On Mon, Dec 5, 2011 at 9:28 AM, Cédric Villemain
<cedric.villemain.debian at gmail.com> wrote:
> Le 5 décembre 2011 10:12, Laura Passigni <passigni at navionics.it> a écrit :
>> Dear all,
>>
>> I’m using slony to have a replica of my database (Postgresql 8.4).
>>
>> Currently I replicate only some tables, not the entire database. I was
>> wondering if it is possible to replicate only some records of a table, for
>> example according to some field’s value. In other words, perform a sort of
>> conditional replica.
>
> It is not in the features list of slony.
> I don't think it is possible without rewriting triggers. Maybe have a
> look at logtrigger().

It ought to become plausible after 2.2 is released.

We have the change (still being integrated in; I think Jan is planning
to get this into master this week; you can take a peek at the repos
for Jan/Steve/Myself at GitHub to see what things are looking like)
that the log data is being split apart rather differently, notably so
that the log data is no longer "cooked" into text, but is, instead,
kept as a series of arrays.

That should make it possible to inject a "hook" function into the log
application function to reinterpret what is to be done with the log
data.

The first aspect that we expect to be pretty easy is to replay data
into a different table/schema; that's extremely useful for the case
where people want to replicate several instances of an application
into a "consolidation database."  This just involves renaming the
target table, and I wouldn't expect this to be terribly difficult to
deal with.  It should be able to be handled pretty cleanly.
(Hopefully.)

But as a broader matter, it would be interesting to add in "hooks" to
do more sophisticated things such as:

- Renaming columns.  Describing the configuration well is likely to be
the hard part.

- Selective replication, based on interpreting the data that is
provided.  Given a "hook" function to override the built-in behaviour,
it should surely be possible to do "just about anything."  Though the
higher the flights of fancy, the wiser this likely isn't...

Rewriting the triggers, as they exist now, seems horribly unattractive
to me.  They're SPI C functions, and there's enough intricacy there
that I'd think it pretty dangerous to hack on them without a pretty
deep understanding of them.  But this changes in 2.2...