Hobbit cluster and DRDB 🔗 link

Hi Brothers,

Just read an article in Linux Journal about using DRDB as a "cluster" filesystem for a redundant installation of sendmail and mysql. This has caused me to think that this could be a way forward as a cluster mecanism for hobbit. Currently I run a rdist job for this and then have a manual intervention for starting up hobbit on the secondary node.

I know from when I last looked at DRDB that the IO performance as a problem with BB but with Hobbit I am prety sure it should work.

I am wondering if any of you have experienced with the DRDB/heartbeat/hobbit and what your impressions are ?

Regards, Thomas


--

Hi... I use drbd and heartbeat for some clusters at work. We run apache,
mysql, qmail, pound (a free load balancer), an Oracle Instance (where we
cannot use our RAC) and also for the admin server of a weblogic server. I'm using drbd 0.7.xx and heartbeat 1.2.x on RHEL3 and RHEL4, 32 and 64
bit. Some of those cluster are running from more then one year without any
problems.

Heartbeat it simple to install and need no mainteinance at all, you can
upgrade it without problems (the 2.0 also support multinode cluster).
DRBD can also be configured to get some performance boost at risk for
data writing on disk (he have 3 mode of disk writing). If you have a free network card you can dedicate it to heartbeat and
drbd with a crossover cable and at least with Linux the networking work
quite well so you can get good performance on a 1gb network card. The only "problem" with drbd is that if you need to upgrade the kernel
you'all also need to recompile the drbd drivers for that kernel or
you'll not see the disks. Heartbeat and drbd will integrate themself
simply without problems. You'll end with a shared disk (that can be only
primary in write access from one node) where you'll install the entire
hobbit home directory, a shared ip address that you will use as ip
address for hobbit.
You will just need one line of configuration in the heartbeat resource
file to put the ip, the device to mount and the script to start hobbit.

Regarding performance I think it will not be so bad but I think it will
depends from how many rrd youre using for graphs. I think that with a gb
network card and a good buffer for io the performance will not be a
problem.

Francesco

Date: 16-08-2006

Yea, I read the article as well. I was attempting  implement BB +DRBD
on RH Linux but the project stopped due to other priority.

During test phase I was able to simluate the outage of a webser server
on A to cause slave server (B) to shoot down A and take over its ip address
via private subnet in between etc ...

I do remember the sync between two nodes was great but I didn't
really deploy bb on bother server so I don't now how it really run.

I am wondering if any of you have experienced with the DRDB/heartbeat/hobbit and what your impressions are ?

I am less interested about BB+DRBD+RHLinux combination
for implementing system monitoring clustering. Instead, I am more
interested about Hobbit+Solaris 10+Sun Cluster software.
Sorry to hijack your subject into mine.

Sun Cluster software is now free for use( see R1)

R1: http://www.sun.com/software/cluster/

tj

Regards, Thomas


-- 

I use hearthbeat (for all our linux cluster) and suncluster (for 2
solaris cluster). I don't have a big knowledge of SunCluster but it
seems to me really more complicated. From what I remember you also need
a fc disk or some scsi disk connected to the 2 machine to work as quorum
device for the cluster and it's a bit intrusive inside the OS kernel
with his drivers (I don't know if with solaris 10 and new version of
suncluster something has changed). 

I know it's more sophisticated but also really more complicated and
difficult to setup related to heartbeat and it's free to download and
use but I don't know how much support you'll get for the free version if
you have a problem.

Hearthbeat setup will take about 5 minutes and all you have to do is to
edit 2 file ... 
And it also run on Solaris (I didn't test it but at least it's what's
written on the website :D)

Francesco

I have setup Hobbit before using a shared directory both via a NetApp NFS mount and an NFS mount from a linux server,  and it worked just fine, so it should also work well with a DRBD filesystem. From what I have seen Hobbit isn't too IO intensive
Heartbeat should be able to easily fire up the hobbit server on another node in the case of failover, but I have never tried a setup like that.  I would be interested to see if the heartbeat takeover is quick enough to not cause any false conn alerts to be sent when Hobbit starts up on the other node (there is a small window where network connectivity isn't there at failover because of arp caches and whatnot).

-Charles

Francesco Duranti wrote:

On Fri, Aug 18, 2006 at 10:59:21AM +0200, Thomas wrote:

I am thinking about adding Mon for local hobbit monitoring, but for that to work I would need some "ping - pong" responce from opening a sesion to the deamon. Anybody know what I could send and what I should expect back ?

   bb 127.0.0.1 "ping" 
returns the text "hobbitd VERSION"


Henrik

Hi again,

Thanks for your answers, I will give it a go and post my results.

I am thinking about adding Mon for local hobbit monitoring, but for that to work I would need some "ping - pong" responce from opening a sesion to the deamon. Anybody know what I could send and what I should expect back ?

BR Thomas

Thomas wrote: