From: | Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> |
---|---|
To: | Dmitry Ivanov <d(dot)ivanov(at)postgrespro(dot)ru> |
Cc: | pgsql-hackers <pgsql-hackers(at)postgresql(dot)org> |
Subject: | Re: Possible problem in Custom Scan API |
Date: | 2017-04-17 19:33:35 |
Message-ID: | 29429.1492457615@sss.pgh.pa.us |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-hackers |
Dmitry Ivanov <d(dot)ivanov(at)postgrespro(dot)ru> writes:
> Tom Lane wrote:
>> I'm coming around to the idea that it'd be better to disable physical
>> tlists for custom scans.
>> However, I'm hesitant to make such a change in the back branches; if
>> there's anyone using custom scans who's negatively affected, they'd be
>> rightfully unhappy. Are you satisfied if we change this in HEAD?
After thinking about this over the weekend, I've come to the conclusion
that back-patching is probably the right thing anyway. The upside of the
use_physical_tlist optimization is pretty marginal even when it applies,
while the downsides of applying it inappropriately can be very large,
as we've discussed.
> I've been thinking about this all along, and it seems that this is a decent
> decision. However, I've made a tiny bugfix patch which allows CustomScans
> to notify the core code that they can handle physical targetlists.
I'm unimpressed by this part --- we couldn't back-patch such a change, and
I think it's unnecessary anyway in 9.6+ because the scan provider could
generate a nondefault pathtarget if it wants this to happen.
regards, tom lane
From | Date | Subject | |
---|---|---|---|
Next Message | Stephen Frost | 2017-04-17 19:41:25 | Re: pg_dump emits ALTER TABLE ONLY partitioned_table |
Previous Message | Erik Rijkers | 2017-04-17 19:16:54 | Re: Logical replication - TRAP: FailedAssertion in pgstat.c |