> No directory _attribute_ caching, which again should mean that we check
> if the parent directory mtime has changed on each lookup.
Ok, sorry for not doing a complete RTFM but is there an attribute then for
disabling directory caching?
> The restore will probably mess with the mtime, in which case you should
> probably apply the above commit (applications cannot change the ctime).
The odd thing is, that we're seeing sync delays of up to 4-5 hours when
doing a restore onto NFS, but also 1-2 hours when doing a backup from
NFS (which is then only reads) - the latter is rather odd as in making cache
syncs that slow, isn't it? Looks like an overload situation maybe?
Cheers,
Stefan