[Kolab-devel] Kolab: Cyrus replication HA option

[Matt Douhan]

On Thursday 21 July 2005 22.02, Kevin Baker wrote:
> Interesting... some questions below
> >
> > 1, Snap shot the fs, dump, copy restore
>
> So the dump is a complete backup,
> The copy is scp to the other server,
> the restore is stop cyrus move files over start cyrus?

Yes

>
> Do you stop Cyrus before the snapshot? Does the snapshot
> require additional space for the mailstore duplicate, or
> is it a small sized snapshot with sym links then rsync?
>
> Does this require a cyrus reconstruct?

No it does not since the dump is complete, and yes it is a small snapshot.

>
> > 2, Snap shot the fs, mount the snapshot and rsynch it to
> > the standby.
>
> Same as above is the snapshot requiring extra storage? IE
> 40GB's mail spool, 40GB snapshot... or 40GB mailspool,
> snapshot 0+changes.

No the snap itself is minimal, it also depends how you do the snap.


>
> > A lot easier then messing with IMAP things, but the lag is
> > as large as the
> > synch time between hosts.
>
> I agree... this has been an on going issue for me with
> Cyrus. Considered moving to DBMail for a while to use
> MySQL replication, but it is not as feature rich.
>
> > but for a normal server, with some 3-400 GB storage, we
> > can dump and copy and
> > restore fairly quickly.
>
> So what is the lag? Is your hot "backup" server 5 minutes
> behind? or....

On smaller servers we are looking at a lag of roughly 10-15 minutes

On some larger machines up to 1 hour, neither of wich is bothering us or the 
customers very much.

>
>
> Thanks so much for your response... I'd love to have a
> method working in production that could be published on
> the wiki. It'd be great to have a "standard" way to
> provide a low cost "hot backup" setup for two server
> clusters. Sort of a RAID-1 for email...

We also have other things like a shared storage with one hot and standby host, 
but that only helps if the host machine goes down, not if the storage itself 
is lost, I like to have total redundancy, storage AND hosts, that is why we 
have chosen the scenario we have.

-- 
Matt Douhan
www.fruitsalad.org