On Fri, Jan 05, 2001 at 10:51:58PM +0100, Ivo Vollrath wrote:
> Hi,
> 
> When I move huge amounts of data from a local drive to a coda 
> volume, I get inconsistencies with the coda server.  My local
> venus cache is about 100M large but the data I am moving is
> larger (e.g. an installation directory of StarOffice).
> It seems that the inconsistencies do not occur when I disable
> write-back caching (by saying "cfs writereconnect").
> 
> Here is a command that produces inconsistencies:
> 
> [/opt]# mv StarOffice5.2/* /coda/opt/i386_linux/StarOffice5.2
> (where /coda/opt/i386_linux/StarOffice5.2 is a coda volume)

This happens because the Coda client, doesn't yet know how to throttle
back the rate at which new files are created. It couldn't push the
modification to the server fast enough and probably ran out of space in
the cache.

During weakly-connected operation (when writeback caching), the client
creates it's own local file-id's. These are replaced by prober global
FID's when trickle-reintegrating to the server. The replacement of local
with global FIDs is prone to small races where the kernel already knows
about the global fid but venus doesn't, or the other way around. This
is one possible cause for inconsistencies.

Jan