Open Stack

I know there are some people upset with me questioning this feature so
late in the development cycle, but I think people need to get over
that. If there are concerns being raised in the community, and a
feature is not released, it's completely valid for me to ask questions
and be unsure about this feature being released after all.

Leaving that argument aside, I want to understand how useful this is,
because it introduces a bit of complexity into Cinder, that I'm not
sure is solving much. Ultimately OpenStack operators are going to use
this feature, as it requires them to create these filters and weigher
functions. So far, it appears only one operator has weighed in on this
feature according to gerrit.

Proposed Documentation:
http://thing.ee/x/doc-20150210/content/driver_filter_weighing.html

Spec: https://review.openstack.org/#/c/129330/11/specs/kilo/filtering-weighing-with-driver-supplied-functions.rst

Now let me reiterate. I say it appears this feature is not solving
much, because the documentation is so vague, and it just recommends me
to consult a vendor for help. In my opinion, this is a red flag on a
feature in OpenStack. If the feature is so complex that we can't give
help in our own documentation, then I'm not sure if we're solving the
problem the right way. It also introduces error prone issues into
Cinder, assuming the OpenStack operator is going to invest time in
writing for each storage backend deployed a filter and goodness
function that will work. A common question people ask is why not solve
this problem in the current filter scheduler Cinder has today. I'm
sure there were debates about how that would work, but might be worth
reconsidering another look if we're getting to this point.

I would like to hear from the operators, who are ultimately going to
be faced with this feature of how useful it would be for them.

--
Mike Perez