Re: Query plans for plpgsql triggers

[Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>]

"Eric B. Ridge" <ebr(at)tcdi(dot)com> writes:
> When is the UPDATE statement inside foo() planned?  When the trigger  
> is first created, or when it's first used per backend, or every time  
> it's used per backend?

First use per backend, ignoring corner cases such as replacing the
function definition.

> I dunno what plan is being generated, but it's gotta be using a  
> sequential scan.

The issue is probably that the planner is seeing a parameterized
query.  Try this:

prepare foo(int8) as update some_other_table SET field = 'value' WHERE id = $1;
explain execute foo(42);

and see what plan you get.  If the id field has sufficiently
discouraging statistics then the planner may think that a seqscan
is the safest plan.  In a "normal" query where you're comparing id
to a constant, the planner can see whether the constant matches any
of the most common values for the column --- if it doesn't then an
indexscan is a good plan.

If you really want a replan every time, you can get it by using
EXECUTE.

			regards, tom lane