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: 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: 2022-11-16T07:39:00Z
Lists: pgsql-hackers

Attachments

On Wed, Nov 16, 2022 at 12:33 PM Masahiko Sawada <sawada.mshk@gmail.com>
wrote:
>
> On Wed, Nov 16, 2022 at 1:46 PM John Naylor
> <john.naylor@enterprisedb.com> wrote:
> >
> >
> > On Tue, Nov 15, 2022 at 11:59 AM Masahiko Sawada <sawada.mshk@gmail.com>
wrote:
> > > Thanks! Please let me know if there is something I can help with.
> >
> > I didn't get very far because the tests fail on 0004 in rt_verify_node:
> >
> > TRAP: failed Assert("n4->chunks[i - 1] < n4->chunks[i]"), File:
"../src/backend/lib/radixtree.c", Line: 2186, PID: 18242
>
> Which tests do you use to get this assertion failure? I've confirmed
> there is a bug in 0005 patch but without it, "make check-world"
> passed.

Hmm, I started over and rebuilt and it didn't reproduce. Not sure what
happened, sorry for the noise.

I'm attaching a test I wrote to stress test branch prediction in search,
and while trying it out I found two possible issues.

It's based on the random int load test, but tests search speed. Run like
this:

select * from bench_search_random_nodes(10 * 1000 * 1000)

It also takes some care to include all the different node kinds,
restricting the possible keys by AND-ing with a filter. Here's a simple
demo:

filter = ((uint64)1<<40)-1;
LOG:  num_keys = 9999967, height = 4, n4 = 17513814, n32 = 6320, n128 =
62663, n256 = 3130

Just using random integers leads to >99% using the smallest node. I wanted
to get close to having the same number of each, but that's difficult while
still using random inputs. I ended up using

filter = (((uint64) 0x7F<<32) | (0x07<<24) | (0xFF<<16) | 0xFF)

which gives

LOG:  num_keys = 9291812, height = 4, n4 = 262144, n32 = 79603, n128 =
182670, n256 = 1024

Which seems okay for the task. One puzzling thing I found while trying
various filters is that sometimes the reported tree height would change.
For example:

filter = (((uint64) 1<<32) | (0xFF<<24));
LOG:  num_keys = 9999944, height = 7, n4 = 47515559, n32 = 6209, n128 =
62632, n256 = 3161

1) Any idea why the tree height would be reported as 7 here? I didn't
expect that.

2) It seems that 0004 actually causes a significant slowdown in this test
(as in the attached, using the second filter above and with turboboost
disabled):

v9 0003: 2062 2051 2050
v9 0004: 2346 2316 2321

That means my idea for the pointer struct might have some problems, at
least as currently implemented. Maybe in the course of separating out and
polishing that piece, an inefficiency will fall out. Or, it might be
another reason to template local and shared separately. Not sure yet. I
also haven't tried to adjust this test for the shared memory case.

--
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