[sheepdog] [PATCH v1 0/4] remove farm from sheepdog

Liu Yuan namei.unix at gmail.com
Mon May 13 10:01:56 CEST 2013


On 05/13/2013 03:43 PM, Joseph Glanville wrote:
> Is the problem how the snapshot is exported out of the cluster or how
> the objects are marked read-only etc?

The problem is where we should store the so called snapshot objects.
Currently it is distributed on all the nodes of the cluster. When the
cluster snapshot is taken, all the objects will go to 'Farm', which has
auto-deduplication and compression, incremental store capability. But
storage place of Farm is fragile because it can't deal with changes of
node topology and can't be easily extended to export to other backup
storage.

> In terms of exporting it it would be really nice if a manifest of
> read-only objects was created along with an interface make reading
> said objects easy.
> The manifest can then dealt with by another tool that understands what
> you are trying to export to.
> 

So the new idea is to put Farm in a central storage instead of
distributed storage of all nodes. As you said, the underlying storage
for farm can be NFS, S3 or Swift and other distributed storage. The
first version will, thought, be limited to local file system but this is
very easy to be extended later.

Thanks,
Yuan



More information about the sheepdog mailing list