On Wed, 03 Feb 2021 at 02:13, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> japin <japinli@hotmail.com> writes:
>> Here's my analysis:
>> 1) In the explain_ExecutorEnd(), it will create a ExplainState on SQL function
>> memory context, which is a long-lived, cause the memory grow up.
>
> Yeah, agreed. I think people looking at this have assumed that the
> ExecutorEnd hook would automatically be running in the executor's
> per-query context, but that's not so; we haven't yet entered
> standard_ExecutorEnd where the context switch is. The caller's
> context is likely to be much longer-lived than the executor's.
>
> I think we should put the switch one level further out than you have
> it here, just to be sure that InstrEndLoop is covered too (that doesn't
> allocate memory, but auto_explain shouldn't assume that). Otherwise
> seems like a good fix.
>
Thanks for your review and clarification.
--
Regrads,
Japin Li.
ChengDu WenWu Information Technology Co.,Ltd.