Re: Sanity Check?

"Larry Rosenman" <ler@lerctr.org> writes:
> SCO is seeing the following failure without -O, but no failure with -O:

> *** ./expected/sanity_check.out Wed Jul 27 17:58:12 2005
> --- ./results/sanity_check.out  Wed Jul 27 18:09:41 2005
> ***************
> *** 17,22 ****
> --- 17,24 ----
>    circle_tbl          | t
>    fast_emp4000        | t
>    func_index_heap     | t
> +  gcircle_tbl         | t
> +  gpolygon_tbl        | t
>    hash_f8_heap        | t
>    hash_i4_heap        | t
>    hash_name_heap      | t

Hmm.  This looks like a race condition in the test to me.  gcircle_tbl
and gpolygon_tbl are temp tables created during the create_index test.
They do have indexes, so if the backend that ran create_index hadn't
managed to delete 'em yet, it'd make sense that they show up in
sanity_check's query.  And in the parallel regression schedule,
create_index does run directly before sanity_check.

Those temp tables are recently introduced, I believe, so the fact that
this hasn't been reported before doesn't mean it can't happen elsewhere
than SCO machines.

It's pretty surprising though that sanity_check manages to execute
a complete database-wide VACUUM before the create_index backend has
managed to drop its couple of temp tables.  So there may be something
wrong with this explanation; or there might be something weird about
the kernel scheduling policy on their machine.
        regards, tom lane