a hardware RNG device physically attached to the host is also possible. The use of the Virtio RNG device is enabled using the hw_rng property in the metadata of the image used to build the instance.
  • Watchdog support has been added, allowing the triggering of instance lifecycle events based on a crash or kernel panic detected within the instance. The watchdog device used is a i6300esb. It is enabled by setting the hw_watchdog_action property in the image properties or flavor extra specifications to a value other than disabled. Supported hw_watchdog_action property values, which denote the action for the watchdog device to take in the event of detecting an instance failure, are poweroff, reset, pause, and none.
  • It is now possible to configure instances to use video driver other than the default. This allows the specification of different video driver models, different amounts of video RAM, and different numbers of video heads. These values are configured by setting the hw_video_model, and hw_video_vram properties respectively in the image metadata. The number of heads is configured from within the guest operating system. Currently supported video driver models are vga, cirrus, vmvga, xen and qxl.
  • Modified kernel arguments can now be provided to booting compute instances. The kernel arguments are retrieved from the os_command_line key in the image metadata as stored in the OpenStack Image Service (Glance), if a value for the key was provided. If no value is provided then the default kernel arguments continue to be used.
  • VirtIO SCSI (virtio-scsi) can now be used instead of VirtIO Block (virtio-blk) to provide block device access for instances. Virtio SCSI is a para-virtualized SCSI controller device designed as a future successor to VirtIO Block and aiming to provide improved scalability and performance. VirtIO SCSI is enabled for a guest instance by setting hw_disk_bus_model to virtio-scsi in the image properties.
  • Changes have been made to the expected format of the /etc/nova/nova.conf configuration file with a view to ensuring that all configuration groups in the file use descriptive names. A number of driver specific flags, including those for the Libvirt driver, have also been moved to their own option groups.

Compute API Enhancements

A number of contributors have been hard at work both at extending the Compute v2 API while work also continued on an updated API that might one day displace it, Compute v3. This has been a hot topic in recent weeks as contributors  discussed ways to carefully balance the desire for API innovation with the needs of operators and users.
In the meantime development of the Compute v2 API continues, some key extensions and changes made in the Icehouse release are:
  • API facilities have been added for defining, listing, and retrieving the details of instance groups. Instance groups provide a facility for grouping related virtual machine instances at boot time and applying policies to determine how they must be scheduled in relation to other members of the group. Currently supported policies are affinity, which indicates all instances in the group should be scheduled to the same host, and anti-affinity, which indicates all instances in the group should be scheduled on separate hosts. Retrieving the details of an instance group using the updated API also returns the list of group members.
  • The Compute API now exposes a mechanism for permanently removing decommissioned compute nodes. Previously these would continue to be listed even where the compute service had been disabled and the system re-provisioned. This functionality is provided by the ExtendedServicesDelete API extension.
  • The Compute API now exposes the hypervisor IP address, allowing it to be retrieved by administrators using the "nova hypervisor-show" command.
  • The Compute API currently supports both XML and JSON formats. Support for the XML format has now been marked as deprecated and will be retired in a future release.

Notifications

Compute support for notifications has been growing with  every release, with more and more actions gradually being modified to generate a notification that operators, users, and orchestration systems can capture to track events. The notable notifications added in the Icehouse release are:
  • Notifications are now generated when a Compute host is enabled, disabled, powered on, shut down, rebooted, put into maintenance mode and taken out of maintenance mode.
  • Notifications are now generated upon the creation and deletion of keypairs.

Scheduler

The compute scheduler is responsible for determining which compute nodes a launched instance will be placed on based on a series of configurable filters and weights. While efforts remain under way to further decouple the scheduler from Nova, this remains an area of rich innovation within the project. In the Icehouse release:
  • Modifications have been made to the scheduler to add an extensible framework allowing for it to make decisions based on resource utilization. In coming releases expect to see more development in this space, particular as it is extended to handle specific resource classes.
  • An initial experimental implementation of a caching scheduler driver was added. The caching scheduler uses the existing facilities for applying scheduler filters and weights but caches the list of available hosts. When a user request is passed to the caching scheduler it attempts to perform scheduling based on the list of cached hosts, with a view to improving scheduler performance.
  • A new scheduler filter, AggregateImagePropertiesIsolation, has been introduced. The new filter schedules instances to hosts based on matching namespaced image properties with host aggregate properties. Hosts that do not belong to any host aggregate remain valid scheduling targets for instances based on all images. The new Compute service configuration keys aggregate_image_properties_isolation_namespace and aggregate_image_properties_isolation_separator are used to determine which image properties are examined by the filter.

Testing

During the Icehouse release cycle work has continued to facilitate third party testing of hypervisor drivers that live in the OpenStack Compute source tree. This allows third parties to provide continuous integration (CI) infrastructure to run regression tests against each proposed OpenStack Compute patch and record the results so that they can be referred to as the code is reviewed. This ensures not only test coverage for these drivers but valuable additional test coverage of the shared componentry provided by the OpenStack Compute project itself.

Upgrades

The Compute services now allow for a level of rolling upgrade, whereby control services can be upgraded to Icehouse while they continue to interact with compute services running code from the Havana release. This allows for a more gradual approach to upgrading an OpenStack cloud, or logical designated subset thereof, than has typically been possible in the past.

Work on stabilizing Icehouse will go on for some time now before the community gathers again in May for OpenStack Summit 2014 in Atlanta to define the design vision for the next six month release cycle. If you want to help test out some of the above features, why not get involved in the upcoming RDO test day using freshly baked packages based on the third Icehouse release milestone?