On Fri, Oct 14, 2016 at 11:54 AM, Andres Freund <andres@anarazel.de> wrote:
> I don't think it's a good idea to do this under the content lock in any
> case. But luckily I don't think we have to do so at all.
>
> Due to pagemode - which is used by pretty much everything iterating over
> heaps, and definitely seqscans - the key test already happens without
> the content lock held, in heapgettup_pagemode():
Yes, you are right. Now after this clarification, it's clear that
even we push down the key we are not evaluating it under buffer
content lock.
As of now, I have done my further analysis by keeping in mind only
pushing 'var op const'. Below are my observations.
#1. If we process the qual in executor, all temp memory allocation are
under "per_tuple_context" which get reset after each tuple process.
But, on other hand if we do that in heap, then that will be under
"per_query_context". This restrict us to pushdown any qual which need
to do memory allocations like "toastable" field.
Is there any other way to handle this ?
#2. Currently quals are ordered based on cost (refer
order_qual_clauses), But once we pushdown some of the quals, then
those quals will always be executed first. Can this create problem ?
consider below example..
create or replace function f1(anyelement) returns bool as
$$
begin
raise notice '%', $1;
return true;
end;
$$ LANGUAGE plpgsql cost 0.01;
select * from t3 where a>1 and f1(b);
In this case "f1(b)" will always be executed as first qual (cost is
set very low by user) hence it will raise notice for all the tuple.
But if we pushdown "a>1" qual to heap then only qualified tuple will
be passed to "f1(b)".
Is it behaviour change ?
I know that, it can also impact the performance, because when user has
given some function with very low cost then that should be executed
first as it may discard most of the tuple.
One solution to this can be..
1. Only pushdown if either all quals are of same cost.
2. Pushdown all quals until we find first non pushable qual (this way
we can maintain the same qual execution order what is there in
existing system).
--
Regards,
Dilip Kumar
EnterpriseDB: http://www.enterprisedb.com
