Some thoughts on replication

Hi folks,

Maybe this belongs to the developers mailing list, but I'm not subscribed to
that list, so I post it here.

I've read through todo list on the website, especially the planned
replication feature. It seems to me that implementing replication into
postgresql is a rather hard and longterm task.

I want to show two possible alternatives, used in a real world application (a
billing system based on Informix) I was involved in a few years ago.

In our case we didn't need (or want) to replicate the whole database, but
only a few tables. The first thing was to use synonym tables. In Infomix one
can just type something like

create synonym mysynonym for table mytable:somedb@somehost;

where the somedb and somehost values are optional. It was then possible to
work on the synonym as if it where a normal table.

Later, we also did table replication based on triggers, a daemon and a
configuration table. The tables which had to be replicated where entered in
the config table, together with their destination. Then a little script was
used to setup the apprpriate insert, update and delete triggers for those
tables. When one of the triggers fired, the daemon contacted its friend on
the other host and did the same action on the remote database. I think there
was also a timestamp in the config table which had to be updated. I don't
know more details anymore, but that was roughly the concept.

Maybe one of those two (or both) is easier and faster to realise than
database replication.

Any comments?

Bye...

    Dirk