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: John Naylor <johncnaylorls@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: 2024-03-21T09:02:47Z
Lists: pgsql-hackers

Attachments

On Thu, Mar 21, 2024 at 4:35 PM John Naylor <johncnaylorls@gmail.com> wrote:
>
> On Thu, Mar 21, 2024 at 1:11 PM Masahiko Sawada <sawada.mshk@gmail.com> wrote:
>
> > Or we can have a new function for dsa.c to set the initial and max
> > segment size (or either one) to the existing DSA area so that
> > TidStoreCreate() can specify them at creation.
>
> I didn't like this very much, because it's splitting an operation
> across an API boundary. The caller already has all the information it
> needs when it creates the DSA. Straw man proposal: it could do the
> same for local memory, then they'd be more similar. But if we made
> local contexts the responsibility of the caller, that would cause
> duplication between creating and resetting.

Fair point.

>
> > In shared TidStore
> > cases, since all memory required by shared radix tree is allocated in
> > the passed-in DSA area and the memory usage is the total segment size
> > allocated in the DSA area
>
> ...plus apparently some overhead, I just found out today, but that's
> beside the point.
>
> On Thu, Mar 21, 2024 at 2:02 PM Masahiko Sawada <sawada.mshk@gmail.com> wrote:
> >
> > Yet another idea is that TidStore creates its own DSA area in
> > TidStoreCreate(). That is, In TidStoreCreate() we create a DSA area
> > (using dsa_create()) and pass it to RT_CREATE(). Also, we need a new
> > API to get the DSA area. The caller (e.g. parallel vacuum) gets the
> > dsa_handle of the DSA and stores it in the shared memory (e.g. in
> > PVShared). TidStoreAttach() will take two arguments: dsa_handle for
> > the DSA area and dsa_pointer for the shared radix tree. This idea
> > still requires controlling min/max segment sizes since dsa_create()
> > uses the 1MB as the initial segment size. But the TidStoreCreate()
> > would be more user friendly.
>
> This seems like an overall simplification, aside from future size
> configuration, so +1 to continue looking into this. If we go this
> route, I'd like to avoid a boolean parameter and cleanly separate
> TidStoreCreateLocal() and TidStoreCreateShared(). Every operation
> after that can introspect, but it's a bit awkward to force these cases
> into the same function. It always was a little bit, but this change
> makes it more so.

I've looked into this idea further. Overall, it looks clean and I
don't see any problem so far in terms of integration with lazy vacuum.
I've attached three patches for discussion and tests.

- 0001 patch makes lazy vacuum use of tidstore.
- 0002 patch makes DSA init/max segment size configurable (borrowed
from another thread).
- 0003 patch makes TidStore create its own DSA area with init/max DSA
segment adjustment (PoC patch).

One thing unclear to me is that this idea will be usable even when we
want to use the tidstore for parallel bitmap scan. Currently, we
create a shared tidbitmap on a DSA area in ParallelExecutorInfo. This
DSA area is used not only for tidbitmap but also for parallel hash
etc. If the tidstore created its own DSA area, parallel bitmap scan
would have to use the tidstore's DSA in addition to the DSA area in
ParallelExecutorInfo. I'm not sure if there are some differences
between these usages in terms of resource manager etc. It seems no
problem but I might be missing something.

Regards,
-- 
Masahiko Sawada
Amazon Web Services: https://aws.amazon.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