On Fri, Jul 23, 1999 at 09:34:28PM -0700, Len Burns wrote:
> I kill the copy, and then see a string of reintegration messages, which
> complete successfully.  Once that has occured, I can copy more files
> to the coda volume, so it definitely has plenty of space.  cfs lv
> /coda shows plenty of space as well.

The connectivity with the server was considered weak (server responses
became sluggish), and Coda started logging all mutating operations, and
postponed writing data back to the server. Normally, log-optimizations
would then get rid of useless stores (temporary files etc) reducing
network traffic and server-load. However, the copying of a large tree
from the local fs to the cache is faster than the reintegration of the
cache with the servers, and you ran out of cachespace on the client.

You can make venus ignore it's network measurements by issueing the
command `cfs strong' before you start the copy. That way you will only
switch to (write-)disconnected mode when the connection to the server
actually dies, which should be infrequent. The command `cfs adaptive'
will let venus use the network measurements again.

We have the beginning of an implementation of writeback caching which we
could (mis)use at some time to stall user applications and reintegrate
part of the operations log when the client is write-disconnected and the
local cache is becoming too full.

> -Len

Jan