Thread: Re: Dramatic slowdown of sql when placed in a function

Re: Dramatic slowdown of sql when placed in a function

From
Jeff Boes
Date:
At some point in time, mike_moran@mac.com (Mike Moran) wrote:

>Hi. I currently have a fairly large query which I have been optimizing
>for later use in a function. There are things in the query which I
>have been keeping constant whilst optimizing, but which are variables
>in the function. When I run this query as sql, with the `variables'
>constant, I get a runtime of about 3 or 4 seconds. However, when I
>place this same sql in an sql function, and then pass my constants
>from before in as arguments, I get a runtime of about 215 seconds.
>
>I am trying to understand how this could be. How does putting the sql
>in a function affect the query optimiser? Would putting it in as a
>plpsql function help? How else can I retain the original speed?

My first guess would be that the indexes being used in the query are
mis-matching on data type compared to your function arguments. For instance,


create function foobar(text) as
'beginselect * from foobar_table where col_a=$1;end' ...

I may have some syntax wrong up there, but the idea is that you are passing in a
parameter of some datatype (text here) and then using it in a select statement
against a column which we will assume is of some datatype other than "text". If
the index your query uses is not picking up the datatype conversion properly,
then you may have a sequential scan instead.

To verify this, you might do these:

EXPLAIN
select * from foobar_table where
col_a=<THE_CONSTANT_VALUE>;

vs.

EXPLAIN
select * from foobar_table where
col_a=<THE_CONSTANT_VALUE>::<THE_PARAMETER_TYPE>;



--
~~~~~~~~~~~~~~~~| Genius may have its limitations, but stupidity is not
Jeff Boes       | thus handicapped.
jboes@qtm.net   |        --Elbert Hubbard (1856-1915), American author




Re: Dramatic slowdown of sql when placed in a function

From
mike_moran@mac.com (Mike Moran)
Date:
Jeff Boes <jboes@nexcerpt.com> wrote in message news:<971d398a3317a5af5086277b2a1487f4@news.teranews.com>...
> At some point in time, mike_moran@mac.com (Mike Moran) wrote:
> 
> >Hi. I currently have a fairly large query which I have been optimizing
> >for later use in a function. There are things in the query which I
> >have been keeping constant whilst optimizing, but which are variables
> >in the function. When I run this query as sql, with the `variables'
> >constant, I get a runtime of about 3 or 4 seconds. However, when I
> >place this same sql in an sql function, and then pass my constants
> >from before in as arguments, I get a runtime of about 215 seconds.
> >
[ ... ]
> 
> My first guess would be that the indexes being used in the query are
> mis-matching on data type compared to your function arguments. For instance,
[ ... ]

Hi. I think it is something like this that is going on. A couple of
the variables are dates which are
specified in the table as 'timestamp without time zone', whilst the
function was using 'timestamp with time zone'. I confirmed the
slowdown by casting the types to the 'slow' type in the original
query.

However, when I change the signature of the function and do a cast of
the variable within the function body I still get the same speed. I
even cast the arguments to the function given at the psql prompt and
still I get the same speed.

I will have to sanity-check this again tomorrow (posting from home)
but I couldn't see anywhere else that I could force the type to be the
same as that specified on the table.

Many thanks,

-- 
Mike