> I don't think it is reasonable to support blobs directly in Slony. They 
> are not well supported in Postgres to begin with (no triggers, no 
> back-link to table, no possible referential integrity, backup/recovery 
> problems).

Well, this leaves postgres with lack of support for a byte array data
type which can be streamed and replicated... of course the user can work
it around, but it surely would be nice to have it built in.

> A possible way would be to move to bytea. I do understand that you don't 
> want to deal with the whole blob data in memory at once, but who said 
> that you have to store the whole object in one bytea column? It is well 
> possible to create a separate table that holds chunks of that data, and 
> the application can access them via cursor. That's close enough to large 
> object reading in my book. And it fully supports insert/update/delete 
> and replication via Slony-I.

I could do this, but I have to support oracle too and at least at some
higher level the behavior should be the same from the application POV.
This might be a PITA...

But I guess I'll go for it as there's no better choice.

Cheers,
Csaba.