[sheepdog] [PATCH v2 08/11] group: add node to sockfd_cache as early as possible

[Hitoshi Mitake]

At Wed, 18 Sep 2013 11:52:16 +0800,
Liu Yuan wrote:
> 
> On Wed, Sep 18, 2013 at 12:34:36AM +0900, Hitoshi Mitake wrote:
> > At Sat, 14 Sep 2013 18:34:28 +0800,
> > Liu Yuan wrote:
> > > 
> > > get_vdis will try to connect to all the nodes include joining node, so before
> > > calling get_vdis, we'd better add joining node in the sockfd cache. This is not
> > > a fatal problem and sockfd cache can actually sort it out with the old code, but
> > > this patch will safe us unnecessary revalidating node after grab failure.
> > > 
> > > Signed-off-by: Liu Yuan <namei.unix at gmail.com>
> > > ---
> > >  sheep/group.c |    3 +--
> > >  1 file changed, 1 insertion(+), 2 deletions(-)
> > > 
> > > diff --git a/sheep/group.c b/sheep/group.c
> > > index 660fc5a..0b03150 100644
> > > --- a/sheep/group.c
> > > +++ b/sheep/group.c
> > > @@ -633,6 +633,7 @@ static void update_cluster_info(const struct cluster_info *cinfo,
> > >  
> > >  	if (node_is_local(joined))
> > >  		sockfd_cache_add_group(nroot);
> > 
> > We should add else here. If the joined node is local, the newly
> > created connections will not be useful.
> > 
> 
> Yes for now, but the erasure code patch series I am currently working on will
> need to add itself to the sockfd cache. So let's keep it as is in this patch.

If the erasure coding cannot be implemented without connecting to
local nodes and it cannot be an optional thing, the "else" shouldn't
be added. But it is hard to imagine such an implementation.

I think eliminating the else after adding the erasure coding isn't too
late.

Thanks,
Hitoshi