Topic: Failover - Fault Tolerance

[Kelvin]

Hi all,

As long as we're wishing, might as well toss in a doozy.

Good job on the software RAID. Now, in addition to that, it would be a very good (no, GREAT) thing to have if SME could have failover capabilities built in.

Eg :

Network has 2 (or more) SME servers connected in the network. Any data saved to one is also saved / replicated to the other (with or without the need for a shared storage system - in fact, not shared is preferrable). If primary server 'dies', the secondary server detects this, makes necessary adjustments to its own configuration, then steps in and takes over.

A bit like clustering but don't really need to go that far. Load balancing is nice but not absolutely necessary. Main aim is fault tolerance / redundancy and not load balancing / load sharing.

Cheers !

Kelvin

[Judy Morgann]

Hi,

greate idea, the same i´m trying to set up. I want to make a rpm of heartbeat for e-smith - if you don´t know, it checks over a second networkline oder serial cable the status of the other server. If one fails the other takes over it´s ip address and services.
At the moment we do it with 4 RedHat Servers, but for some customers that want to replace theire NT-Server with e-smith i´ll try it.
To keep the data up to date on each server set up a cron job that copies the data via ssh/rsync from the master to the backup server.
One problem that i haven´t a solution at the moment is how to copy the users/samba-users/machineaccounts (when using e-smith as pdc logon-server).
Somebody any ideas?
I think Failover would be a very nice and often used feature of e-smith.

greetings
judy

[jeff martin]

you might check out   http://www.ifost.org.au/~peterw/    to get you started.

Just make sure you let me know when your rpm is ready so I can try it out : )

Thanks,
Jeff

[guestHH]

Hi,
Gave Unison at http://www.ifost.org.au/~peterw/ a try.

Works very well (so far only tested i-bay -> ibay on the same machine, need to test ibay -> winbox) , and a rpm is not really needed.

Just to let you know.

Thanks,

guestHH

[Kelvin]

Unison would be great as a sync / backup tool. Unfortunately, it cannot (and was not intended to) do server failovers. What we really need is a simple (if such a thing is possible to be made 'simple' !) server failover mechanism.

Even in the Windows world, simple 'failover' software don't go far enough - so much so that I question their use of the term 'failover'. Most of them are much like Unison, file sync utilities. I have not come across one that does true Server failover (ie. Server 1 dies, Server 2 takes over as if it were Server 1 - much like what hot spare in a RAID pack does) without going to the extent of configuring a server cluster.

So, if SME can have true (and simple !) failover functionality, it would really be revolutionary !

Cheers,

Kelvin

[koolmoose]

Kelvin wrote:
>
> Hi all,
>
> As long as we're wishing, might as well toss in a doozy.
>
> Good job on the software RAID. Now, in addition to that, it
> would be a very good (no, GREAT) thing to have if SME could
> have failover capabilities built in.
>
> Eg :
>
> Network has 2 (or more) SME servers connected in the network.
> Any data saved to one is also saved / replicated to the other
> (with or without the need for a shared storage system - in
> fact, not shared is preferrable). If primary server 'dies',
> the secondary server detects this, makes necessary
> adjustments to its own configuration, then steps in and takes
> over.
>
> A bit like clustering but don't really need to go that far.
> Load balancing is nice but not absolutely necessary. Main aim
> is fault tolerance / redundancy and not load balancing / load
> sharing.
>
I would be most interested in having such a system do the backup functions that we expect a tape drive to do and have the machine in a different room on the same network.  It seems to me that an older machine with a same size drive could be used as a 'hot spare' with rsync backups done as 'nice' processes to avoid unnecessary load. The main issue is loss of data and reasonable recovery time to me. Even a machine that ran once or twice a day would be useful.

[Hsing-Foo]

Kelvin,

Did you take a look at HeartBeat (http://www.linux-ha.org) ?

It works quit well.

JFYI

RequestedDeletion