Add support for tuple routing to foreign partitions

From: Etsuro Fujita <fujita(dot)etsuro(at)lab(dot)ntt(dot)co(dot)jp>
To: pgsql-hackers(at)postgresql(dot)org
Cc: Langote_Amit_f8(at)lab(dot)ntt(dot)co(dot)jp
Subject: Add support for tuple routing to foreign partitions
Date: 2017-06-29 11:20:41
Message-ID: 29906a26-da12-8c86-4fb9-d8f88442f2b9@lab.ntt.co.jp
Views: Raw Message | Whole Thread | Download mbox
Thread:
Lists: pgsql-hackers

Hi,

Here is a patch for $subject. This is based on Amit's original patch
(patch '0007-Tuple-routing-for-partitioned-tables-15.patch' in [1]).
Main changes are:

* I like Amit's idea that for each partition that is a foreign table,
the FDW performs query planning in the same way as in non-partition
cases, but the changes to make_modifytable() in the original patch that
creates a working-copy of Query to pass to PlanForeignModify() seem
unacceptable. So, I modified the changes so that we create
more-valid-looking copies of Query and ModifyTable with the foreign
partition as target, as I said before. In relation to this, I also
allowed expand_inherited_rtentry() to build an RTE and AppendRelInfo for
each partition of a partitioned table that is an INSERT target, as
mentioned by Amit in [1], by modifying transformInsertStmt() so that we
set the inh flag for the target table's RTE to true when the target
table is a partitioned table. The partition RTEs are not only
referenced by FDWs, but used in selecting relation aliases for EXPLAIN
(see below) as well as extracting plan dependencies in setref.c so that
we force rebuilding of the plan on any change to partitions. (We do
replanning on FDW table-option changes to foreign partitions, currently.
See regression tests in the attached patch.)

* The original patch added tuple routing to foreign partitions in COPY
FROM, but I'm not sure the changes to BeginCopy() in the patch are the
right way to go. For example, the patch passes a dummy empty Query to
PlanForeignModify() to make FDWs perform query planning, but I think
FDWs would be surprised by the Query. Currently, we don't support COPY
to foreign tables, so ISTM that it's better to revisit this issue when
adding support for COPY to foreign tables. So, I dropped the COPY part.

* I modified explain.c so that EXPLAIN for an INSERT into a partitioned
table displays partitions just below the ModifyTable node, and shows
remote queries for foreign partitions in the same way as EXPLAIN for
UPDATE/DELETE cases. Here is an example:

postgres=# explain verbose insert into pt values (1), (2);
QUERY PLAN
-------------------------------------------------------------------
Insert on public.pt (cost=0.00..0.03 rows=2 width=4)
Foreign Insert on public.fp1
Remote SQL: INSERT INTO public.t1(a) VALUES ($1)
Foreign Insert on public.fp2
Remote SQL: INSERT INTO public.t2(a) VALUES ($1)
-> Values Scan on "*VALUES*" (cost=0.00..0.03 rows=2 width=4)
Output: "*VALUES*".column1
(7 rows)

Comments are welcome! Anyway, I'll add this to the next commitfest.

Best regards,
Etsuro Fujita

[1]
https://www.postgresql.org/message-id/aa874eaf-cd3b-0f75-c647-6c0ea823781e%40lab.ntt.co.jp

Attachment Content-Type Size
tuple-routing-to-foreign-partitions-v1.patch text/plain 32.6 KB

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Etsuro Fujita 2017-06-29 11:28:21 Re: Declarative partitioning - another take
Previous Message Masahiko Sawada 2017-06-29 10:09:19 Re: What does it mean by XLOG_BACKUP_RECORD?