plpgsql functions and the planner
Do SQL statements inside of plpgsql functions get planned upon every
execution, only when the function is first executed/defined, or something
else entirely?
For example, suppose I have a table foo and a function bar. Function bar
executes some SQL statements (select/insert/update) against table foo using
various indexed columns. When the function is created and first executed,
table foo is near empty so the statements in function bar use seqscans.
Then table foo is filled with many rows. Now, when bar is executed again,
will PG (8.3.1) know that a seqscan is no longer reasonable?
On Sun, Apr 27, 2008 at 2:06 AM, Matthew Dennis <mdennis@merfer.net> wrote:
Do SQL statements inside of plpgsql functions get planned upon every
execution, only when the function is first executed/defined, or something
else entirely?
They are planned on first execution and the plan is cached for later use.
-Doug
"Matthew Dennis" <mdennis@merfer.net> writes:
Do SQL statements inside of plpgsql functions get planned upon every
execution, only when the function is first executed/defined, or something
else entirely?
First executed per session.
Now, when bar is executed again, will PG (8.3.1) know that a seqscan is no
longer reasonable?
It won't notice until someone runs ANALYZE on that table. autovacuum should
notice that it's necessary and run but might not run promptly enough for your
purposes. You might have to run analyze within your function.
Before 8.3 it still wouldn't get replanned until you started a new session but
8.3 is more clever.
--
Gregory Stark
EnterpriseDB http://www.enterprisedb.com
Ask me about EnterpriseDB's 24x7 Postgres support!