Horton

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

Masahiko Sawada <sawada.mshk@gmail.com>

View thread View on postgr.es
From: Masahiko Sawada <sawada.mshk@gmail.com>
To: John Naylor <john.naylor@enterprisedb.com>
Cc: Nathan Bossart <nathandbossart@gmail.com>, 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: 2023-03-09T06:51:05Z
Lists: pgsql-hackers

Attachments

On Wed, Mar 8, 2023 at 1:40 PM John Naylor <john.naylor@enterprisedb.com> wrote:
>
>

> On Tue, Mar 7, 2023 at 8:25 AM Masahiko Sawada <sawada.mshk@gmail.com> wrote:
>
> > > 1. Make it optional to track chunk memory space by a template parameter. It might be tiny compared to everything else that vacuum does. That would allow other users to avoid that overhead.
> > > 2. When context block usage exceeds the limit (rare), make the additional effort to get the precise usage -- I'm not sure such a top-down facility exists, and I'm not feeling well enough today to study this further.
> >
> > I prefer option (1) as it's straight forward. I mentioned a similar
> > idea before[1]. RT_MEMORY_USAGE() is defined only when the macro is
> > defined. It might be worth checking if there is visible overhead of
> > tracking chunk memory space. IIRC we've not evaluated it yet.
>
> Ok, let's try this -- I can test and profile later this week.

Thanks!

I've attached the new version patches. I merged improvements and fixes
I did in the v29 patch. 0007 through 0010 are updates from v29. The
main change made in v30 is to make the memory measurement and
RT_MEMORY_USAGE() optional, which is done in 0007 patch. The 0008 and
0009 patches are the updates for tidstore and the vacuum integration
patches. Here are results of quick tests (an average of 3 executions):

query: select * from bench_load_random_int(10 * 1000 * 1000)

* w/ RT_MEASURE_MEMORY_USAGE:
 mem_allocated | load_ms
---------------+---------
    1996512000 |    3305
(1 row)

* w/o RT_MEASURE_MEMORY_USAGE:
 mem_allocated | load_ms
---------------+---------
             0 |    3258
(1 row)

It seems to be within a noise level but I agree to make it optional.

Apart from the memory measurement stuff, I've done another todo item
on my list; adding min max classes for node3 and node125. I've done
that in 0010 patch, and here is a quick test result:

query: select * from bench_load_random_int(10 * 1000 * 1000)

* w/ 0000 patch
 mem_allocated | load_ms
---------------+---------
    1268630080 |    3275
(1 row)

* w/o 0000 patch
 mem_allocated | load_ms
---------------+---------
    1996512000 |    3214
(1 row)

That's a good improvement on the memory usage, without a noticeable
performance overhead. FYI CLASS_3_MIN has 1 fanout and is 24 bytes in
size, and CLASS_125_MIN has 61 fanouts and is 768 bytes in size.

Regards,

-- 
Masahiko Sawada
Amazon Web Services: https://aws.amazon.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