Yet another mail-restore question...

[Simon Matter]

Earl Shannon schrieb:
> 
> Hello,
> 
> Sadly we've a little experience in IMAP server recovery.
> Most of what I'm listing makes common sense but I'll say
> it anyway.

Hi,

I'm currently planning a new mailsystem so I'm interested in quick
recovery in case something goes wrong.

My idea was the following:
- Using XFS or ext3 filesystem (XFS is the preferred one)
- Using LVM
- Having two servers. One real server and a 'snapshot' server.
The snapshot server is built with cheap, big IDE drives. Using rsync,
snapshots (well, sort of) are taken periodically and in the case of
failure, data can be restored quickly from this box.

Simon

> 
> How to quickly get back a once working server depends upon
> whats wrong with it. We've not had any problems with software
> failing but have had hardware bite us(me) in the butt.
> The major concern we have is restoring a server in the event
> it goes completely south. We were looking at this recently
> from the wrong side of the event. The RAID controller in
> the server was not a redundant controller and its battery
> was failing. When I attempted to replace the battery I only
> made things worse. The machine lost the file system with
> the mail store and configuration files on it. While 40 Gb
> of data may not sound like much by todays standards anymore,
> wait until you lose it and are looking at recovering it.
> ( And with a 10 Mb quota limit currently, you figure out
> how many users were somewhat upset. And yes, we are very
> overbooked for quota on our machines. Believe it or not
> the students can get by with the 10 Mb limit. Staff on the
> other hand, who have 20 Mb, are still not satisfied. )
> We took a two pronged approach to see which would come back
> faster, rebuilding the filesystem on the current machine
> using fsck and selected restorations, or simply building
> a new machine and restoring to it.  We actually had the
> fsck'ed machine back before the restoration was done, but
> decided that a reconcstruct on each individual mailbox should
> also be done. That took almost another twelve hours.
> 
> Building the new machine was pretty straight forward in this
> case since we would have been able to simply take the
> system disk from the old machine and put it in the new machine
> and just rebooted with the new RAID unit. But we also have
> an IMAP install "kit" we can use to create a new IMAP server
> when we need one. We would then have done the restoration to
> the newly installed machine. I don't know from experience if
> any of the Bare Metal recovery methods would have done the
> restoration any faster. I do believe that the system disk if
> lost can probably be recovered significantly faster via Bare
> Metal methods ( Such as what Veritas markets ) but we do not
> yet have the technology in house here.
> 
> This has made me painfully aware that large file systems,
> while nice to store lots of stuff, have severe drawbacks when
> it comes to replacing them from backup. We are wanting to
> get transitioned to a journaling file system in order to help
> reduce fsck times on the occasional/unscheduled reboot.
> 
> The bottom line is you have to know your system and its requirements
> for recovery. Making public a document about how to restore
> service in the event of a disaster would only give you
> an outline to use at your site since you would probably use
> different backup methods, have different OS you were using,
> etc. And you should really be able to come up with the outline
> yourself.
> 
> Regards,
> Earl Shannon
> --
> Systems Programmer, Computing Services, Information Technology
> NC State University.
> http://www.earl.ncsu.edu
> 
> Bryntez wrote:
> >
> > Has anybody on the list made a detailed doc about
> > howto restore the maildb in case of disaster ?
> > I mean a short, quick note with step-by-step commands
> > to execute, to quickly get back in business ?
> >
> > If for some reason the system crashes, it sure would have
> > been nice to have a doc at hand to quickly restore the
> > imap-system, saving time, and not having to browse the whole
> > documentation when the sweat are dripping and the stress-
> > level are at a dangerous level :-)
> >
> > A copy of such doc would have been reassuring....
> >
> > (Running Cyrus imap/sasl 2.1.5 on RedHat 7.3)
> >
> > Regards
> > bryntez