As it stands right now, there are some problems with our UFO implementation; 1)
it’s based on
OpenStack Swift-1.4.8 (Essex), which is rapidly becoming obsolete, if it isn’t
already; 2) it
doesn’t use Keystone, it uses tempauth, which doesn’t scale; 3) the
implementation is a heavily
patched version of swift — a patch that is so intrusive that will make it very
difficult to port
UFO to the current version of Swift (1.7.4, Folsom) and later; and 4) fixing
performance bugs is
hard, we think it'll be easier to fix them if we clean up the implementation
first.
Not only is the patch so big that it’s hard to port, but it has been deemed by
some to be —
essentially — a fork of Swift. Needless to say, we don’t want to maintain a
fork in perpetuity,
nor do we want to make moving to newer versions hard (because that would be yet
another fork),
and we want to be good Open Source citizens and not unnecessarily fork things
that don’t need
to be forked. And theory says if we make porting to new versions simple and
easy, then we can
minimize the lag between OpenStack’s releases and our own releases of UFO. And
if we can make
UFO be completely version agnostic when it comes to Swift; then, when new
versions of Swift are
released, UFO will automatically work with it with no effort on our part.
To that end, we’re working on rewriting UFO. Instead of patching the Swift
sources, we’re
implementing a set of layered subclasses that contain the UFO functionality
outside of Swift
per se. This implementation does rely on a patch of code that has been accepted
into the Swift
1.7.5 source. It’s referred to it as the constraints-config-backport patch.
We’re lobbying the
Fedora community to accept the constraints-config-backport patch into the
openstack-swift RPMs
that they distribute so that we do not have to provide our lightly patched
version of Swift in
the glusterfs RPMs.
These changes are already on the HEAD of the glusterfs git tree. I have a set
of RPMs in an
experimental YUM repo at
http://repos.fedorapeople.org/repos/kkeithle/glusterfs/XX-ufo/ and
instructions for setting up UFO are in my blog article at
http://www.gluster.org/2012/11/a-better-
ufo-is-coming-soon/
Ultimately we'd like to get buy-in from the community and move toward shipping
this new 1.1
implementation of UFO relatively soon. The sooner we ship this — still based on
Swift 1.4.8 —
the sooner we can roll out the next one, based on Swift 1.7.4 (or later).
So give it a try, let us know what you think. If you have questions, please
don't hesitate to ask.