Re: ATTACH/DETACH PARTITION CONCURRENTLY

From: Robert Haas <robertmhaas(at)gmail(dot)com>
To: Alvaro Herrera <alvherre(at)2ndquadrant(dot)com>
Cc: Michael Paquier <michael(at)paquier(dot)xyz>, Sergei Kornilov <sk(at)zsrv(dot)org>, Amit Langote <langote_amit_f8(at)lab(dot)ntt(dot)co(dot)jp>, David Rowley <david(dot)rowley(at)2ndquadrant(dot)com>, Andres Freund <andres(at)anarazel(dot)de>, PostgreSQL Hackers <pgsql-hackers(at)lists(dot)postgresql(dot)org>, Simon Riggs <simon(at)2ndquadrant(dot)com>
Subject: Re: ATTACH/DETACH PARTITION CONCURRENTLY
Date: 2019-01-29 18:59:39
Message-ID: CA+TgmoZGa==6n7qnqXnj2WP507oWWh3YV8A7=b1tkJ-xy0dsQw@mail.gmail.com
Views: Raw Message | Whole Thread | Download mbox
Thread:
Lists: pgsql-hackers

On Fri, Jan 25, 2019 at 4:18 PM Alvaro Herrera <alvherre(at)2ndquadrant(dot)com> wrote:
> > I wrote a little patch that stores the relation OIDs of the partitions
> > into the PartitionedPruneRelInfo and then, at execution time, does an
> > Assert() that what it gets matches what existed at plan time. I
> > figured that a good start would be to find a test case where this
> > fails with concurrent DDL allowed, but I haven't so far succeeded in
> > devising one. To make the Assert() fail, I need to come up with a
> > case where concurrent DDL has caused the PartitionDesc to be rebuilt
> > but without causing an update to the plan. If I use prepared queries
> > inside of a transaction block, [...]
>
> > I also had the idea of trying to use a cursor, because if I could
> > start execution of a query, [...]
>
> Those are the ways I thought of, and the reason for the shape of some of
> those .spec tests. I wasn't able to hit the situation.

I've managed to come up with a test case that seems to hit this case.

Preparation:

create table foo (a int, b text, primary key (a)) partition by range (a);
create table foo1 partition of foo for values from (0) to (1000);
create table foo2 partition of foo for values from (1000) to (2000);
insert into foo1 values (1, 'one');
insert into foo2 values (1001, 'two');
alter system set plan_cache_mode = force_generic_plan;
select pg_reload_conf();

$ cat >x
alter table foo detach partition foo2;
alter table foo attach partition foo2 for values from (1000) to (2000);
^D

Window #1:

prepare foo as select * from foo where a = $1;
explain execute foo(1500);
\watch 0.01

Window #2:

$ pgbench -n -f x -T 60

Boom:

TRAP: FailedAssertion("!(partdesc->nparts == pinfo->nparts)", File:
"execPartition.c", Line: 1631)

I don't know how to reduce this to something reliable enough to
include it in the regression tests, and maybe we don't really need
that, but it's good to know that this is not a purely theoretical
problem. I think next I'll try to write some code to make
execPartition.c able to cope with the situation when it arises.

(My draft/WIP patches attached, if you're interested.)

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company

Attachment Content-Type Size
0008-Drop-lock-level.patch application/octet-stream 749 bytes
0006-Adapt-the-executor-to-use-a-PartitionDirectory.patch application/octet-stream 7.0 KB
0005-Adapt-the-optimizer-to-use-a-PartitionDirectory.patch application/octet-stream 6.0 KB
0007-relid_map-crosschecks.patch application/octet-stream 5.3 KB
0004-Postpone-old-context-removal.patch application/octet-stream 1.5 KB
0003-Initial-cut-at-PartitionDirectory.patch application/octet-stream 4.1 KB
0002-Ensure-that-RelationBuildPartitionDesc-sees-a-consis.patch application/octet-stream 6.1 KB
0001-Move-code-for-managing-PartitionDescs-into-a-new-fil.patch application/octet-stream 23.0 KB

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Andres Freund 2019-01-29 19:25:41 Re: Why are we PageInit'ing buffers in RelationAddExtraBlocks()?
Previous Message Andrey Borodin 2019-01-29 18:58:06 Re: Covering GiST indexes