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

[MORITA Kazutaka]

At Sat, 11 May 2013 18:29:06 +0800,
Kai Zhang wrote:
> 
> Currently, farm is used to make cluster-wide distributed snapshot.
> However, it is hard to reload a distributed snapshot to a new cluster.
> 
> A new idea is to make cluster-wide snapshot by a single node:
> 1. read vdi info
> 2. read all objects belongs to snapshot vdi (replicas of same object will be read only once)
> 3. save these objects to local file path in farm-style sha1 file
> 
> In addition, this process can be extended in 2 ways:
> 1. performance of snapshotting can be improved by parallizing of reading objects
> 2. objects can be wrote to other device for backup

IIUC, the single node stores all the snapshot data to its local filesystem.  If
so, I think the node can be a bottleneck and doubt that it improves the snapshot
performance.

> 
> This set of patches is to remove farm from sheepdog.
> 
> After this patch, cluster snapshot does not work temporarily.
> However, it will be available soon with more useful features.

Removing farm looks overkill to me.  There are some features (e.g. data
de-duplication) which are planned to be implemented in farm.

Anyway, I'd like to see your snapshot implementation before disabling the
current farm snapshot.

Thanks,

Kazutaka