Difference between revisions of "InstanceMigration"
m (Text replace - "__NOTOC__" to "") |
|||
Line 1: | Line 1: | ||
− | + | ||
<!-- ##(see the [[SpecSpec]] for an explanation) --> | <!-- ##(see the [[SpecSpec]] for an explanation) --> | ||
* '''Launchpad Entry''': [[NovaSpec]]:migration | * '''Launchpad Entry''': [[NovaSpec]]:migration |
Revision as of 23:31, 17 February 2013
- Launchpad Entry: NovaSpec:migration
- Created: 2010-09-01
- Contributors: JesseAndrews
Contents
Summary
Develop a feature to allow cloud administrators to perform maintenance tasks of physical servers. This feature will allow admins to shutdown an instance and move it to a new host node.
Release Note
An internal admin tool to move instances from a compute node has been added to OpenStack compute.
Rationale
Hardware fails. A user's instance shouldn't be permanently lost when hardware must be removed from the pool.
User stories
A member of the operations team notices that the hardware on a server is failing. To protect the users, the instances are migrated and the system is sent to the datacenter team.
Assumptions
- more than one node exists
- compute nodes are able to rsync/scp between each other (server setup should be limited ssh between nodes to only rsync) -- or the image is moved to the object store.
Implementation
a nova-manage command will be added that:
- shutdown the instance
- determines a new server to migrate the instance (via the scheduler)
- the compute node that receives the run command retrieves the image from the existing server.
- the network configuration is kept the same, and the new VM is restarted
Optimization:
- when you take a snapshot, it creates a base and a delta (cow) file
- the base snapshot could be copied before the instance is shutdown, then the delta (cow) could be moved after shutdown - which would be much smaller and faster.
- potentially pause VM instead of shutdown