I'm not entirely sure about the answers here but I'll see if I can add
something useful to the discussion...
For starters, qmail-1.03/INTERNALS says the following:
> The queue is designed to be crashproof, provided that the underlying
> filesystem is crashproof.
Whether NFS is considered crashproof or not, I don't know. I don't think it
really is.
The questions posed in the archive were:
> 1. Are inode numbers consistent across NFS?
> 2. Are named pipes possible across NFS?
> 3. Are link() and rename() defined to be atomic across NFS?
1. They appear to be, at least on Linux. I'm not sure if this is required
by the NFS specification, or if the NFS inode is related to the "real" inode
on the disk.
2. Yes (at least on Linux). However, they aren't a part of NFS; they're
handled by the kernel. Named pipes are merely an extension of the anonymous
pipe facility using the filesystem as the namespace. You can put the pipes
wherever you want and once they are opened, I would think you could unmount
the disk and pipe operations would still succeed. I'm not sure about that
though.
3. They may be so defined; however, a better question is "on what
implementations can this be guaranteed?" Also, I'm sure the semantics are
different depending on which version of NFS you're running.
My conclusion - Mounting the queue dir over NFS may very well be possible
assuming you can count on certain things. However, there's no way I would
EVER do it, and you're probably asking for trouble if you do.
I'm sure you've thought about this issue and have good reasons for wanting
to do this, but I'm curious to know why you need to put queues on each of
these machines with "limited space." If these machines were diskless (or
near-diskless) workstations, you'd be better off setting each machine up to
relay to a central hub.
You say this is for a server setup though, so I'm assuming the machines
aren't diskless. Given that these are servers, they must have at least some
reasonable amount of local disk in them, so you've almost certainly got
enough room to put a queue on them.
shag
=====
Judd Bourgeois | CNM Network +1 (805) 520-7170
Software Architect | 1900 Los Angeles Avenue, 2nd Floor
[EMAIL PROTECTED] | Simi Valley, CA 93065
Quidquid latine dictum sit, altum viditur.
----- Original Message -----
From: Curtis Generous <[EMAIL PROTECTED]>
To: <[EMAIL PROTECTED]>
Sent: Mon 1 Nov 1999 14.04
Subject: NFS mounted /var/qmail/queue directory
> Is it possible (and safe) to use an NFS mounted /var/qmail/queue/*
directory?
>
> [NOTE: I'm _not_ advocating sharing /var/qmail/queue/* tree between
> several qmail servers via NFS, but instead want to use an NFS filesystem
> for the /var/qmail/queue/* tree instead of using local disks which have
> limited space on them in our configuration.]
>
> We have a NetApp 760 filer with lots of disks, and it makes better
> sense to use those disks rather than buying external disk packs to
> attach to our QMAIL servers (on which we already use NFS mounted
> MAILDIR for user mailboxes).
>
> There was a thread on this topic a couple of years ago but no resolution
> to this issue. Here is that URL:
>
>
http://www.ornl.gov/its/archives/mailing-lists/qmail/1997/04/msg00619.html
>
> Would Appreciate any feedback/comments on this issue.
>
> Thanks,
>
> --curtis
>
