Alvaro Herrera <alvherre@commandprompt.com> writes:
> Hmm ... that would be strange. Off-the-cuff idea: we introduced code to
> advance relfrozenxid in CLUSTER, TRUNCATE and table-rewriting forms of
> ALTER TABLE. Perhaps the problem is that we're neglecting to update it
> for the toast table there. AFAIR I analyzed the cases and they were all
> handled, but perhaps I forgot something.
I found a smoking gun ...
regression=# create table foo (f1 serial primary key, f2 text);
NOTICE: CREATE TABLE will create implicit sequence "foo_f1_seq" for serial column "foo.f1"
NOTICE: CREATE TABLE / PRIMARY KEY will create implicit index "foo_pkey" for table "foo"
CREATE TABLE
regression=# insert into foo values(1,repeat('xyzzy',100000));
INSERT 0 1
regression=# insert into foo values(2,repeat('xqzzy',100000));
INSERT 0 1
regression=# select relname, relkind, relfrozenxid from pg_class order by oid desc limit 6; relname |
relkind| relfrozenxid
-----------------------+---------+--------------foo_pkey | i | 0pg_toast_707220_index | i
| 0pg_toast_707220 | t | 119421foo | r |
119421foo_f1_seq | S | 0xmlview5 | v | 0
(6 rows)
regression=# cluster foo_pkey on foo;
CLUSTER
regression=# select relname, relkind, relfrozenxid from pg_class order by oid desc limit 6; relname |
relkind| relfrozenxid
-----------------------+---------+--------------pg_toast_707231_index | i | 0pg_toast_707231 | t
| 119424foo_pkey | i | 0foo | r |
4195086720foo_f1_seq | S | 0xmlview5 | v | 0
(6 rows)
So something is out of whack in CLUSTER. However it only seems to be
broken in HEAD, so I'm not sure this helps to explain the original
report. (Speculation: this is related to the rewrite to make CLUSTER
MVCC-safe?)
regards, tom lane