Horton

Re: BitmapHeapScan streaming read user and prelim refactoring

Melanie Plageman <melanieplageman@gmail.com>

View thread View on postgr.es
From: Melanie Plageman <melanieplageman@gmail.com>
To: Thomas Munro <thomas.munro@gmail.com>
Cc: Tomas Vondra <tomas@vondra.me>, Nazir Bilal Yavuz <byavuz81@gmail.com>, Dilip Kumar <dilipbalaut@gmail.com>, Heikki Linnakangas <hlinnaka@iki.fi>, Andres Freund <andres@anarazel.de>, Pg Hackers <pgsql-hackers@postgresql.org>
Date: 2025-02-21T22:00:49Z
Lists: pgsql-hackers

Attachments

Thanks for the review!

While editing this, I found a few things I could improve. For one, I
actually wasn't using the parallel state member that I had added as a
parameter to one of the table AM callbacks. So, I was able to remove
that.

On Thu, Feb 20, 2025 at 7:32 PM Thomas Munro <thomas.munro@gmail.com> wrote:
>
> On Fri, Feb 21, 2025 at 11:17 AM Melanie Plageman
> <melanieplageman@gmail.com> wrote:
> > [v31-0001-Delay-extraction-of-TIDBitmap-per-page-offsets.patch]
>
> +/*
> + * The tuple OffsetNumbers extracted from a single page in the bitmap.
> + */
> +typedef OffsetNumber TBMOffsets[TBM_MAX_TUPLES_PER_PAGE];
>
> It's used for a stack object, but it's also used as a degraded pointer here:
>
> +extern void tbm_extract_page_tuple(TBMIterateResult *iteritem,
> +
> TBMOffsets offsets);
>
> Maybe a personal style question but degraded pointers aren't my
> favourite C feature.  A couple of alternatives, I am not sure how good
> they are:  Would this output variable be better wrapped up in a new
> struct for better API clarity?  If so, shouldn't such a struct also
> hold ntuples itself, since that really goes with the offsets it's
> holding?  If so, could iteritem->ntuples then be replaced with a
> boolean iteritem->lossy, since all you really need to be able to check
> is whether it has offsets to give you, instead of using -1 and -2 for
> that?

Yea, I think this is a good point. I made a new struct which has
ntuples and the offsets array as you suggested. I think it's much
nicer than using -2 :)

I do worry a bit about the ntuples value in the new struct being
uninitialized. Before, tbm_iterate() filled in
TBMIterateResult.ntuples, so you knew that once you had a
TBMIterateResult at all, ntuples was valid. Now you make a
TBMPageOffsets and until you call tbm_extract_page_tuple() on it,
TBMPageOffsets->ntuples is garbage. It's probably okay though. I have
the caller initialize TBMPageOffsets to -1 when I think this might be
an issue.

> Or as a variation without a new struct, would it be better to
> take that output array as a plain pointer to OffsetNumber and
> max_offsets, and return ntuples (ie how many were filled in, cf commit
> f6bef362), for the caller to use in a loop or stash somewhere, with an
> additional comment that TBM_MAX_TUPLES_PER_PAGE is guaranteed to be
> enough, for convenience eg for arrays on the stack?

I thought about doing this, but in the end I decided against this
approach. If we wanted to make it easy to palloc arrays of different
sizes and have tbm_extract_page_tuple extract that many tuples into
the array, we'd have to change more of the code anyway because
tbm_extract_page_tuple is what tells us how many tuples there are to
begin with. We could save this somewhere while filling in the
PagetableEntries initially, but that is a bigger change.

And, passing a length would also mean more callers would have to know
about TBM_MAX_TUPLES_PER_PAGE, which I think is already kind of a hack
since it is defined as MaxHeapTuplesPerPage.

- Melanie

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.