push data instead of pull

[Martin Pool]

On 20 Feb 2002, tim.conway at philips.com wrote:
> > 
> > "Connection reset by peer" often means the remote rsync crashed. 
> > 
> > If you see this message when the only problem is that the remote
> > module is read only, that's a bug.
> That was my point.  ro and rw were both /tmp.  i made ro first, tested it, 
> duplicated Gautam's error, copied it, changed "yes" to "no" on the "read 
> only = " line, and got success.  The really odd thing was tht i got 
> meaningful error on ro if i was tracing the rsyncd (started as "truss -f 
> rsync --daemon", "-f" meaning follow forks).  I'd suppose that slowed the 
> daemon slightly in some way, alleviating a race condition. 

I suspect that on encountering an error we do not correctly drain all
error messages that are in flight on the network.  I'm not sure.

-- 
Martin