[Typo3-dev] Mirroring typo3.org etc.

[Jan-Hendrik Heuing [netfielders]]

> > > please go ahead. personally I would like to know how others solve
> > > this non-trivial issue.
> >
> > With commercial solutions !
> >
> > All which has been discussed here is no option at all, as you can't
> > keep installations in sync properly over distance. There would be a
> > solution:
> >
> > For the filesystem, search for block-devices. There are open source
> > tools which do that one way, there are commercial tools which do it
> > both ways. But this is far to expensive ! Seriously. This is even much
> > more then the mysql-solution we are going to use.
>
> We(hostsharing.net/german) will evaluate DRBD pretty soon, which seems
> to be very promissing. It will probably take a couple of month though
> till we come up with a result.
> http://www.drbd.org/start.html
> http://www.linbit.com/de/article/articleview/54/1/6

If you have any good results with this, please tell me. Could be a
realtime-backup alternativ.  I didn't find anything when I have been looking
around. At least nothing which did operate their website for more then a
half a year, of if more, kept doing so... ;) Most of these projects are dead
projects.

> > For the database backend, there is emic networks, providing a good
> > commercial solution, which we are testing at the moment. We'll go
> > online with that within the next 2 weeks if our tests work out fine.
> > MySQL has been the bottleneck with the huge database, which ended up
> > in a loop while exporting data via mysqldump ;) (Raid, Filesystem,
> > myisamchk, all went fine and did not report errors)
> >
> > -> www.emicnetworks.com
> >
> > Believe me, there is no proper way syncing two installations in
> > realtime for a proper failover. If you come up with a _working_
> > version of something the like, it could be discussed again. But I
> > don't think this will be possible with the tools available.
>
> Well, we are facing the same problem and are still looking for a gpl'ed
> solution, since payware is no real choice for our business. There
> must be a solution. Anyway if you are facing that much of a problem with
> mysql, DRBD might not be of any help I believe. Have you
> had a look at DRBD yet? Haven't you tried heartbeat already for some
> purpose? Can't find that anymore.

We use a lot of heartbeat things. It's no choice for master-replication of
mysql though. One network outage, and replication MIGHT be gone if  a client
send a single update or insert to the slave. We actualy used it, and it
didn't spit out any errors that it did not replicate anymore, which it
should do in any way in the logs. Well, that turned out to failover to a 3
weeks old database. We skipped that approach immediatly.
For mysql the emic-solution looks like a good solution. Problem at the
moment is that it is much slower than mysql straight away, which is no
option.
We did use failover for mailservers, webservers etc..., now we switched over
to a load balancer. Not because of problems, more like because it's more
comfortable.
mod_backend is no option either. After 150GB/month went through, mod_backend
just did not work properly anymore as it should. We skipped that as well,
especially because traffic went up twice as much within a couple of month.
If you work with a few GBs, no problem... Loadbalancer are much better in
many ways...


> Actually this is HA(High availability) not "mirroring" but afaik it can
> go over distance as long as there is a proper bandwith.

HA over distance is difficult. It could work with some sort of load-balancer
approach we are using right now. That could not only used internaly, but
also externaly as a director. We will offer that as a service in combination
with DNS-Server. So you could have a DNS Server, and not use round robin,
but point to one IP, which is our load balancer. That passes through to one
or more of your real servers, if one fails it's taken out of the "cluster"
(it's not a real cluster thought).
We'll only set it up if there is a market for it. We'll probably start it at
some point, test the service for a year. If there are enough customers, we
leave it. Otherwise we'd switch over to DNS round robin or whatever, depends
on the clients.

JH