Horton

Re: BitmapHeapScan streaming read user and prelim refactoring

Tomas Vondra <tomas@vondra.me>

View thread View on postgr.es
From: Tomas Vondra <tomas@vondra.me>
To: Melanie Plageman <melanieplageman@gmail.com>
Cc: Nazir Bilal Yavuz <byavuz81@gmail.com>, Dilip Kumar <dilipbalaut@gmail.com>, Heikki Linnakangas <hlinnaka@iki.fi>, Thomas Munro <thomas.munro@gmail.com>, Andres Freund <andres@anarazel.de>, Pg Hackers <pgsql-hackers@postgresql.org>
Date: 2025-02-13T12:08:28Z
Lists: pgsql-hackers 

On 2/13/25 01:40, Melanie Plageman wrote:
> On Sun, Feb 9, 2025 at 9:27 AM Tomas Vondra <tomas@vondra.me> wrote:
>>
>> For the nvme RAID (device: raid-nvme), it's looks almost exactly the
>> same, except that with parallel query (page 27) there's a clear area of
>> regression with eic=1 (look for "column" of red cells). That's a bit
>> unfortunate, because eic=1 is the default value.
> 
> So, I feel pretty confident after even more analysis today with Thomas
> Munro that all of the parallel cases with effective_io_concurrency ==
> 1 are because master cheats effective_io_concurrency. Therefore, I'm
> not too worried about those for now. We probably will have to increase
> effective_io_concurrency before merging this patch, though.
> 
> Thomas mentioned this to me off-list, and I think he's right. We
> likely need to rethink the way parallel bitmap heap scan workers get
> block assignments for reading and prefetching to make it more similar
> to parallel sequential scan. The workers should probably get
> assignments of a range of blocks. On master, each worker does end up
> issuing reads/fadvises for a bunch of blocks in a row -- even though
> that isn't the intent of the parallel bitmap table scan
> implementation. We are losing some of that with the patch -- but only
> because it is behaving as you would expect given the implementation
> and design. I don't consider that a blocker, though.
> 

Agreed. If this is due to master doing something wrong (and either
prefetching more, or failing to prefetch), that's not the fault of this
patch. People might perceive this as a regression, but increasing the
default eic value would fix that.

BTW I recall we ran into issues with prefetching and parallel workers
about a year ago [1]. Is this the same issue, or something new?

[1]
https://www.postgresql.org/message-id/20240315211449.en2jcmdqxv5o6tlz%40alap3.anarazel.de

>> For the SATA SSD RAID (device: raid-sata), it's similar - a couple
>> regressions for eic=0, just like for NVMe. But then there's also a
>> couple regressions for higher eic values, usually for queries with
>> selective conditions.
> 
> I started trying to reproduce the regressions you saw on your SATA
> devices for higher eic values. I was focusing just on serial bitmap
> heap scan -- since that doesn't have the problems mentioned above. I
> don't have a SATA drive, but I used dmsetup to add a 10ms delay to my
> nvme drive. Unfortunately, that did the opposite of reproduce the
> issue. With dm delay 10ms, the patch is 5 times faster than master.
> 

I'm not quite sure at which point does dm-setup add the delay. Does it
affect the fadvise call too, or just the I/O if the page is not already
in page cache?

regards

-- 
Tomas Vondra

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.