[sheepdog] [PATCH] tests: add a test to test vdi snapshot functionality concurrently

[MORITA Kazutaka]

At Thu, 20 Sep 2012 19:47:43 +0800,
Liu Yuan wrote:
> 
> On 09/20/2012 07:40 AM, MORITA Kazutaka wrote:
> > But I still wonder if we should rollback to the old cache semantics
> > since it looks a bit confusing to me.  The below patch introduces a
> > new flag SD_FLAG_CMD_CACHE to skip object cache.  With this patch, all
> > collie commands other than 'vdi read' will try to read objects
> > directly from the backing store.  Doesn't it work as you expect?
> 
> I am afraid not. sd_write_object() should also support direct write as sd_read_object()
> because sd_write_object() will also pull in a copy at local node. And also other IO requests (I am
> not sure which but I think there is) from collie will need direct mode. That is, only 
> 'collie vdi write/read' can make request on object cache.

Then, adding a direct option to sd_write_object() arguments and using
SD_FLAG_CMD_DIRECT for non vdi_write operaitons looks enough?  If yes,
I'll write the patch for 0.5.0.

Thanks,

Kazutaka