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: 2023-02-27T17:07:23Z
Lists: pgsql-hackers

Attachments

On Fri, Feb 24, 2023 at 3:41 PM Masahiko Sawada <sawada.mshk@gmail.com>
wrote:
>
> In v29 vacuum took twice as long (286 s vs. 573 s)?

Not sure what happened there, and clearly I was looking at the wrong number
:/
I scripted the test for reproducibility and ran it three times. Also
included some variations (attached):

UUID times look comparable here, so no speedup or regression:

master:
system usage: CPU: user: 216.05 s, system: 35.81 s, elapsed: 634.22 s
system usage: CPU: user: 173.71 s, system: 31.24 s, elapsed: 599.04 s
system usage: CPU: user: 171.16 s, system: 30.21 s, elapsed: 583.21 s

v29:
system usage: CPU: user:  93.47 s, system: 40.92 s, elapsed: 594.10 s
system usage: CPU: user:  99.58 s, system: 44.73 s, elapsed: 606.80 s
system usage: CPU: user:  96.29 s, system: 42.74 s, elapsed: 600.10 s

Then, I tried sequential integers, which is a much more favorable access
pattern in general, and the new tid storage shows substantial improvement:

master:
system usage: CPU: user: 100.39 s, system: 7.79 s, elapsed: 121.57 s
system usage: CPU: user: 104.90 s, system: 8.81 s, elapsed: 124.24 s
system usage: CPU: user:  95.04 s, system: 7.55 s, elapsed: 116.44 s

v29:
system usage: CPU: user:  24.57 s, system: 8.53 s, elapsed: 61.07 s
system usage: CPU: user:  23.18 s, system: 8.25 s, elapsed: 58.99 s
system usage: CPU: user:  23.20 s, system: 8.98 s, elapsed: 66.86 s

That's fast enough that I thought an improvement would show up even with
standard WAL logging (no separate attachment, since it's a trivial change).
Seems a bit faster:

master:
system usage: CPU: user: 152.27 s, system: 11.76 s, elapsed: 216.86 s
system usage: CPU: user: 137.25 s, system: 11.07 s, elapsed: 213.62 s
system usage: CPU: user: 149.48 s, system: 12.15 s, elapsed: 220.96 s

v29:
system usage: CPU: user: 40.88 s, system: 15.99 s, elapsed: 170.98 s
system usage: CPU: user: 41.33 s, system: 15.45 s, elapsed: 166.75 s
system usage: CPU: user: 41.51 s, system: 18.20 s, elapsed: 203.94 s

There is more we could test here, but I feel better about these numbers.

In the next few days, I'll resume style review and list the remaining
issues we need to address.

--
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