[sheepdog-users] [sheepdog] qustions about sheepdog presentation

[Liu Yuan]

On Wed, Oct 22, 2014 at 11:15:49AM +0200, Valerio Pachera wrote:
> 2014-10-22 11:02 GMT+02:00 Vasiliy Tolstov <v.tolstov at selfip.ru>:
> > Hi.
> > I find this slides
> > http://sheepdog.github.io/sheepdog/_static/sheepdog-openstack.pdf
> > Page 9 contains gateway node with shared cache on ssd. What data
> > placed on this ssd ?
> 
> qemu sends a read request;
> sheep request from the other nodes the required object;
> it stores a copy of that object in the object cache (ssd);
> the next time qemu is sending a read request for the same object, it
> will read it directly from it's ssd.
> 
> Qemu sends a write request;
> sheep write a new object in the object cache (ssd) instead of writing
> that on other nodes;
> after a certain amount of time, it's going to replicate this object on
> the other nodes ad remove it from the cache.

A minor addition,

- the data will be replicated to other nodes if a) timeout b) Guest send 'SYNC'
- we write a new object and it will be marked as dirty in cache and after
  replication to other nodes, it will be marked as clean and stay in the cache
- the clean objects could be removed from cache if the cache if near full.

> It's evident that if the node dies before replicating the object in
> the cache, you loose data.

We won't lose the data if the cache is still alive(failred node can reboot)
because object cache is persistent cache.

Thanks
Yuan