No, I'm not asking how to do replication in Postgres, or when replication
will be written. I'm just trying to understand how replication would
work.
And yes - This is a (slightly) serious question. If I can understand how
replication works I would love to start figuring out how get Postgres to
do it...
Q: Does replication normally work bidirectionally?
If Yes:
If so how do database systems deal with disconnected activity where
inserts into one of the databases can conflict with data inserted
into the other (ie unique indexed)?
Is it acceptable to force all inserts to go to one database, and
then allow them to replicate to slave servers [and is this what
Oracle and the others do]? If so does/should the user connect to
the master directly or does the backend handle passing the query up
to the "master" and reports success to the user once the transaction
has been accepted/committed by the master?
If No:
Does each server cache the requests and then commit them as soon as
it can communicate with the other server?
How does a client know if their transaction was actually accepted?
How are conflicts resolved? Does a human have to get involved?
Like I said above I'm trying to understand this so I can figure out how it
could be implemented in Postgres.
Chris