Horton

Re: [PoC] Improve dead tuple storage for lazy vacuum

John Naylor <john.naylor@enterprisedb.com>

View thread View on postgr.es
From: John Naylor <john.naylor@enterprisedb.com>
To: Masahiko Sawada <sawada.mshk@gmail.com>
Cc: Andres Freund <andres@anarazel.de>, Matthias van de Meent <boekewurm+postgres@gmail.com>, Yura Sokolov <y.sokolov@postgrespro.ru>, PostgreSQL-development <pgsql-hackers@postgresql.org>
Date: 2022-08-15T13:39:27Z
Lists: pgsql-hackers 
On Mon, Aug 15, 2022 at 12:39 PM Masahiko Sawada <sawada.mshk@gmail.com> wrote:
>
> On Fri, Jul 22, 2022 at 10:43 AM Masahiko Sawada <sawada.mshk@gmail.com> wrote:
> >
> > On Tue, Jul 19, 2022 at 1:30 PM John Naylor
> > <john.naylor@enterprisedb.com> wrote:
> > >
> > >
> > >
> > > On Tue, Jul 19, 2022 at 9:11 AM Masahiko Sawada <sawada.mshk@gmail.com> wrote:
> > >
> > > > I’d like to keep the first version simple. We can improve it and add
> > > > more optimizations later. Using radix tree for vacuum TID storage
> > > > would still be a big win comparing to using a flat array, even without
> > > > all these optimizations. In terms of single-value leaves method, I'm
> > > > also concerned about an extra pointer traversal and extra memory
> > > > allocation. It's most flexible but multi-value leaves method is also
> > > > flexible enough for many use cases. Using the single-value method
> > > > seems to be too much as the first step for me.
> > > >
> > > > Overall, using 64-bit keys and 64-bit values would be a reasonable
> > > > choice for me as the first step . It can cover wider use cases
> > > > including vacuum TID use cases. And possibly it can cover use cases by
> > > > combining a hash table or using tree of tree, for example.
> > >
> > > These two aspects would also bring it closer to Andres' prototype, which 1) makes review easier and 2) easier to preserve optimization work already done, so +1 from me.
> >
> > Thanks.
> >
> > I've updated the patch. It now implements 64-bit keys, 64-bit values,
> > and the multi-value leaves method. I've tried to remove duplicated
> > codes but we might find a better way to do that.
> >
>
> With the recent changes related to simd, I'm going to split the patch
> into at least two parts: introduce other simd optimized functions used
> by the radix tree and the radix tree implementation. Particularly we
> need two functions for radix tree: a function like pg_lfind32 but for
> 8 bits integers and return the index, and a function that returns the
> index of the first element that is >= key.

I recommend looking at

https://www.postgresql.org/message-id/CAFBsxsESLUyJ5spfOSyPrOvKUEYYNqsBosue9SV1j8ecgNXSKA%40mail.gmail.com

since I did the work just now for searching bytes and returning a
bool, buth = and <=. Should be pretty close. Also, i believe if you
left this for last as a possible refactoring, it might save some work.
In any case, I'll take a look at the latest patch next month.

-- 
John Naylor
EDB: http://www.enterprisedb.com

Commits

  1. radixtree: Fix SIGSEGV at update of embeddable value to non-embeddable.

  2. Get rid of anonymous struct

  3. Teach radix tree to embed values at runtime

  4. Teach TID store to skip bitmap for small numbers of offsets

  5. Use bump context for TID bitmaps stored by vacuum

  6. Fix alignment of stack variable

  7. Use TidStore for dead tuple TIDs storage during lazy vacuum.

  8. Rethink create and attach APIs of shared TidStore.

  9. Fix inconsistent function prototypes with function definitions.

  10. Fix a calculation in TidStoreCreate().

  11. Fix potential integer handling issue in radixtree.h.

  12. Add TIDStore, to store sets of TIDs (ItemPointerData) efficiently.

  13. Fix link error for test_radixtree module on Windows

  14. Blind attempt to fix ODR violations

  15. Fix incorrect format specifier for int64

  16. Fix redefinition of typedefs

  17. Add template for adaptive radix tree

  18. Fix signedness error in 9f225e992 for gcc

  19. Introduce helper SIMD functions for small byte arrays

  20. Optimize vacuuming of relations with no indexes.

  21. Add bound check before bsearch() for performance

  22. Allocate consecutive blocks during parallel seqscans