[Linux-HA] RE: One-way failover?

[Marian Marinov]

On Sunday 07 January 2007 21:27, Alexey Zilber wrote:
> Hi Marian,
>
>    That sounds encouraging.  We're currently using innodb exclusively (we
> absolutely must have transactional support, which is also why mysql DBD
> clusering is out of the question).  This looks like it might be a viable
> solutions then.  The other option I have is to stay with the master->slave
> config, and create a custom script that will do a point-in-time recovery
> back to the master from the slave's binlogs.  That might get a bit messy
> and the downtime might be more then 6-8 minutes (there's also an amanda
> style mysql point-in-time tool that's free).  One advantage though is that
> you don't have to do the innodb recovery that you would have to do with
> DRBD. Decisions, decisions.  Please keep me posted, and if possible, share
> the config :-)
>
> -Alex

Hello Alex,

If you have a second/third drive on the master node of DRBD cluster you can 
simply use LVM over the DRBD and use LVM's snapshots option to create point 
in time replicas. This can save you a lot of time during DR.

In fact my problem was similar to yours. I have a MySQL DB which can not be 
used with the ndbcluster engine of MySQL so I came up with this solution 
(Active/Passive DRBD cluster). While I'm using both machines I use only one 
of them for the MySQL and on the second I have configured port forwarding to 
the first on port 3306.

Regards
  Marian

>
> On 1/6/07, Marian Marinov <mm at yuhu.biz> wrote:
> > On Sunday 07 January 2007 00:40, Alexey Zilber wrote:
> > > Hi Eddie,
> > >
> > >   I'm aware of 5.0 multi-master.  We actually had 4.1 setup as
> >
> > multi-master
> >
> > > (but using it as a master-slave).   The danger with this, is that at
> >
> > some
> >
> > > point someone executed some sql to drop the databases, then the vpn
> >
> > between
> >
> > > the servers was down.  Once that was fixed, you can imagine the
> >
> > carnage...
> >
> > > What about DRDB with mysql, is that doable?
> >
> > Hello Alexey,
> > I have installed mysql over a DRBD 2 weeks ago and I'm currently testing
> > it
> > for entering in production.
> >
> > For now the max downtime that I had during simulated server downs is
> > 6-8min
> > during the checking of a big innodb DB. I had no issues with databases
> > using
> > MyISAM.
> >
> > Best regards,
> >   Marian
> >
> > > I'll check out the sourceforge project on monday too.. thanks!
> > >
> > > -Alex
> > >
> > > On 1/6/07, Eddie C <edlinuxguru at gmail.com> wrote:
> > > > There is a project on sourceforge called mysql-ha. Its not part of
> >
> > linux
> >
> > > > HA
> > > > put it deals with switching servers from master to slave in the event
> >
> > of
> >
> > > > a failure. Seems like never mysql 5.0 with multi-master is a better
> >
> > way
> >
> > > > to go.
> > > >
> > > > There is a great article on onlamp about it. I posted it to this list
> >
> > a
> >
> > > > while back.
> > > > Edward
> > > >
> > > > On 1/6/07, Luettgen, John (N-ENSCO) <john.luettgen at lmco.com> wrote:
> > > > > Alex, not configurable, but still possible.  Simply having the node
> >
> > "a"
> >
> > > > > not start HA, if the other node has the services running.  Not
> > > >
> > > > complicated
> > > >
> > > > > to do, just a bit of logic to consult the other node to check it
> >
> > state
> >
> > > > and
> > > >
> > > > > service.
> > > > >
> > > > > OK, John
> > > > >
> > > > >
> > > > > From: "Alexey Zilber" <alexeyzilber at gmail.com>
> > > > > Subject: [Linux-HA] One-way failover?
> > > > > To: linux-ha at lists.linux-ha.org
> > > > > Message-ID:
> > > > >        
> > > > > <c42337150701051318q601001b7qcd1593517deafafc at mail.gmail.com
> > > > >
> > > > > Content-Type: text/plain; charset=ISO-8859-1; format=flowed
> > > > >
> > > > > Hi All,
> > > > >
> > > > >   Is it possible to do a one-way failover with Heartbeat?  Here's
> >
> > the
> >
> > > > > scenario.  We are running mysql on node a and the slave on node
> >
> > b.  We
> >
> > > > > want
> > > > > to failover one way, ie. node a fails, node b takes over.  If node
> > > > > b fails,
> > > > > regardless if node a is up or not, we don't want any services
> > > > > coming
> > > >
> > > > back
> > > >
> > > > > up.  This will allow us to move data back to the master (node a)
> > > > > and
> > > >
> > > > bring
> > > >
> > > > > it up manually.
> > > > >
> > > > > Is this scenario possible?
> > > > >
> > > > > Thanks!
> > > > > Alex
> > > > >
> > > > > _______________________________________________
> > > > > Linux-HA mailing list
> > > > > Linux-HA at lists.linux-ha.org
> > > > > http://lists.linux-ha.org/mailman/listinfo/linux-ha
> > > > > See also: http://linux-ha.org/ReportingProblems
> > > >
> > > > _______________________________________________
> > > > Linux-HA mailing list
> > > > Linux-HA at lists.linux-ha.org
> > > > http://lists.linux-ha.org/mailman/listinfo/linux-ha
> > > > See also: http://linux-ha.org/ReportingProblems
> > >
> > > _______________________________________________
> > > Linux-HA mailing list
> > > Linux-HA at lists.linux-ha.org
> > > http://lists.linux-ha.org/mailman/listinfo/linux-ha
> > > See also: http://linux-ha.org/ReportingProblems
> >
> > _______________________________________________
> > Linux-HA mailing list
> > Linux-HA at lists.linux-ha.org
> > http://lists.linux-ha.org/mailman/listinfo/linux-ha
> > See also: http://linux-ha.org/ReportingProblems
>
> _______________________________________________
> Linux-HA mailing list
> Linux-HA at lists.linux-ha.org
> http://lists.linux-ha.org/mailman/listinfo/linux-ha
> See also: http://linux-ha.org/ReportingProblems