Andrew Dunstan <andrew@dunslane.net> writes:
> On 02/02/2014 08:54 PM, Craig Ringer wrote:
>> The attached patch deletes the context after use, bringing performance
>> back into line. It should be backpatched to 9.3.
> Hmm. I guess I was assuming that the tmp_cxt would be cleaned up at the
> end of the function since it's a child of the CurrentMemoryContext.
The executor does MemoryContextReset, not
MemoryContextResetAndDeleteChildren, on the per-tuple context. That means
that child contexts will be reset, not deleted. I seem to recall some
discussions about changing that, or even redefining MemoryContextReset to
automatically delete child contexts; but it would take a fair amount of
research to be sure such a change was safe.
regards, tom lane