Re: lazy_truncate_heap()

Simon Riggs wrote:
> On Wed, 2008-12-31 at 21:45 +0200, Heikki Linnakangas wrote:
>>> Can I fix?
>> Yes please.
>
> Fix attached.

> --- 183,192 ----
> * number of pages. Otherwise, the time taken isn't worth it.
> */
> possibly_freeable = vacrelstats->rel_pages - vacrelstats->nonempty_pages;
> ! if (vacrelstats->tuples_deleted > 0 &&
> ! (possibly_freeable >= REL_TRUNCATE_MINIMUM ||
> ! (possibly_freeable >= vacrelstats->rel_pages / REL_TRUNCATE_FRACTION &&
> ! possibly_freeable > 0)))
> lazy_truncate_heap(onerel, vacrelstats);
>

Where did that "tuples_deleted > 0" condition come from? It seems
counter-productive; if a previous vacuum failed to acquire the lock,
subsequent vacuums wouldn't even try if they don't remove any tuples.
How about simply:

***************
*** 183,190 ****
* number of pages. Otherwise, the time taken isn't worth it.
*/
possibly_freeable = vacrelstats->rel_pages -
vacrelstats->nonempty_pages;
! if (possibly_freeable >= REL_TRUNCATE_MINIMUM ||
! possibly_freeable >= vacrelstats->rel_pages / REL_TRUNCATE_FRACTION)
lazy_truncate_heap(onerel, vacrelstats);

/* Vacuum the Free Space Map */
--- 183,191 ----
* number of pages. Otherwise, the time taken isn't worth it.
*/
possibly_freeable = vacrelstats->rel_pages -
vacrelstats->nonempty_pages;
! if (possibly_freeable > 0 &&
! (possibly_freeable >= REL_TRUNCATE_MINIMUM ||
! possibly_freeable >= vacrelstats->rel_pages / REL_TRUNCATE_FRACTION))
lazy_truncate_heap(onerel, vacrelstats);

/* Vacuum the Free Space Map */

--
Heikki Linnakangas
EnterpriseDB http://www.enterprisedb.com