Horton

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

Masahiko Sawada <sawada.mshk@gmail.com>

View thread View on postgr.es
From: Masahiko Sawada <sawada.mshk@gmail.com>
To: Andres Freund <andres@anarazel.de>
Cc: PostgreSQL-development <pgsql-hackers@postgresql.org>
Date: 2021-07-10T12:11:20Z
Lists: pgsql-hackers 
On Sat, Jul 10, 2021 at 2:17 AM Andres Freund <andres@anarazel.de> wrote:
>
> Hi,
>
> On 2021-07-07 20:46:38 +0900, Masahiko Sawada wrote:
> > Currently, the TIDs of dead tuples are stored in an array that is
> > collectively allocated at the start of lazy vacuum and TID lookup uses
> > bsearch(). There are the following challenges and limitations:
>
> > So I prototyped a new data structure dedicated to storing dead tuples
> > during lazy vacuum while borrowing the idea from Roaring Bitmap[2].
> > The authors provide an implementation of Roaring Bitmap[3]  (Apache
> > 2.0 license). But I've implemented this idea from scratch because we
> > need to integrate it with Dynamic Shared Memory/Area to support
> > parallel vacuum and need to support ItemPointerData, 6-bytes integer
> > in total, whereas the implementation supports only 4-bytes integers.
> > Also, when it comes to vacuum, we neither need to compute the
> > intersection, the union, nor the difference between sets, but need
> > only an existence check.
> >
> > The data structure is somewhat similar to TIDBitmap. It consists of
> > the hash table and the container area; the hash table has entries per
> > block and each block entry allocates its memory space, called a
> > container, in the container area to store its offset numbers. The
> > container area is actually an array of bytes and can be enlarged as
> > needed. In the container area, the data representation of offset
> > numbers varies depending on their cardinality. It has three container
> > types: array, bitmap, and run.
>
> How are you thinking of implementing iteration efficiently for rtbm? The
> second heap pass needs that obviously... I think the only option would
> be to qsort the whole thing?

Yes, I'm thinking that the iteration of rtbm is somewhat similar to
tbm. That is, we iterate and collect hash table entries and do qsort
hash entries by the block number. Then fetch the entry along with its
container one by one in order of the block number.

Regards,

-- 
Masahiko Sawada
EDB:  https://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