Interesting LDAP replication scenario

["Geoffrey L. Wright" <geoff@akwebart.com>]

The title says it all.  I'm fairly comfortable with developing 
against OpenLDAP and have worked on several fairly large deploys.  
But all have been very straitforward from a replication 
standpoint.  I have in front of me now a replication task that 
looks fairly tricky -- at least to me...

A picture is worth a thousand words, so here's one that will 
hopefully make my situation a bit more clear:

http://www.northernwastes.org/~geoff/rep_diagram.png

Now here's a verbal summary of my scenario:

I have a number of "districts", each which contains a number of 
separate physical sites.  All sites in a given district have a 
common suffix and need to share information.  Each district has a 
separate database.  Right now I have replication working in a 
pretty standard way:


1. An LDAP client submits a modify request of some sort to the 
slave slapd server.

2. THE slave slapd refers the LDAP client the master slapd.

3. THE master slapd does the modify and writes the change to the 
replication log.

4. slurpd sends the change to all slave slapd servers in the district.


So far so good.  Everything is pretty much by the book.  But here's 
my problem:

Many of my sites are in very rural locations and have very 
tentative connections to the rest of the world.  Connection 
downtime measured in days is possible in some cases.  At many 
individual sites downtime measured in hours is unfortunately fairly 
common.  When new information is added to any individual site the 
site slapd instance I need it to to be accessible to the users at 
that site immediately.  There are currently about a hundred of 
individual sites, with more to be added over time.

Here's what I would _like_ to see:


1. The LDAP client submits a modify request to a site (master? 
slave?) slapd server

2. This site slapd server immediate does the modify then (if it is 
available) passes the request to the master slapd server at home 
base.  If not, the site slapd instance periodically checks back 
until such time as the home base master slapd server becomes 
available and then passes on the request.

3. The home base master slapd does the modify and writes the change 
to the replication log.

4. slurpd sends the change to all site (master? slave?) slapd 
servers in the district _except_ the original location.


Can this scenario be made to work?  Is this really the best way to 
do things, or is there another way?

Any help. thoughts or suggestions at this point would be greatly 
appreciated...


//glw