Horton

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

John Naylor <johncnaylorls@gmail.com>

View thread View on postgr.es
From: John Naylor <johncnaylorls@gmail.com>
To: Masahiko Sawada <sawada.mshk@gmail.com>
Cc: PostgreSQL-development <pgsql-hackers@postgresql.org>, Nathan Bossart <nathandbossart@gmail.com>, Andres Freund <andres@anarazel.de>, Matthias van de Meent <boekewurm+postgres@gmail.com>, Yura Sokolov <y.sokolov@postgrespro.ru>
Date: 2024-03-19T09:40:06Z
Lists: pgsql-hackers 
On Tue, Mar 19, 2024 at 10:24 AM Masahiko Sawada <sawada.mshk@gmail.com> wrote:
>
> On Tue, Mar 19, 2024 at 8:35 AM John Naylor <johncnaylorls@gmail.com> wrote:
> >
> > On Mon, Mar 18, 2024 at 11:12 AM Masahiko Sawada <sawada.mshk@gmail.com> wrote:
> > >
> > > On Sun, Mar 17, 2024 at 11:46 AM John Naylor <johncnaylorls@gmail.com> wrote:

> > It might also be worth reducing the number of blocks in the random
> > test -- multiple runs will have different offsets anyway.
>
> Yes. If we reduce the number of blocks from 1000 to 100, the
> regression test took on my environment:
>
> 1000 blocks : 516 ms
> 100 blocks   : 228 ms

Sounds good.

> Removed some unnecessary variables in 0002 patch.

Looks good.

> So the MaxBlocktableEntrySize calculation would be as follows?
>
> #define MaxBlocktableEntrySize \
>     offsetof(BlocktableEntry, words) + \
>         (sizeof(bitmapword) * \
>         WORDS_PER_PAGE(Min(MaxOffsetNumber, \
>                            BITS_PER_BITMAPWORD * PG_INT8_MAX - 1))))
>
> I've made this change in the 0003 patch.

This is okay, but one side effect is that we have both an assert and
an elog, for different limits. I think we'll need a separate #define
to help. But for now, I don't want to hold up tidstore further with
this because I believe almost everything else in v74 is in pretty good
shape. I'll save this for later as a part of the optimization I
proposed.

Remaining things I noticed:

+#define RT_PREFIX local_rt
+#define RT_PREFIX shared_rt

Prefixes for simplehash, for example, don't have "sh" -- maybe "local/shared_ts"

+ /* MemoryContext where the radix tree uses */

s/where/that/

+/*
+ * Lock support functions.
+ *
+ * We can use the radix tree's lock for shared TidStore as the data we
+ * need to protect is only the shared radix tree.
+ */
+void
+TidStoreLockExclusive(TidStore *ts)

Talking about multiple things, so maybe a blank line after the comment.

With those, I think you can go ahead and squash all the tidstore
patches except for 0003 and commit it.

> While reviewing the vacuum patch, I realized that we always pass
> LWTRANCHE_SHARED_TIDSTORE to RT_CREATE(), and the wait event related
> to the tidstore is therefore always the same. I think it would be
> better to make the caller of TidStoreCreate() specify the tranch_id
> and pass it to RT_CREATE(). That way, the caller can specify their own
> wait event for tidstore. The 0008 patch tried this idea. dshash.c does
> the same idea.

Sounds reasonable. I'll just note that src/include/storage/lwlock.h
still has an entry for LWTRANCHE_SHARED_TIDSTORE.

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