[sheepdog-users] plan for Sheepdog 1.0 and new organization

[Liu Yuan]

On 05/24/2013 05:29 PM, Valerio Pachera wrote:
> Sheepdog does already many things.
> For release 1.0 I would focus on heavy testing of the most important futures.
> This means that it's ok to me if recovering slow down the cluster or
> if guests do not get reconnected after a temporary host disconnection
> etc.
> 
> Recovering method has recently changed and snapshots too.
> Snapshot reliability it's a key future that would save me/us lot's of
> trouble with backups.
> Correct me if I'm wrong but, because of the latest changes, snapshots
> may have to be consider almost like a new future, right?

It is just a new design of the old things.

It is not new, it was introduced with farm store, which supports the
cluster wide snapshot at day 1. But it was hard to put into production
use because it stores data locally in the cluster. This means if your
cluster dies, the backup data are gone too.

The new design of cluster wide snapshots just move farm out of sheep and
have it focus on the storing of snapshot objects only and teach it to
store those data in another storage system. This will completely
separate backup data from running cluster.

> 
> For example, I would not consider Multi Device a stable future right now.
> If 1.0 was going to release tomorrow, I would not include that future.
> Luckily we have time to test and fix it.
> 

MD is almost done to be used in production. What we miss now is:
 1 Fix 'node info'
 2 Add a new command to allow users to change the node weight manually.
We can't do it transparently to user because re-weighting will trigger a
cluster recovery. So plug/unplug disks as default won't cause cluster
wide recovery, but we can activate reweighting if we want.

> I think we need a table with the list of futures (see "What is cooking
> in master");
> a "testing rate / maturity level";

Looks good to me. I'll add an entry for it at monthly 'what is cooking'
report

By the way, the next release was changed to be as '0.6'.

Thanks,
Yuan