[Linux-HA] Can linux-ha manage a dual-uplink scenario

[Andreas Kurz]

On 1/29/07, Dirk Jagdmann <doj at cubic.org> wrote:
> Hello,
>
> I'd like to make a server fault tolerant by using two computers and
> provide some failover if one of them fails. As the servers need to
> communicate with other clients and server with the network I have to
> deal with the issue, that the network (and not my server) fails. However
> I'd like to do a server failover only, when the server itself (and not
> the connecting network) goes out of service, because the server contains
> complex databases were failover is not trivial. My servers are embedded
> in a collection of other servers and existing network infrastructure,
> which is basically already built redundantly.
>
> +---------+                           +---------+
> |Server A1|+------+           +------+|Server A2|
> +---------+       |           |       +---------+
>                    |           |
> +---------+       |           |       +---------+
> |Server B1|+---+  |           |  +---+|Server B2|
> +---------+    |  |           |  |    +---------+
>                +-------+   +-------+
>                |Switch1|+-+|Switch2|
>                +-------+   +-------+
>                 |    |       |    |
>              +--+    |       |    |
>              |       +----------+ |
>              |  +------------+  | |
>              |  |               | |
>            1a|  |1b           2a| |2b
>          +----------+       +----------+
>          |My Server1|       |My Server2|
>          +----------+       +----------+
>
> Server A1 and A2 both provide the same functionality, the same with B1
> and B2. So if either A1 or Switch1 fails my servers 1 and 2 can still
> reach A2. The High Availabilty restriction of this project is, that one
> component (either server or network [network includes the cable as
> well]) may fail. If two components fails simultaneously the project may
> suffer from (partial) failure.
>
> What I'd like to accomplish is, that if MyServer1 is currently active
> and Switch1 fails, I'd like to stay on MyServer1 and continue execution
> with ServerA2 and ServerB2 via Switch2. Further I have to avoid a
> Multi-Master (or Multi-Active) Situation, so my Servers 1 and 2 have to
> communicate and detect wheter they are both alive and then decide on a
> master node.

Use the ethernet bonding driver in active-backup mode for your interfaces.

>
> What I'd like to have is a configuration/software that runs on MyServer1
> and MyServer2 and detects which Servers a alive, then decide which
> server should be active/master. Then on the active server a shared IP
> address is configured for on is its interfaces (a or b). If the software
> detects a network failure on the configured interface it should switch
> the IP address to the other interface.

This is exactly what heartbeat can do for you.

>
> Can this be done with the current linux-ha components? Or would this
> require some enhancements to certain parts of linux-ha?

Of course ;-) ... heartbeat & bonding ... works fine for me.

>
> --
> ---> Dirk Jagdmann ^ doj / cubic
> ----> http://cubic.org/~doj
> -----> http://llg.cubic.org
> _______________________________________________
> 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
>