[Freeipa-users] Need for some pull-style replication, or an alternate solution

[Petr Spacek]

On 20.8.2014 10:58, Dmitri Pal wrote:
> On 08/19/2014 07:55 PM, Joshua J. Kugler wrote:
>> A replica must connect to the master for initial setup; after that, the master
>> pushes to the replica.
>>
>> j
>>
>> On Tuesday, August 19, 2014 09:26:11 Ludwig Krispenz wrote:
>>> What's wrong with your scenario B: master(s) in internal network, they
>>> can contact consumers in DMZ and remote rack and replicate to them.
>>> What do you mean by "to contact for setup" ?
>>>
>>> Ludwig
>>>
>>> On 08/19/2014 03:12 AM, Joshua J. Kugler wrote:
>>>> So, we have a need for replication, but the existing push-only methodology
>>>> doesn't work for us. I suppose our problems could be attributed to over-
>>>> zealous network rules, but it is what it is. :) I'd love to change our
>>>> network policy, but we aren't in charge of network policy, and there is
>>>> no way I'm swaying the person that is.
>>>>
>>>> Topology:
>>>> 1) DMZ environments 1,...,n
>>>> 2) An Internal network
>>>> 3) A remote rack in a data center.
>>>>
>>>> Rules (by "talk" I mean initiate connections to):
>>>> 1) DMZs can talk to each other, somewhat.
>>>> 2) The Internal network can talk to the DMZs
>>>> 3) The DMZ *cannot* connect to the Internal network
>>>> 4) The remote rack of course cannot contact the Internal network, but can
>>>> contact the DMZs.
>>>>
>>>> Scenario A, Master in a DMZ:
>>>>    - Slave in the Internal network could contact the DMZ master for replica
>>>>
>>>> setup, but the Master could not contact the slave to push updates
>>>>
>>>>    - Slave in the remote rack could contact master in DMZ, but incoming to
>>>>
>>>> remote rack is very restrictive, so it is possible that master couldn't
>>>> push.
>>>>
>>>> Scenario B: Master in the Internal network.
>>>>
>>>>    - Slaves in DMZ and remote rack couldn't contact master for setup,
>>>>    although
>>>>
>>>> the master could contact the slaves to push updates.
>>>>
>>>> Scenario C: Master in remote rack
>>>>
>>>>    - Not acceptable as remote rack is a testing rack, and may go down at
>>>>    any
>>>>
>>>> time.
>>>>
>>>> So, the best solution, from my current understanding is being able to
>>>> somehow>
>>>> do pull-updates for replicas, because then we could have this:
>>>>    - Master in DMZs
>>>>    - Slaves in Internal network can contact Master in DMZ for replica setup
>>>>    and>
>>>> updates
>>>>
>>>>    - Slaves in remote rack can contact Master in DMZ for replica setup and
>>>>
>>>> updates
>>>>
>>>> Any feedback is appreciated, especially if I'm missing something...obvious
>>>> or minor.
>>>>
>>>> j
> I think you capture the problem correctly. There is unfortunately no solution
> for this at the moment.
> We considered looking into read only replicas but this is not exactly what
> would help here either since changes to RO replicas would be rerouted to the
> real masters and thos need to be accessible from DMZ or remote req in your
> case - so it will be inbound connection here.
> I am not sure there is a way to help you here with the current software. The
> only option I see is a two different domains - internal and external with some
> manually set trust in between. You bight be able to sync people in some way
> with some scripts but still that would be quite fragile.
> Are the users operating inside the FW and in DMZ/remote are really same users?

Or IPA-to-IPA trust? :-)

Joshua, if you want to experiment:

Ludwig said earlier in this thread that 389 replication will work fine if 
master is inside internal network (and thus able to contact other replicas in 
DMZ or external network).

It seems to me that main *technical* problem is "replica setup" phase where 
replica contacts the original master and not the other way around.

You can use e.g. SSH from master to replica and do some tricks with port 
forwarding and iptables/routing table so the replica will be able to contact 
the master inside internal network. (That will breach all policies you have, 
of course.)

If you want to experiment even more, you can try to use port forwarding for 
replica setup and then close the hole. 389 replication should work because 
master will connect to replica and not the other way around. I'm not sure what 
else will break...

-- 
Petr^2 Spacek