Horton

Re: Problems with plan estimates in postgres_fdw

Etsuro Fujita <fujita.etsuro@lab.ntt.co.jp>

View thread View on postgr.es
From: Etsuro Fujita <fujita.etsuro@lab.ntt.co.jp>
To: Antonin Houska <ah@cybertec.at>
Cc: Tom Lane <tgl@sss.pgh.pa.us>, Andrew Gierth <andrew@tao11.riddles.org.uk>, pgsql-hackers@postgresql.org
Date: 2019-03-05T08:11:18Z
Lists: pgsql-hackers 
(2019/03/02 1:14), Antonin Houska wrote:
> Etsuro Fujita<fujita.etsuro@lab.ntt.co.jp>  wrote:
>> (2019/03/01 20:00), Antonin Houska wrote:

>> Sorry, my explanation was not enough again, but I showed that query ("SELECT
>> a+b, random() FROM foreign_table GROUP BY a+b ORDER BY a+b;") to explain why
>> the following code bit is needed:
>>
>> +       /*
>> +        * If this includes an UPPERREL_ORDERED step, the given target, which
>> +        * would be the final target to be applied to the resulting path,
>> might
>> +        * have different expressions from the underlying relation's reltarget
>> +        * (see make_sort_input_target()); adjust tlist eval costs.
>> +        */
>> +       if (fpextra&&   fpextra->target != foreignrel->reltarget)
>> +       {
>> +               QualCost        oldcost = foreignrel->reltarget->cost;
>> +               QualCost        newcost = fpextra->target->cost;
>> +
>> +               startup_cost += newcost.startup - oldcost.startup;
>> +               total_cost += newcost.startup - oldcost.startup;
>> +               total_cost += (newcost.per_tuple - oldcost.per_tuple) * rows;
>> +       }
>
> Maybe I undestand now. Do the expressions (newcost.* - oldcost.*) reflect the
> fact that, for the query
>
> 	SELECT a+b, random() FROM foreign_table GROUP BY a+b ORDER BY a+b;
>
> the UPPERREL_ORDERED stage only needs to evaluate the random() function
> because (a + b) was already evaluated during the UPPERREL_GROUP_AGG stage?

Yeah, I think so.

Best regards,
Etsuro Fujita

Commits

  1. postgres_fdw: Perform the (FINAL, NULL) upperrel operations remotely.

  2. Refactor create_limit_path() to share cost adjustment code with FDWs.

  3. postgres_fdw: Modify regression tests for EPQ-related planning problems.

  4. postgres_fdw: Perform the (ORDERED, NULL) upperrel operations remotely.

  5. Save PathTargets for distinct/ordered relations in root->upper_targets[].

  6. Split create_foreignscan_path() into three functions.

  7. Fix test case for 'outer pathkeys do not match mergeclauses' fix.

  8. postgres_fdw: Avoid 'outer pathkeys do not match mergeclauses' error.