On Tue, Mar 23, 2021 at 1:31 PM Michael Paquier <michael@paquier.xyz> wrote:
>
> On Mon, Mar 22, 2021 at 12:17:37PM +0900, Masahiko Sawada wrote:
> > I've updated the patch. I saved the index names at the beginning of
> > heap_vacuum_rel() for autovacuum logging, and add indstats and
> > nindexes to LVRelStats. Some functions still have nindexes as a
> > function argument but it seems to make sense since it corresponds the
> > list of index relations (*Irel). Please review the patch.
>
> Going back to that, the structure of the static APIs in this file make
> the whole logic a bit hard to follow, but the whole set of changes you
> have done here makes sense. It took me a moment to recall and
> understand why it is safe to free *stats at the end of
> vacuum_one_index() and if the index stats array actually pointed to
> the DSM segment correctly within the shared stats.
>
> I think that there is more consolidation possible within LVRelStats,
> but let's leave that for another day, if there is any need for such a
> move.
While studying your patch (v5-index_stat_log.patch) I found we can
polish the parallel vacuum code in some places. I'll try it another
day.
>
> To keep it short. Sold.
Thank you!
Regards,
--
Masahiko Sawada
EDB: https://www.enterprisedb.com/
