Horton

Re: Allow CLUSTER, VACUUM FULL and REINDEX to change tablespace on the fly

Michael Paquier <michael@paquier.xyz>

View thread View on postgr.es
From: Michael Paquier <michael@paquier.xyz>
To: Justin Pryzby <pryzby@telsasoft.com>
Cc: Alexey Kondratov <a.kondratov@postgrespro.ru>, Alvaro Herrera <alvherre@alvh.no-ip.org>, Peter Eisentraut <peter.eisentraut@enterprisedb.com>, Masahiko Sawada <masahiko.sawada@2ndquadrant.com>, Steve Singer <steve@ssinger.info>, pgsql-hackers@lists.postgresql.org, Robert Haas <robertmhaas@gmail.com>, Alexander Korotkov <a.korotkov@postgrespro.ru>, Masahiko Sawada <sawada.mshk@gmail.com>, Jose Luis Tallon <jltallon@adv-solutions.net>
Date: 2021-02-15T02:58:02Z
Lists: pgsql-hackers 
On Sun, Feb 14, 2021 at 08:10:50PM -0600, Justin Pryzby wrote:
> Isn't this dead code ?

Nope, it's not dead.  Those two code paths can be hit when attempting
a reidex with a tablespace move directly on toast tables and indexes,
see:
=# create table aa (a text);
CREATE TABLE
=# select relname from pg_class where oid > 16000;
       relname
----------------------
 aa
 pg_toast_16385
 pg_toast_16385_index
(3 rows)
=# reindex (concurrently, tablespace pg_default) table
      pg_toast.pg_toast_16385;
ERROR:  0A000: cannot move system relation "pg_toast_16385"
LOCATION:  ReindexRelationConcurrently, indexcmds.c:3295
=# reindex (concurrently, tablespace pg_default) index
      pg_toast.pg_toast_16385_index;
ERROR:  0A000: cannot move system relation "pg_toast_16385_index"
LOCATION:  ReindexRelationConcurrently, indexcmds.c:3439

It is easy to save the relation name using \gset in a regression test,
but we had better keep a reference to the relation name in the error
message so this would not be really portable.  Using a PL function to
do that with a CATCH block would not work either as CONCURRENTLY
cannot be run in a transaction block.  This leaves 090_reindexdb.pl,
but I was not really convinced that this was worth the extra test
cycles (I am aware of the --tablespace option missing in reindexdb,
someone I know was trying to get that done for the next CF).
--
Michael

Commits

  1. Add TABLESPACE option to REINDEX

  2. Refactor code in tablecmds.c to check and process tablespace moves

  3. Refactor option handling of CLUSTER, REINDEX and VACUUM

  4. pg_dump: Don't use enums for defining bit mask values

  5. Refactor CLUSTER and REINDEX grammar to use DefElem for option lists

  6. Refactor parsing rules for option lists of EXPLAIN, VACUUM and ANALYZE

  7. Improve tab completion of REINDEX in psql

  8. Fix possible crash during FATAL exit from reindexing.