pgsql: Use bump memory context for tuplesorts
Use bump memory context for tuplesorts
29f6a959c added a bump allocator type for efficient compact allocations.
Here we make use of this for non-bounded tuplesorts to store tuples.
This is very space efficient when storing narrow tuples due to bump.c
not having chunk headers. This means we can fit more tuples in work_mem
before spilling to disk, or perform an in-memory sort touching fewer
cacheline.
Author: David Rowley
Reviewed-by: Nathan Bossart
Reviewed-by: Matthias van de Meent
Reviewed-by: Tomas Vondra
Reviewed-by: John Naylor
Discussion: /messages/by-id/CAApHDvqGSpCU95TmM=Bp=6xjL_nLys4zdZOpfNyWBk97Xrdj2w@mail.gmail.com
Branch
------
master
Details
-------
https://git.postgresql.org/pg/commitdiff/6ed83d5fa55cf6e6c9d1be34ec10730c48eba763
Modified Files
--------------
src/backend/utils/sort/tuplesort.c | 52 +++++++++++++++++-------------
src/backend/utils/sort/tuplesortvariants.c | 38 +++++++++++++++++++---
src/include/utils/tuplesort.h | 21 +++++++++---
3 files changed, 78 insertions(+), 33 deletions(-)
On Sun, Apr 7, 2024 at 8:32 AM David Rowley <drowley@postgresql.org> wrote:
Use bump memory context for tuplesorts
29f6a959c added a bump allocator type for efficient compact allocations.
Here we make use of this for non-bounded tuplesorts to store tuples.
This is very space efficient when storing narrow tuples due to bump.c
not having chunk headers. This means we can fit more tuples in work_mem
before spilling to disk, or perform an in-memory sort touching fewer
cacheline.
Looks like this assert is tripping on grison [1]https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=grison&dt=2024-04-07%2014%3A10%3A09.
running bootstrap script ... TRAP: failed Assert("total_allocated ==
context->mem_allocated"), File: "bump.c", Line: 808, PID: 30248
[1]: https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=grison&dt=2024-04-07%2014%3A10%3A09
Melanie Plageman <melanieplageman@gmail.com> writes:
Looks like this assert is tripping on grison [1].
running bootstrap script ... TRAP: failed Assert("total_allocated ==
context->mem_allocated"), File: "bump.c", Line: 808, PID: 30248
The same on mamba. However, I failed to duplicate it in an x86
(32-bit) VM. I think the critical factor on the failing machines
is that MAXALIGN is 8 which is more than their sizeof(pointer).
regards, tom lane
Hi,
On 2024-04-07 13:52:28 -0400, Tom Lane wrote:
Melanie Plageman <melanieplageman@gmail.com> writes:
Looks like this assert is tripping on grison [1].
running bootstrap script ... TRAP: failed Assert("total_allocated ==
context->mem_allocated"), File: "bump.c", Line: 808, PID: 30248The same on mamba. However, I failed to duplicate it in an x86
(32-bit) VM. I think the critical factor on the failing machines
is that MAXALIGN is 8 which is more than their sizeof(pointer).
For posterity: I think the above is correct, and that I found the concrete
reason this causes issues on the thread related to the patch. I posted there
because I figured it out after looking at Tomas' gdb output:
/messages/by-id/20240407210924.fvwfwbzms3bvtfcd@awork3.anarazel.de
- Andres