Thread: Prepared statement vs. parameterized

Prepared statement vs. parameterized

From
"Bucky Jordan"
Date:
I've only used the PG JDBC driver for pretty standard things in the
past, and on pretty small db's. However, the current project I'm working
on will be using rather large data sets and likely make use of some more
advanced features. So, I have a few questions related to JDBC
parameterized statements and their relation to prepared statements on
the PG backend.

It appears that parameterized statements are always prepared- is this
correct? If so, why (JDBC spec, performance..)?

As a general rule, we always use parameterized statements (courtesy of
Hibernate). So, if the above is true, then all of our queries are also
prepared. After reading previous discussions on -jdbc and -performance,
it appears that this is very often not so good for performance since the
PG backend basically has to plan a query without values.

Now, on a small database, this wouldn't be much of an issue, especially
as the planner gets better at picking reasonable plans without values
(seems pretty tricky to me..). However, as your data sizes grow (> 500
million records or so) I would think it's critical to have the values
when planning. (I'm starting a project with very large data sizes, so I
will soon be testing some of this..)

So my question is, given the above, would it possibly makes sense to be
able to use parameterized statements without preparing them? Maybe this
could be a user configurable option? (if not, the other request is
really more for the back-end to improve the performance of prepared
statements... which seems quite a bit more complicated).

Thanks,

Bucky

Re: Prepared statement vs. parameterized

From
Kris Jurka
Date:

On Fri, 27 Oct 2006, Bucky Jordan wrote:

> I've only used the PG JDBC driver for pretty standard things in the
> past, and on pretty small db's. However, the current project I'm working
> on will be using rather large data sets and likely make use of some more
> advanced features. So, I have a few questions related to JDBC
> parameterized statements and their relation to prepared statements on
> the PG backend.
>
> It appears that parameterized statements are always prepared- is this
> correct? If so, why (JDBC spec, performance..)?

They are always prepared (using the V3 protocol), but there are two
methods of prepared statement execution.  A statement can be prepared with
an explicit name (that is good for the rest of the connection) or on the
"unnamed" statement (that is good until reused).  Statements prepared on
the unnamed statement use the bound parameter values for planning purposes
so you should get performance equivalent to a non-parameterized query.
Statements prepared with explicit names do use generic planning
placeholders and can get bad query plans if you have uneven data
distributions.  When a PreparedStatement object is first created it will
use the unnamed statement for its first several uses.  Once the driver
detects that it is being reused again and again it will switch over to a
named statement.  This execution count is kept in the PreparedStatement
object and no pooling is done, so you have to look at your
PreparedStatement object lifetime to determine if it's being reused or
created again.  The switchover point is configured (or disabled) via a URL
parameter prepareThreshold[1] or may be set on each individual connection
or statement by using pg specific java code.

Kris Jurka

[1]
http://jdbc.postgresql.org/documentation/81/connect.html#connection-parameters