Horton

Re: Reducing the chunk header sizes on all memory context types

David Rowley <dgrowleyml@gmail.com>

View thread View on postgr.es
From: David Rowley <dgrowleyml@gmail.com>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: Andres Freund <andres@anarazel.de>, Julien Rouhaud <rjuju123@gmail.com>, Robert Haas <robertmhaas@gmail.com>, Yura Sokolov <y.sokolov@postgrespro.ru>, PostgreSQL Developers <pgsql-hackers@lists.postgresql.org>
Date: 2022-10-10T22:30:51Z
Lists: pgsql-hackers 
On Tue, 11 Oct 2022 at 10:07, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Yeah, slab.c hasn't any distinction between large and small chunks,
> so we have to just pick one policy or the other.  I'd hoped to get
> away with the more robust runtime test on the basis that slab allocation
> is not used so much that this'd result in any noticeable performance
> change.  SlabRealloc, at least, is not used *at all* per the code
> coverage tests, and if we're there at all we should be highly suspicious
> that something is wrong.  However, I could be wrong about SlabFree,
> and if you're going to hold my feet to the fire then I'll change it
> rather than try to produce some performance evidence.

The main reason I brought it up was that only yesterday I was looking
into fixing the slowness of the Slab allocator. It's currently quite
far behind the performance of both generation.c and aset.c and it
would be very nice to bring it up to at least be on-par with those.
Ideally there would be some performance advantages of the fixed-sized
chunks. I'd just rather not have any additional things go in to make
that goal harder to reach.

The proposed patches in [1] do aim to make additional usages of the
slab allocator, and I have a feeling that we'll want to fix the
performance of slab.c before those. Perhaps the Asserts are a better
option if we're to get the proposed radix tree implementation.

David

[1] https://postgr.es/m/CAD21AoD3w76wERs_Lq7_uA6+gTaoOERPji+Yz8Ac6aui4JwvTg@mail.gmail.com

Commits

  1. Harden memory context allocators against bogus chunk pointers.

  2. Improve our ability to detect bogus pointers passed to pfree et al.

  3. Remove MemoryContextContains().

  4. Remove uses of MemoryContextContains in nodeAgg.c and nodeWindowAgg.c.

  5. Temporarily make MemoryContextContains return false

  6. Make MemoryContextContains work correctly again

  7. Make more effort to put a sentinel at the end of allocated memory

  8. Fix some possibly latent bugs in slab.c

  9. Various cleanups of the new memory context header code

  10. Revert "Add missing padding from MemoryChunk struct"

  11. Use MAXALIGN() in calculations using sizeof(SlabBlock)

  12. Add missing padding from MemoryChunk struct

  13. Improve performance of and reduce overheads of memory management