Re: Inheriting table AMs for partitioned tables

On Tue, 5 Mar 2019 at 19:08, Andres Freund <andres(at)anarazel(dot)de> wrote:
>
> On 2019-03-05 16:01:50 +1300, David Rowley wrote:
> > I'd suggest it's made to work the same way as ca4103025dfe26 made
> > tablespaces work.
>
> Hm, is that actually correct? Because as far as I can tell that doesn't
> have the necessary pg_dump code to make this behaviour persistent:
>
> CREATE TABLESPACE frak LOCATION '/tmp/frak';
> CREATE TABLE test_tablespace (a text, b int) PARTITION BY list (a) TABLESPACE frak ;
> CREATE TABLE test_tablespace_1 PARTITION OF test_tablespace FOR VALUES in ('a');
> CREATE TABLE test_tablespace_2 PARTITION OF test_tablespace FOR VALUES in ('b') TABLESPACE pg_default;
> CREATE TABLE test_tablespace_3 PARTITION OF test_tablespace FOR VALUES in ('c') TABLESPACE frak;
>
> SELECT relname, relkind, reltablespace FROM pg_class WHERE relname LIKE 'test_tablespace%' ORDER BY 1;
> ┌───────────────────┬─────────┬───────────────┐
> │ relname │ relkind │ reltablespace │
> ├───────────────────┼─────────┼───────────────┤
> │ test_tablespace │ p │ 16384 │
> │ test_tablespace_1 │ r │ 16384 │
> │ test_tablespace_2 │ r │ 0 │
> │ test_tablespace_3 │ r │ 16384 │
> └───────────────────┴─────────┴───────────────┘

[pg_dump/pg_restore]

> ┌───────────────────┬─────────┬───────────────┐
> │ relname │ relkind │ reltablespace │
> ├───────────────────┼─────────┼───────────────┤
> │ test_tablespace │ p │ 16384 │
> │ test_tablespace_1 │ r │ 16384 │
> │ test_tablespace_2 │ r │ 16384 │
> │ test_tablespace_3 │ r │ 16384 │
> └───────────────────┴─────────┴───────────────┘

frak... that's a bit busted. I can't instantly think of a fix, but I
see the same problem does not seem to exist for partition indexes, so
that's a relief since that's already in PG11.

I'll take this up on another thread once I have something good to report.

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