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: 2024-01-29T11:48:27Z
Lists: pgsql-hackers 
On Mon, Jan 29, 2024 at 2:29 PM Masahiko Sawada <sawada.mshk@gmail.com> wrote:

> > > +/*
> > > + * Calculate the slab blocksize so that we can allocate at least 32 chunks
> > > + * from the block.
> > > + */
> > > +#define RT_SLAB_BLOCK_SIZE(size) \
> > > + Max((SLAB_DEFAULT_BLOCK_SIZE / (size)) * (size), (size) * 32)
> > >
> > > The first parameter seems to be trying to make the block size exact,
> > > but that's not right, because of the chunk header, and maybe
> > > alignment. If the default block size is big enough to waste only a
> > > tiny amount of space, let's just use that as-is.

> If we use SLAB_DEFAULT_BLOCK_SIZE (8kB) for each node class, we waste
> [snip]
> We might want to calculate a better slab block size for node256 at least.

I meant the macro could probably be

Max(SLAB_DEFAULT_BLOCK_SIZE, (size) * N)

(Right now N=32). I also realize I didn't answer your question earlier
about block sizes being powers of two. I was talking about PG in
general -- I was thinking all block sizes were powers of two. If
that's true, I'm not sure if it's because programmers find the macro
calculations easy to reason about, or if there was an implementation
reason for it (e.g. libc behavior). 32*2088 bytes is about 65kB, or
just above a power of two, so if we did  round that up it would be
128kB.

> > > + * TODO: The caller must be certain that no other backend will attempt to
> > > + * access the TidStore before calling this function. Other backend must
> > > + * explicitly call TidStoreDetach() to free up backend-local memory associated
> > > + * with the TidStore. The backend that calls TidStoreDestroy() must not call
> > > + * TidStoreDetach().
> > >
> > > Do we need to do anything now?
> >
> > No, will remove it.
> >
>
> I misunderstood something. I think the above statement is still true
> but we don't need to do anything at this stage. It's a typical usage
> that the leader destroys the shared data after confirming all workers
> are detached. It's not a TODO but probably a NOTE.

Okay.

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