Subject: Re: adding partitioned tables to publications
To: Amit Langote <amitlangote09@gmail.com>
References: 
 <CA+HiwqH=Y85vRK3mOdjEkqFK+E=ST=eQiHdpj43L=_eJMOOznQ@mail.gmail.com>
 <CA+HiwqEF4K+SxBZkU-=q6krGD74ysZZZv5z10_2C=+UcE2vTYw@mail.gmail.com>
 <CA+FpmFf2HVQYRQNh1ZvWPwatyr8_5n4pHRzNtmD5zVqs1zoLbQ@mail.gmail.com>
 <CA+HiwqEgn5da9KvWP4y0rBjcMtHdRao01Lr6TgTw78+UGceJwQ@mail.gmail.com>
 <037d3133-1689-108c-360d-b782e3209cfb@2ndquadrant.com>
 <CA+HiwqHf8_bcyHt=6R4O_Ju4P_Wz6wu1MkOAe14URdpdafS_KQ@mail.gmail.com>
 <CA+HiwqEQb5-26-=-K=vrWKQhvn7kqOfS-Vgj+H2YMw5XtdUUQw@mail.gmail.com>
 <eed6fb07-94d6-a2a5-33e1-84679dca9363@2ndquadrant.com>
 <CA+HiwqF7JD0CLkN12KNuGsf5GfaLuVb-Nc6+nTyL_iOhrSuA0g@mail.gmail.com>
 <CA+HiwqFWK3raKUU5Fp789jBsju-pYVpAiw0OsuO5sMX1CkYwFQ@mail.gmail.com>
Cc: Rafia Sabih <rafia.pghackers@gmail.com>,
 PostgreSQL-development <pgsql-hackers@postgresql.org>
From: Peter Eisentraut <peter.eisentraut@2ndquadrant.com>
Organization: 2ndQuadrant
Message-ID: <b1e7d3c5-d4f8-2bc6-80a1-18b18c5db347@2ndquadrant.com>
Date: Wed, 20 Nov 2019 08:55:39 +0100
User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.14; rv:52.0)
 Gecko/20100101 Thunderbird/52.9.1
MIME-Version: 1.0
In-Reply-To: 
 <CA+HiwqFWK3raKUU5Fp789jBsju-pYVpAiw0OsuO5sMX1CkYwFQ@mail.gmail.com>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 2019-11-18 09:53, Amit Langote wrote:
> I have spent some time hacking on this.  With the attached updated
> patch, adding a partitioned table to publication results in publishing
> the inserts, updates, deletes of the table's leaf partitions as
> inserts, updates, deletes of the table itself (it all happens inside
> pgoutput).  So, the replication target table doesn't necessarily have
> to be a partitioned table and even if it is partitioned its partitions
> don't have to match one-to-one.
> 
> One restriction remains though: partitioned tables on a subscriber
> can't accept updates and deletes, because we'd need to map those to
> updates and deletes of their partitions, including handling a tuple
> possibly moving from one partition to another during an update.

Right.  Without that second part, the first part isn't really that 
useful yet, is it?

I'm not sure what your intent with this patch is now.  I thought the 
previous behavior -- add a partitioned table to a publication and its 
leaf tables appear in the replication output -- was pretty welcome.  Do 
we not want that anymore?

There should probably be an option to pick the behavior, like we do in 
pg_dump.

What happens when you add a leaf table directly to a publication?  Is it 
replicated under its own identity or under its ancestor partitioned 
table?  (What if both the leaf table and a partitioned table are 
publication members?)

-- 
Peter Eisentraut              http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services