[SOLVED] Zimbra on DRBD

[The Fold]

I didn't get a chance to actually. I couldn't get the whole thing working so I was asked to drop it and work on something else

[hizar2000]

got this working quite nicely... up and running for about 2 weeks now.. HowTo on my blog..

ZIMBRA High Availablity Server with DRBD and HEARTBEAT Safestream

[mek1]

Great write up!

[jsmith]

Very nice write up!

I would love to hear how it has been working in a month or two.
Also how rolling upgrades (upgrade the secondary, fail-over, upgrade the "new" secondary) go??

Hopefully I will get a chance to try the same things at some point this year time permitting!!

[hizar2000]

sorry blog moved.. here is the new link for the howto..

safestream - All Articles

[bmw]

Yes, we have two dual quad-core Intels with heartbeat running and have tested it thoroughly for failsafe. Works great!

You do have to remember that the 'inactive' server does not activate Zimbra or the mail/file/data stores until it is 'active'. Also, when you are working on a server be sure never to log into the inactive server and try to do work like transfers or importing on the active server. It will crash it with a "split brain" error.

Don't have time to go into all the intricacies. Will get back to this thread next week.

[bjared]

Under /opt, we have 360GB. 169GB of this is /opt/zimbra/backups. Now, we want to replicate mail over the WAN ... obviously, a DRBD device that gets 85GB written to it each day needs some significant bandwidth to keep up. (169GB is for two days' worth of full Zimbra NE backups)

So, I'm thinking of making /opt a DRBD device, and making /opt/zimbra/backup it's own standard device. The amount of new mail received during the day should be small enough to even use a T1 as bandwidth for replication.

Of course, I won't be using HA. I'll simply have a clone of the machine with zimbra
turned off, and probably the disk unmounted as well, so nobody messes with it.

In case of failure on the primary, we simply shut the primary down (if it's not down
already), mount /opt on the failover box and start zimbra.

Some quick DNS changes and flushing of DNS cache...and users will at least be able
to send e-mail. With a little more work, we can have mail coming into it as well, as long as our anti-spam/AV service provider is quick to respond...

Does this sound like a feasible approach? I don't think there are many holes in that plan...but I could be missing something.

--Brian

[hizar2000]

Originally Posted by hizar2000

sorry blog moved.. here is the new link for the howto..

safestream - All Articles

Just a quick update.. running stable now for almost 3 months without a hitch.. carried out some maintainance on the servers a few weeks ago by manually failing over the service and everything was very graceful with no interuptions reported.

Updated link to the Howto here: data-server.org

[lassaad]

Hi all,

I just installed a cluster with ZCS / heartbeat / DRBD,

my problem is the failover:

1 - if I stop heartbeat on the primary node, takeover ok

2 - if I do "/etc/init.d/zimbra stop" : no effect: there is no takeover

any ideas ?

thank's in advance.

[sems]

Originally Posted by lassaad

Hi all,

I just installed a cluster with ZCS / heartbeat / DRBD,

my problem is the failover:

1 - if I stop heartbeat on the primary node, takeover ok

2 - if I do "/etc/init.d/zimbra stop" : no effect: there is no takeover

any ideas ?

thank's in advance.

Heartbeat does in R1 mode does not check the status of each resource, whether they are up or down. So, you'll need to implement something, like MON package to monitor zimbra status and if needed stop heartbeat accordingly.

Hizar,

In your setup, I haven't seen anything related to split-brain on DRBD. That has been my major issue (I have been running an active/passive cluster since March, but it always comes back as a problem). How do you deal with it on yours? I have been searching on the net for possible ways to prevent split-brain to happen in the first place. I found some articles suggesting to increase heartbeat's drbddisk script's retry from 6 to 20 and increasing timeout values (see Otherworldly Gate: STONITH with DRBD and Heartbeat), I haven't tried it yet but I am open to suggestions.