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: 2023-12-07T03:27:00Z
Lists: pgsql-hackers 
On Mon, Nov 27, 2023 at 1:45 PM Masahiko Sawada <sawada.mshk@gmail.com> wrote:
>
> On Sat, Oct 28, 2023 at 5:56 PM John Naylor <johncnaylorls@gmail.com> wrote:

> bool
> RT_SET(RT_RADIX_TREE *tree, uint64 key, RT_VALUE_TYPE *value_p);
> or for variable-length value support,
> RT_SET(RT_RADIX_TREE *tree, uint64 key, RT_VALUE_TYPE *value_p, size_t sz);
>
> If an entry already exists, update its value to 'value_p' and return
> true. Otherwise set the value and return false.

> RT_VALUE_TYPE
> RT_INSERT(RT_RADIX_TREE *tree, uint64 key, size_t sz, bool *found);
>
> If the entry already exists, replace the value with a new empty value
> with sz bytes and set "found" to true. Otherwise, insert an empty
> value, return its pointer, and set "found" to false.
>
> We probably will find a better name but I use RT_INSERT() for
> discussion. RT_INSERT() returns an empty slot regardless of existing
> values. It can be used to insert a new value or to replace the value
> with a larger value.

Looking at TidStoreSetBlockOffsets again (in particular how it works
with RT_GET), and thinking about issues we've discussed, I think
RT_SET is sufficient for vacuum. Here's how it could work:

TidStoreSetBlockOffsets could have a stack variable that's "almost
always" large enough. When not, it can allocate in its own context. It
sets the necessary bits there. Then, it passes the pointer to RT_SET
with the number of bytes to copy. That seems very simple.

At some future time, we can add a new function with the complex
business about getting the current value to modify it, with the
re-alloc'ing that it might require.

In other words, from both an API perspective and a performance
perspective, it makes sense for tid store to have a simple "set"
interface for vacuum that can be optimized for its characteristics
(insert only, ordered offsets). And also a more complex one for bitmap
scan (setting/unsetting bits of existing values, in any order). They
can share the same iteration interface, key types, and value types.

What do you think, Masahiko?

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