[sheepdog] Production VS Testing

[Liu Yuan]

On 12/11/2012 03:45 PM, Valerio Pachera wrote:
> 'This software is experimental and developing software'
> 

Maybe we can call it beta to make it not scare away the possible users.

> Hi guys, in this period I'm testing sheepdog and I tested it a bit in the 2011.
> I have to say I wasn't able to corrupt my guest image after all my
> happy testing :-)
> Ok, it was only a single guest doing almost nothing and only 3 nodes,
> but this seems a very good resault to me.
> 

We have running more than 20 nodes for about 6 months for production
use. Yeah, we have experienced bugs and even hang and upgrading things.
I think we can say it is rather stable right now. For stable, I mean the
sheepdog can tolerate high volume of concurrent IO operations from
Guests throughout all the nodes and serving IO request as expect all the
time even during nodes changes. And no known defects that need a long
time fix, and potential bugs can fixed in a rather small time window
because core code is stable and compete.

> What I'm wondering about this project:
> - is there a road map and a precise set of test that will move it out
> of 'testing' phase?

We have tested fio, iozones,even runs hadoop's terasort and more
importantly, running sheepdog cluster month for production use. It shows
Sheepdog is reliable.

For current development roadmap, it is the features rather than core
code fixes comes at the first place. For e.g, a better iSCSI support and
other protocol support, better integrated into existing cloud stack such
as openstack, etc.

> - i wish to help testing to have a production ready software, may you
> tell me wich kind of tests I can do?

I think, simply go ahead with whatever you want from Sheepdog and report
bugs or request features to the community.

> - is there a known situation where sheep dog fail it's job?

For large scale, for e.g, more than several thousand nodes, I think
Sheepdog will much works to accomplish it.

Thanks,
Yuan