On 26 January 2018 at 01:13, Wolf Humble <wolfhumble(a)gmail.com> wrote:
On more question about teaming -- that wouldn't work in my case
since
I want ALL IPs to answer at the same time -- but just curious:
It seems like teaming with activebackup would work even for interfaces
that are in different broadcast domains, as long as you don't mind
that only IPs connected to eth1 will answer when eth1 and eth2 are up,
and only IPs connected to eth2 willl answer when eth1 is down. (From
what I remember when testing the activebackup mode earlier in my
setup).
Do I understand that correctly?
That's almost correct.
It will depend on which is the *active slave* at the time, which is a
separate thing to linkstate. You can have both slave links up and
switch which is the active one. You can change whether a preferred
slave always takes over when it's up. See the prio/sticky options on
the active-backup runner. Of course a slave must actually be up to be
selected as the active slave!
Once the active slave is decided, the system will only receive traffic
from one set of IPs (either aaa+bbb or ccc+ddd) depending on which
slave is up, hence the system will only reply to those IPs speaking in
the active slave. You're right that the incoming traffic will just
happen to be from different subnets.
However, there's nothing to stop the system trying to use its own IP
in the wrong network, say sending ARP requests for other hosts in
network aaa out the team. Whether anything responds to those requests
will depend on which slave is up.
It would also be tricky to manage which default gateway is in use
depending on which slave is up, so it would be a pretty terrible
solution for failover.
With your interface config, you may be more interested in policy
routing and multiple upstream gateways:
http://lartc.org/howto/lartc.rpdb.multiple-links.html
Jamie