On 2013-10-15 11:02:39 -0400, Robert Haas wrote:
> >> If the plugin interface isn't rich enough to provide a convenient way
> >> to avoid that, then it needs to be fixed so that it is, because it
> >> will be a common requirement.
> >
> > Oh, it surely is possibly to avoid repeating it. The output plugin
> > interface simply gives you a relcache entry, that contains everything
> > necessary.
> > The output plugin would need to keep track of whether it has output data
> > for a specific relation and it would need to check whether the table
> > definition has changed, but I don't see how we could avoid that?
>
> Well, it might be nice if there were a callback for, hey, schema has
> changed! Seems like a lot of plugins will want to know that for one
> reason or another, and rechecking for every tuple sounds expensive.
I don't really see how we could provide that in any useful manner. We
could provide a callback that is called whenever another transaction has
changed the schema, but there's nothing easily to be done about schema
changes by the replayed transaction itself. And those are the only ones
where meaningful schema changes can happen since the locks the source
transaction has held will prevent most other schema changes.
As much as I hate such code, I guess checking (and possibly storing) the
ctid||xmin of the pg_class row is the easiest thing we could do :(.
Greetings,
Andres Freund
-- Andres Freund http://www.2ndQuadrant.com/PostgreSQL Development, 24x7 Support, Training &
Services
