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: Julien Rouhaud <rjuju123@gmail.com>, Andres Freund <andres@anarazel.de>, Robert Haas <robertmhaas@gmail.com>, Yura Sokolov <y.sokolov@postgrespro.ru>, PostgreSQL Developers <pgsql-hackers@lists.postgresql.org>
Date: 2022-09-08T00:13:11Z
Lists: pgsql-hackers

Attachments

On Thu, 8 Sept 2022 at 09:32, David Rowley <dgrowleyml@gmail.com> wrote:
>
> On Thu, 8 Sept 2022 at 03:08, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> > Step 4 is annoyingly expensive, but perhaps not too awful given
> > the way we step up alloc block sizes.  We should make sure that
> > any context we want to use MemoryContextContains with is allowed
> > to let its blocks grow large, so that there can't be too many
> > of them.
>
> I'll go code up your idea and see if doing that triggers any other ideas.

I've attached a very much draft grade patch for this.  I have a couple
of thoughts:

1. I should remove all the Assert(MemoryContextContains(context,
ret)); I littered around mcxt.c. This function is not as cheap as it
once was and I'm expecting that Assert to be a bit too expensive now.
2. I changed the header comment in MemoryContextContains again, but I
removed the part about false positives since I don't believe that is
possible now.  What I do think is just as possible as it was before is
a segfault.  We're still accessing the 8 bytes prior to the given
pointer and there's a decent chance that would segfault when working
with a pointer which was returned by malloc.  I imagine I'm not the
only C programmer around that dislikes writing comments along the
lines of "this might segfault, but..."
3. For external chunks, I'd coded MemoryChunk to put a magic number in
the 60 free bits of the hdrmask.  Since we still need to call
MemoryChunkIsExternal on the given pointer, that function will Assert
that the magic number matches if the external chunk bit is set. We
can't expect that magic number check to pass when the external bit
just happens to be on because it's not a MemoryChunk we're looking at.
For now I commented out those Asserts to make the tests pass. Not sure
what's best there, maybe another version of MemoryChunkIsExternal or
export the underlying macro.  I'm currently more focused on what I
wrote in #2.

David

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