Re: Metadata queries on Ceph/RADOS

[Josh Durgin <josh.durgin@dreamhost.com>]

On 11/22/2011 12:11 AM, Atish Kathpal wrote:
> Hello
>
> Does RADOS support metadata queries? I have gone through librados and
> the "rados" command line tool, and there doesn't seem to be a way for
> me to say, query the object store saying: "Give me all objects with
> Xattr Video = 1". The only way I can think of is to write some of my
> own code using librados.

That's right, RADOS operates on a single-object transaction level. 
There's no way for it to support consistent reads of multiple objects 
without rewriting much of the OSD - it's not designed for that. For your 
use case, you could store your metadata in a trivial map (tmap) object 
using librados' tmap_* functions. This is what radosgw uses for storing 
which objects are in which buckets. These are slow for large mappings, 
but we have plans to fix that  (http://tracker.newdream.net/issues/1571).

> Any pointers if the system has a metadata service or something that I
> can query?
>
> (I have seen few diagrams showing a metadata cluster as part of Ceph
> which supports metadata operations, how can I leverage that? reference
> diagram: http://www.usenix.org/event/osdi06/tech/full_papers/weil/weil_html/fig/overview.png)

The metadata cluster is for the Ceph distruted file system, which is 
another layer on top of RADOS.