Re: unique indexes on partitioned tables

From: Alvaro Herrera <alvherre(at)alvh(dot)no-ip(dot)org>
To: Peter Eisentraut <peter(dot)eisentraut(at)2ndquadrant(dot)com>, Jaime Casanova <jaime(dot)casanova(at)2ndquadrant(dot)com>
Cc: Jesper Pedersen <jesper(dot)pedersen(at)redhat(dot)com>, Amit Langote <Langote_Amit_f8(at)lab(dot)ntt(dot)co(dot)jp>, Pg Hackers <pgsql-hackers(at)postgresql(dot)org>
Subject: Re: unique indexes on partitioned tables
Date: 2018-02-19 20:45:31
Message-ID: 20180219204531.tsblwxwtz332pp7e@alvherre.pgsql
Views: Raw Message | Whole Thread | Download mbox | Resend email
Lists: pgsql-hackers

I pushed this now, with fixes for the last few comments there were.

Peter Eisentraut wrote:

> I don't understand the variable name "third". I don't see a "first" or
> "second" nearby.

Hah. That was referring to variables "myself" and "referenced". I
changed the variable name to "parentConstr".

> I find some of the columns in pg_constraint confusing. For a primary
> key on a partitioned table, for the PK on the partition I get
> conislocal = false, coninhcount = 1, connoinherit = true
> The last part is confusing to me.

Yeah, I think it was patently wrong. I changed it so that connoinherit
becomes true in this case.

Alvaro Herrera wrote:
> Jaime Casanova wrote:

> > also noted that if you:
> >
> > """
> > create table t1(i int) partition by hash (i);
> > create table t1_0 partition of t1 for values with (modulus 2, remainder 0);
> > create table t1_1 partition of t1 for values with (modulus 2, remainder 1);
> > create unique index on t1(i);
> > alter table t1 add primary key using index t1_i_idx ;
> > """
> >
> > the ALTER TABLE ADD PK does not recurse to partitions, which maybe is
> > perfectly fine because i'm using USING INDEX but it feels like an
> > oversight to me
> Ouch. Yeah, this is a bug. I'll try to come up with something.

After looking at it for a few minutes I determined that adding this
feature requires some more work: you need to iterate on all partitions,
obtain the corresponding index, cons up a few fake parse nodes, then
recurse to create the PK in the children. I think this should be doable
with a couple dozen lines of code, but it's a refinement that can be
added on top. Care to submit a patch? In the meantime, I added an
ereport(ERROR) to avoid leaving the system in an inconsistent state
(that probably would not even be reproduced correctly by pg_dump).

Álvaro Herrera
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

In response to


Browse pgsql-hackers by date

  From Date Subject
Next Message Arthur Zakirov 2018-02-19 21:01:30 Re: [PROPOSAL] Nepali Snowball dictionary
Previous Message Brent Kerby 2018-02-19 20:42:31 Option to ensure monotonic timestamps