Bugzilla – Bug 7065
better performing blankspace: physical partition leases
Last modified: 2010-07-29 14:55:27
You need to
before you can comment on or make changes to this bug.
In order to support very large temporary space partitions for VMs, and fast
access to them, allow the administrator to configure a list of physical
partitions to lease out to incoming Nimbus VMs on each node.
This is an optional and preliminary feature.
The partition will be presented to a configured /dev device inside every Nimbus
VM that is started on that node. If there is a conflict with the mountpoints
the user's request contains, the administrator has a choice of rejecting the
request or not providing the blankspace.
Each partition will be formatted at creation time (destroy needs to be very
fast (especially in the pilot case)). That can be pretty dangerous if the
wrong things are formatted so all of the crucial parts of this will go via sudo
and some configuration will need to take place directly in the libexec script
instead of in the etc directory.
First version will only work on Xen deployments.
There will not be any zeroing out, that could take a very long time, but it
would be the right thing to do. In the future this could support a pool of
partitions larger than the max amount of simultaneous VMs... some would be
getting zeroed out in the background?
If you really wanted zeroing out despite the time it adds to a launch, you
could replace the current FORMAT_COMMAND with your own script that did more
than just mkfs.
Committed implementation. Needs testing with libvirt on the cluster.
Tested/fixed by DL