> > > Hi,
> > > 
> > > I would like to build a HA Solution with 2 servers with 1 in
> > > hot-standy i.e. fail-over. The advice so far was to use GlusterFS
> > > (replicated) for the filesystem and mysql replication (master-slave)
> > > for the database. The purpose is a web server (apache) with a typo3
> > > CMS.
> > > 
> > > In the event of a failure I need to run a script to perform the actual
> > > failover (switch ip via web request (Hetzner robot) and change mysql
> > > replication settings).
> > > 
> > > Would you recommend to use corosync/pacemaker to monitor the
> > 
> > database,
> > 
> > > gluster and apache or should I simply check the availability of a DB
> > > generated Web Page from a script in a loop?
> > > 
> > > Any other recommendations?
> > 
> > Yes!
> > Set up real cluster, use DRBD to replicate the data between the nodes and
> > pacamaker to monitor the resources (IP, filesystem, database and
> > webserver).
> > pacemaker will do the failover in case of problems.
> > 
> > See: http://www.linbit.at/training/webseminare-auf-abruf/mysql-
> > replikation-
> > mit-pacemaker/
> > 
> > > Since I don't have a lot of corosync experience it seems to me like a
> > > huge task (overkill) for a simple monitoring of a failover scenario.
> > > 
> > > What do you think?
> > 
> > I heard there is a very good book from O'Reilly ;-)
> > 
> > Greetings,
> > 
> > --
> > Dr. Michael Schwartzkopff
> 
> Hi Michael,
> 
> the problems starts when I start to think about fencing: I am working with
> a standard root server, it simply doesn't have a STONITH hardware. I've
> read various threads which basically say if you don't have proper fencing
> when it's all your fault ... You don't really suggest to use the
> ssh-stonith for production ... 

STONITH idea for Hetzner:
http://lists.linux-ha.org/pipermail/linux-ha/2011-May/043187.html

> I could externally trigger a hardware reset
> of a machine and reroute the ip to the other server. Would that be a valid
> stonith action? I did actually like the book, but fencing takes some fun
> out of the whole setup (makes it a lot more complicated). Also I tend to
> believe it's best to use the database replication methods if available
> rather than do it via the file system.

DRBD is block replication, not file system sync.

> I didn't like drbd too much, since there is no easy way to gain access to
> the filesystem on the slave machine. I never got the whole stack in dual
> primary mode with ocfs2 running. Gluster easily provides for this.
> 
> Stefan
> 
> P.s.  :-) I did actually like the book, but the fencing part is a killer

I would be happy on a 5-star comment on amazon ;-)

Do as you like. But setting up a Linux Cluster including fencing and DRBD is 
not a big deal. And if your compare it with writing your own failover scripts 
...

-- 
Dr. Michael Schwartzkopff
Guardinistr. 63
81375 München

Tel: (0163) 172 50 98

Attachment: signature.asc
Description: This is a digitally signed message part.

_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems

Reply via email to