Horton

Re: BitmapHeapScan streaming read user and prelim refactoring

Jakub Wartak <jakub.wartak@enterprisedb.com>

View thread View on postgr.es
From: Jakub Wartak <jakub.wartak@enterprisedb.com>
To: Melanie Plageman <melanieplageman@gmail.com>
Cc: Andres Freund <andres@anarazel.de>, Thomas Munro <thomas.munro@gmail.com>, Tomas Vondra <tomas@vondra.me>, Nazir Bilal Yavuz <byavuz81@gmail.com>, Dilip Kumar <dilipbalaut@gmail.com>, Heikki Linnakangas <hlinnaka@iki.fi>, Pg Hackers <pgsql-hackers@postgresql.org>
Date: 2025-03-17T07:43:55Z
Lists: pgsql-hackers 
On Thu, Mar 13, 2025 at 9:34 PM Melanie Plageman
<melanieplageman@gmail.com> wrote:
> On Thu, Mar 13, 2025 at 5:46 AM Jakub Wartak
> <jakub.wartak@enterprisedb.com> wrote:
> >
> > Cool, anything > 1 is just better. Just quick question, so now we have:
> >
> > #define DEFAULT_EFFECTIVE_IO_CONCURRENCY 16
> > #define DEFAULT_MAINTENANCE_IO_CONCURRENCY 10
> >
> > Shouldn't maintenance be now also at the same value (16) instead of
> > 10? Also, fc34b0d9de27 (5 years ago by Thomas) states about m_io_c:
> > "Use the new setting in heapam.c instead of the hard-coded formula
> > effective_io_concurrency + 10 introduced by commit 558a9165e08.", so
> > there was always assumption that m_io_c > e_io_c (?) Now it's kind of
> > inconsistent to see bitmap heap scans pushing more IOPs than
> > recovery(!)/ANALYZE/VACUUM or am I missing something? No pressure to
> > change, just asking what Your thoughts are.
>
> Yea, I'm not really sure what if any relationship the two GUCs should
> have to one another.
> As long as they are in the same ballpark, since
> they are tuning for the same machine, I don't see any reason their
> values need to be adjusted together. However, if it is confusing for
> maintenance_io_concurrency default to be 10 while
> effective_io_concurrency default is 16, I'm happy to change it.

Hi Melanie,

dunno, I've just asked if it isn't suspicious to anyone except me else
that e_io_c > m_io_c rather than e_io_c <= m_io_c. My understanding
was always that to tune max IO queue depth you would do this:
a. up to N backends (up to max_connections; usually much lower) * e_io_c
b. autovacuum_max_workers * m_io_c
c. just one (standby/recovering) * m_io_c

The thing (for me) is: if we are allowing for much higher IOPS "a"
scenario, then why standby cannot use just the same (if not higher)
IOPS for prefetching in "c" scenario. After all, it is a much more
critical and sensitive thing (lag).

> I just don't quite know what I would write in the commit message. "it seemed confusing that they weren't the same?"

"To keep with the historical legacy of maintenance_io_concurrency
being traditionally higher than effective_io_concurrency ..."
OK, so right, nobody else commented on this, so maybe it's just me and
it was just a question after all.

-J.

Commits

  1. Fix bitmapheapscan incorrect recheck of NULL tuples

  2. Increase default maintenance_io_concurrency to 16

  3. Separate TBM[Shared|Private]Iterator and TBMIterateResult

  4. Improve read_stream.c advice for dense streams.

  5. Increase default effective_io_concurrency to 16

  6. Delay extraction of TIDBitmap per page offsets

  7. Add lossy indicator to TBMIterateResult

  8. Move BitmapTableScan per-scan setup into a helper

  9. Add and use BitmapHeapScanDescData struct

  10. Fix bitmap table scan crash on iterator release

  11. Bitmap Table Scans use unified TBMIterator

  12. Add common interface for TBMIterators

  13. Make table_scan_bitmap_next_block() async-friendly

  14. Move EXPLAIN counter increment to heapam_scan_bitmap_next_block

  15. Refactor tidstore.c iterator buffering.

  16. BitmapHeapScan: Remove incorrect assert and reset field

  17. Change BitmapAdjustPrefetchIterator to accept BlockNumber

  18. BitmapHeapScan: Use correct recheck flag for skip_fetch

  19. BitmapHeapScan: Push skip_fetch optimization into table AM

  20. BitmapHeapScan: postpone setting can_skip_fetch

  21. BitmapHeapScan: begin scan after bitmap creation

  22. Fix EXPLAIN Bitmap heap scan to count pages with no visible tuples

  23. Remove redundant snapshot copying from parallel leader to workers

  24. Remove some obsolete smgrcloseall() calls.

  25. Remove the "snapshot too old" feature.

  26. Compute XID horizon for page level index vacuum on primary.