Re: Handling dropped attributes in pglogical_proto

On 29/09/16 05:33, Michael Paquier wrote:
> On Wed, Sep 28, 2016 at 11:25 PM, Konstantin Knizhnik
> <k(dot)knizhnik(at)postgrespro(dot)ru> wrote:
>> But if table was just altered and some attribute was removed from the table,
>> then rel->natts can be greater than natts.
>
> This is part of pglogical, so you may want to reply on the dedicated
> thread or send directly a patch to them. By the way, this code path
> may need to care as well about attisdropped. It is never good not to
> check for it.
>

Agreed this does not belong to hackers and should reported on github.

But just as note the rel variable is not Relation, it's
PGLogicalRelation which gets populated by relation message from the
upstream so if you observe the behavior mentioned in the original email,
you are probably doing something unexpected there (Konstantin is not
using vanilla pglogical). Also the attmap should never contain
attisdropped attributes.

--
Petr Jelinek http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services