From: José Luis Tallón (jltallonadv-solutions.net)
Date: Mon Feb 09 2004 - 08:45:17 CST

• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

At 11:49 09/02/2004, Christian Hammers wrote:
>Hello
>
>Today I got the following error on a server cluster of two machines that
>are running Debian woody with DRBD partitions and NFS on top of them
>(one machine acting as nfs server and both having mounted their spool in
>different directories in this nfs share...the machines are linked with a
>crosscabel between intel e1000 gigabit nics)

Postfix queues are not to be used on top of networked filesystems unless
they are able to behave *exactly* like a local filesystem. This excludes
explicitly OCFS, AFS, Coda, and the like and NFS most of the time ( there
has been a thread on this lately )

It is ok to *deliver* mail over NFS( when using Maildirs, that is ),
though. ( AFS and the like will possibly work for delivery, too )

>Can anybody tell me what could be the cause of this error?

NFS is not completely like a local filesystem ( does not deliver all the
promises a POSIX filesystem should and which Postfix needs to work reliably
), at least not on Linux.

>Is it known to happen in certain circumstances?

Yes.

You'd better keep the queue on a local filesystem ( i won't enter the
quasi-religious war between Ext3/ReiserFS/XFS/whatever ); If more speed
that the current disk subsystem can deliver is needed, you can always use
SSD or RAID 10 ( RAID 1 on top of RAID 0 ) for the queue. Anything else is
likely to lose mail sometime.

>BTW: If somebody knows a NFS mailing list it would be nice to know..

Dunno, sorry.

>bye,
>
>-christian-

Take care, Christian.

         J.L.

>postfix/local[14486]: warning: active/2/2C0768CE83: unexpected EOF in
>data, record type 78 length 76

some problem reading a queuefile -- don't know the exact reason. Solution:
don't put the queue on an NFS-mounted FS.

• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]