On Friday 17 April 2009 14:01:07 Wietse Venema wrote: > Wietse Venema: > > Melvyn Sopacua: > > > On Friday 17 April 2009 01:23:20 Wietse Venema wrote:
<snip replication> > > > Also, any technical objections against moving shared files into an nfs > > > mounted directory and adjusting main.cf to look there? > > > > NFS is not suitable for write-sharing. It is OK only for sharing > > read-only files, or when there is a single writer who is also the > > sole reader. > > In case this gets mis-interpreted: I was talking about NFS sharing > files that are overwritten, or that are updated in place. > > Maildir does not have the above problems. It does not overwrite > files, and it does not update files in place - rather it writes a > file first in a different place and then atomically hardlinks the > file into its final place. > > Mailbox files, on the other hand, are usually overwritten, and they > are updated in place. This almost works reliably, especially if > you use dotlock files and turn off NFS attribute caching. > And I was talking about OP's case of sharing configuration files. I do a lot more with nfs, where multiple machines can write to the same file. As long as I do it in sequence, the other ones will see the changes. Two people editing the same file falls into shooting yourself in the foot category and is not specific to NFS or even two different machines. My concern was more with the machine where postmap is /not/ run, whether it will pick up the changes in a timely fashion. From my experience with FreeBSD nfs, I would say so, but maybe there are implementations that lie/cache stat information. I've inherited a similar setup with two incoming mailhubs in a round-robin, where sharing mostly static config files would save me some work, hence my interest. -- Melvyn Sopacua
