The pacemaker-cloud project demonstrates the current community work in providing application service high availability in a cloud environment.
- Name: Steven Dake
- Email: <email@example.com>
- Targeted release: Fedora 16
- Last updated: (July 20th, 2011)
- Percentage of completion: 95%
We will not be able to achieve 100% complete until we do an update that adds the F16 iso support which requires a bootable rawhide iso image. The remainder of the software is complete.
The software provides a user interface shell called pcloudsh which provides:
- Create deployables including:
- Create a JEOS image of F14, F15, F16, RHEL6
- Create an assembly of F14, F15, F16, RHEL6
- Add assemblies to a deployable
- Add managed resources to an assembly
- Launch a deployable, including all of its assembly images
- Provides user interface feedback when an application or assembly fails and describe which corrective actions are taken.
The software provides daemons and init scripts which provide high availability of the deployables configured in the system:
- Kill/restart applications if a failure is detected.
- Kill and restart assemblies if an assembly failure is detected.
- JEOS - just enough operating system - the bare minimum operating system required to boot a virtual machine image
- Assembly - Composition of a JEOS image and managed resources
- Deployable - Collection of assemblies that represent all virtual machines required to provide a specific service
- Resource - Daemon application, such as Apache's httpd service, which is managed for high availability
- high availability - Applying the techniques of:
- monitoring a component for failure
- forcibly terminating a component when a failure has been detected
- restarting the failed component
- providing notification to system administration so they may repair the underlying fault
- See Pacemaker Cloud Project Slides for more details.
Benefit to Fedora
This feature provides a preview of high availability for cloud environments using a building block that is reusable in other cloud management systems. This feature provides only single node deployable high availability, but for F17 we plan to integrate with other distributed cloud management tools such as Aeolus.
This is a standalone package but has several dependencies on other parts of Fedora 16. We are in good shape relatng to dependencies, however, systemd is not LSB compliant currently resulting in our software not being able to provide high availability for F15 or Rawhide guests.
We are nearing code completion for the single node case and have some basic packaging done.
How To Test
yum install pacemaker-cloud chkconfig pacemaker-cloud on service pacemaker-cloud start
We have a test suite that can be run which provides automated validation the software functions properly.
Manually the following operations can be done:
root# pcloudsh pcloudsh# jeos_create F14 x86_64 pcloudsh# assembly_create assy1 F14 x86_64 pcloudsh# assembly_clone assy1 assy2 pcloudsh# assembly_clone assy1 assy3 pcloudsh# assembly_resource_add httpd httpd assy1 pcloudsh# assembly_resource_add httpd httpd assy2 pcloudsh# assembly_resource_add httpd httpd assy3 pcloudsh# deployable_create dep1 pcloudsh# deployable_assembly_add dep1 assy1 pcloudsh# deployable_assembly_add dep1 assy2 pcloudsh# deployable_assembly_add dep1 assy3 pcloudsh# deployable_start dep1
Keep pcloudsh running and in another shell:
- verify application restart works properly:
- login to one of the assemblies and killall -9 httpd
- verify that httpd is restarted via pacemaker-cloud
- verify deployable restart works properly:
- Open the virtual machine manager GUI
- Use the force off functionality on an assembly
- The virtual machine manager should display that the assembly is restarted
- Login to the restarted virtual machine and verify httpd was restarted properly
- verify pcloudsh displays feedback
- verify failed applications indicate they are failed and restarted
- verify failed assemblies indicate they are failed and restarted
The audience will notice a shell with comands which can be used to create, launch, and monitor deployables single node.
Previously packaged in Fedora rawhide:
glib2 dbus-glib libxml2 libuuid libqb pacemaker-libs qmf libxslt qpid-cpp-server qpid-cpp-client python-qmf matahari-service matahari-host libqb oz systemd
Dependency with broken functionality: systemd - systemd guests don't work properly because systemd is not LSB compliant. A patch to resolve this issue has been merged upstream and tested working with the current pacemaker-cloud code in a f15 JEOS + upstream patch on top of latest rawhide systemd rpm.
If this feature is not ready by July 26, it can moved to a later Fedora version. If systemd is not LSB compliant by July 26, appropriate release notes should indicate that systemd is in the progress of updating its build with upstream packages. More then likely this will just be fixed as part of f16 release of systemd.
Pacemaker-Cloud provides high availability for application services inside virtual machines on a single node. This feature provides a shell for creating virtual machine images, associating resources with the virtual machines, and combining these images into a deployable. A deployable can then be launched and monitored for high availability. If virtual machines or applications fail, these components will be restarted reducing MTTR (mean time to repair) improving availability over manual operator restart.
Fedora guest virtual machines using systemd are currently non-functional until the following bugzilla is merged into rawhide: See systemd defect 702621 discussion.