| From: | Chao Li <li(dot)evan(dot)chao(at)gmail(dot)com> |
|---|---|
| To: | cca5507 <cca5507(at)qq(dot)com> |
| Cc: | pgsql-hackers <pgsql-hackers(at)lists(dot)postgresql(dot)org> |
| Subject: | Re: Avoid copying MinimalTuple twice in ExecHashSkewTableInsert() |
| Date: | 2026-11-20 08:21:05 |
| Message-ID: | EC4AB28B-4079-4549-8EFF-88BE2719AA11@gmail.com |
| Views: | Whole Thread | Raw Message | Download mbox | Resend email |
| Thread: | |
| Lists: | pgsql-hackers |
> On Nov 18, 2026, at 22:56, cca5507 <cca5507(at)qq(dot)com> wrote:
>
> Hi,
>
> We currently copy MinimalTuple twice in ExecHashSkewTableInsert() if the input
> slot is not a MinimalTupleTableSlot. We can avoid this by using
> ExecCopySlotMinimalTupleExtra().
>
> Patch attached.
>
> --
> Regards,
> ChangAo Chen
> <v1-0001-Avoid-copying-MinimalTuple-twice-in-ExecHashSkewT.patch>
The current code does:
* ExecFetchSlotMinimalTuple(slot, &shouldFree)
* if slot->tts_ops->get_minimal_tuple is not NULL, slot->tts_ops->get_minimal_tuple(slot);
* else slot->tts_ops->copy_minimal_tuple(slot, 0);
* hashTuple = (HashJoinTuple) MemoryContextAlloc()
* memcpy(HJTUPLE_MINTUPLE(hashTuple), tuple, tuple->t_len);
* if (shouldFree) heap_free_minimal_tuple(tuple);
The patch does:
* tuple = ExecCopySlotMinimalTupleExtra(slot, HJTUPLE_OVERHEAD);
* slot->tts_ops->copy_minimal_tuple(slot, extra);
* hashTuple = (HashJoinTuple) ((char *) tuple - HJTUPLE_OVERHEAD);
When slot->tts_ops->get_minimal_tuple is not NULL, the old and new code do almost the same amount of work. But when slot->tts_ops->get_minimal_tuple is NULL, this patch saves an allocation, a memcpy(), and a pfree().
So, this patch can improve performance a little bit. Plus, it simplifies the code, so I think it's an improvement.
In summary, v1 looks good to me. I also tried to build and run tests, everything passed.
Best regards,
--
Chao Li (Evan)
HighGo Software Co., Ltd.
https://www.highgo.com/
| From | Date | Subject | |
|---|---|---|---|
| Next Message | Kyotaro Horiguchi | 2026-11-20 08:22:10 | Re: [PATCH v1] Preserve implicit composite casts in ROW deparsing |
| Previous Message | Xuneng Zhou | 2026-11-20 08:19:27 | Re: [BUG] Take a long time to reach consistent after pg_rewind |