Re: BUG #4244: ALTER TABLE ... NO INHERIT problem

The following bug has been logged online:

Bug reference: 4244
Logged by: Tatsuhito Kasahara
Email address: kasahara(dot)tatsuhito(at)oss(dot)ntt(dot)co(dot)jp
PostgreSQL version: 8.3.3
Operating system: RedHat Linux 5.2
Description: ALTER TABLE ... NO INHERIT problem
Details:

Hi.

I performed "NO INHERIT" command to separate a table, then I got following
error.
This problem is reproducible on simple test.

:Session A
CREATE TABLE p_tbl (id int primary key);
CREATE TABLE c1_tbl (LIKE p_tbl INCLUDING INDEXES, CHECK (id between 1 and
100)) INHERITS(p_tbl);
CREATE TABLE c2_tbl (LIKE p_tbl INCLUDING INDEXES, CHECK (id between 101
and 200)) INHERITS(p_tbl);
INSERT INTO c1_tbl SELECT generate_series(1,100);
INSERT INTO c2_tbl SELECT generate_series(101,200);
BEGIN;
ALTER TABLE c1_tbl NO INHERIT p_tbl;
:Session B
SELECT * FROM p_tbl;
(Blocking)
:Session A
COMMIT;
:Session B (error occures)
ERROR: could not find inherited attribute "id" of relation "c1_tbl"
LOCATION: make_inh_translation_lists, prepunion.c:996

The comment in tablecmds.c says "AccessShareLock is probably enough".
But it actually needs more strong lock mode ?

ATExecDropInherit(Relation rel, RangeVar *parent)
/*
* AccessShareLock on the parent is probably enough, seeing that DROP
* TABLE doesn't lock parent tables at all. We need some lock since we'll
* be inspecting the parent's schema.
*/
parent_rel = heap_openrv(parent, AccessShareLock);

Best Regards.

"Tatsuhito Kasahara" <kasahara(dot)tatsuhito(at)oss(dot)ntt(dot)co(dot)jp> writes:
> The comment in tablecmds.c says "AccessShareLock is probably enough".
> But it actually needs more strong lock mode ?

> ATExecDropInherit(Relation rel, RangeVar *parent)
> /*
> * AccessShareLock on the parent is probably enough, seeing that DROP
> * TABLE doesn't lock parent tables at all. We need some lock since we'll
> * be inspecting the parent's schema.
> */
> parent_rel = heap_openrv(parent, AccessShareLock);

Well, as the comment points out, things are just as bad if you do
"DROP TABLE c1_tbl" instead of ALTER NO INHERIT.

I'm not sure there is a lot we can do about this. Taking a lock
on the parent table that is strong enough to block a SELECT would
just convert the problem into a deadlock (since a SELECT on the
parent will lock parent before child). Hardly seems like an
improvement.

regards, tom lane