On Fri, Mar 9, 2018 at 8:43 AM, Alexander Korotkov
<a.korotkov@postgrespro.ru> wrote:
> Hi!
>
Sorry for my late reply.
> I'd like to propose a revised patch based on various ideas upthread.
Thank you for proposing the patch!
>
> This patch works as following.
>
> 1) B-tree meta page is extended with 2 additional parameters:
> * btm_oldest_btpo_xact – oldest btpo_xact among of deleted pages,
> * btm_last_cleanup_num_heap_tuples – number of heap tuples during last
> cleanup scan.
>
> 2) These parameters are reset during btbulkdelete() and set during
> btvacuumcleanup().
Can't we set these parameters even during btbulkdelete()? By keeping
them up to date, we will able to avoid an unnecessary cleanup vacuums
even after index bulk-delete.
>
> 3) Index scans during second and subsequent btvacuumcleanup() happen only if
> btm_oldest_btpo_xact is older than RecentGlobalXmin
> OR num_heap_tuples >= btm_last_cleanup_num_heap_tuples(1 +
> vacuum_cleanup_index_scale_factor).
>
> In other words btvacuumcleanup() scans the index only if there are
> recyclable pages,
> or index statistics is stalled (inserted more than
> vacuum_cleanup_index_scale_factor
> since last index statistics collection).
>
> 4) vacuum_cleanup_index_scale_factor can be set either by GUC or reloption.
> Default value is 0.1. So, by default cleanup scan is triggered after
> increasing of
> table size by 10%.
>
> 5) Since new fields are added to the metapage, BTREE_VERSION is bumped.
> In order to support pg_upgrade, read of previous metapage version is
> supported.
> On metapage rewrite, it's upgraded to the new version.
>
> So, since we don't skip scan of recyclable pages, there is no risk of xid
> wraparound.
> Risk of stalled statistics is also small, because
> vacuum_cleanup_index_scale_factor
> default value is quite low. User can increase
> vacuum_cleanup_index_scale_factor
> on his own risk and have less load of B-tree cleanup scan bought by more gap
> in
> index statistics.
Agreed.
I've not reviewed the code deeply yet but the regression test of this
patch seems to wrongly pass the regression tests and bt_metap()
function of pageinspect needs to be updated. Attached an updated patch
fixed these issue. Will review the patch again.
Regards,
--
Masahiko Sawada
NIPPON TELEGRAPH AND TELEPHONE CORPORATION
NTT Open Source Software Center
