Larry Rosenman <ler@lerctr.org> writes:
> Would it make sense for ONLY drop database to have the above loop?
Not really. We'd just be papering over the most-easily-observable
consequence of readdir's malfeasance. There'd still be problems
like basebackups omitting files, missed fsyncs potentially leading
to data loss, etc.
I am sort of wondering though why we've not heard reports of this
many times before. Did FreeBSD recently change something in this
area? Also, if as they argue it's a fundamental problem in the NFS
protocol, why are we failing to reproduce it with other clients?
regards, tom lane