Re: why there is not VACUUM FULL CONCURRENTLY?

On Fri, Apr 11, 2025 at 5:28 PM Antonin Houska <ah@cybertec.at> wrote:
>
> Please check the next version [1]. Thanks for your input.
>
> [1] https://www.postgresql.org/message-id/97795.1744363522%40localhost
>

Hi, I’ve briefly experimented with v13-0001.

EXPLAIN tab complete:
explain (verbose O
OFF  ON

since we already touched the tab-complete for repack.
We can do it similarly.
you may see src/bin/psql/tab-complete.in.c line 4288.

------------------------------
Currently REPACK Synopsis section
looks like the screenshot attached.

make it one line
REPACK [ ( option [, ...] ) ] [ table_name [ USING INDEX index_name ] ]
would look intuitive, IMHO.
------------------------------
+repack_index_specification:
+ USING INDEX name { $$ = $3; }
+ | /*EMPTY*/ { $$ = NULL; }
+ ;

in gram.y line 4685, we have
ExistingIndex:   USING INDEX name                    { $$ = $3; }
        ;

so here, we can change it to

repack_index_specification:
            ExistingIndex
            | /*EMPTY*/                { $$ = NULL; }

-------------------------------------------------
+static List *
+get_tables_to_repack(MemoryContext repack_context)
+{
+ relrelation = table_open(RelationRelationId, AccessShareLock);
+ scan = table_beginscan_catalog(relrelation, 0, NULL);
+ while ((tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
+ {
+ RelToCluster *rtc;
+ Form_pg_class relrelation = (Form_pg_class) GETSTRUCT(tuple);
+ Oid relid = relrelation->oid;
+
+ /* Only interested in relations. */
+ if (get_rel_relkind(relid) != RELKIND_RELATION)
+ continue;


The doc said (Without a table_name, REPACK processes every table and
materialized view...)
but seems plain ``REPACK(verbose) ; ``
will not process materialized view?