Обсуждение: avoid prepared statements on complex queries?

> If only there was a way to have parameterized queries without using prepared
> statements....

Take a look at the prepareThreshold connection string parameter.
---
Maciek Sakrejda | System Architect | Truviso

1065 E. Hillsdale Blvd., Suite 215
Foster City, CA 94404
(650) 242-3500 Main
www.truviso.com

>I tried playing around with the prepareThreshold parameter, by setting it in the JDBC connection string.  I tried
settingit to zero, which should cause it to not use a prepared statement, right?  Query is still >slow.  I also tried
settingit to 1 and 3, and in all cases it's slow.  Is there any way to verify if postgres is using a planned query? 

I think 0 should do it, yes (technically, it's using an unnamed
prepared statement rather than not using one at all, but this is the
behavior you're looking for). To verify, you should be able to change
log_min_messages (to debug2?) and see all plan/execute steps in the
server logs. Alternately, if you're not using SSL, you could just fire
up Wireshark and see what's on the wire (it has a built-in PostgreSQL
protocol plugin, so this is pretty straightforward). In either case,
the unnamed prepared statements should have no name (in Prepare, Bind,
and Execute messages); if memory serves, the named ones are something
like S_1, S_2, etc..

> "However even when plans are not cached, you will still run into the issue
> that an unnamed statement using out-of-line parameter values may generate a
> less efficient plan than an unnamed statement using inline parameter values,
> because there is no way to tell the planner at the protocol level "I am
> really only ever using this query once, please give me a specific plan for
> these values and don't worry about generating a plan that is correct for
> other values too"."
> Maybe that's what's happening?

Maybe, although I've never seen that happen and as I understand, the
driver always sends parameters along with the statement when using
unnamed statements, so I'm not sure what would trigger this. Perhaps
Oliver will chime in (he's still active on the list).

You may also want to try an EXPLAIN ANALYZE on your query in both
protocol versions and see what the plan differences are.

---
Maciek Sakrejda | System Architect | Truviso

1065 E. Hillsdale Blvd., Suite 215
Foster City, CA 94404
(650) 242-3500 Main
www.truviso.com

On Wed, 16 Nov 2011, Maciek Sakrejda wrote:

> Maybe, although I've never seen that happen and as I understand, the
> driver always sends parameters along with the statement when using
> unnamed statements, so I'm not sure what would trigger this.

The standard example is "WHERE col LIKE ?".  If the parameter is something
like 'abcdef%' then the query can potentially be converted to a range
scan, but only if knows the parameter value.  When preparing this with an
unnamed statement the server has the parameter value, but cannot assume
that the same unnamed statement won't be used with a different parameter
value later and must come up with a plan that is correct for all possible
parameters even if it's not ideal for the current one.

Kris Jurka

On Tue, 15 Nov 2011, Anish Kejariwal wrote:

> I'm running into an issue where a complex query is performing significantly
> slower if I use a prepared statement.
>
> Some notes about the query:
> -I'm calling a postgres function, via: select * from foo(?,?,?,?,?,?,?)
> -I'm getting back 4 million results
> -the underlying query is quite complex joining several tables, and the core
> table is partitioned over a couple hundred tables.
>
> Here's the java code:
>
> conn.setAutoCommit(false);
> pstmt = conn.prepareStatement("select * from foo(?,?,?,?,?,?,?)");
> pstmt.setFetchSize(1000);
>
> //and, then I do the usual thing to set parameters in the prepared statement
> stmt.setInt(1, SOME_NUMBER);
> pstmt.setArray(2, genepoolConn.createArrayOf("integer", ARRAY_OF_INTEGERS));
> //and so on
>
> So, obviously when using a prepared statement, postgres is coming up with
> the wrong execution plan when the the parameter list is unexpectedly large.

You haven't really shown that.  Unless you issue the same query, manually
interpolating in the query parameters and get a much faster time, all
you've shown is that things get slower with a larger number of values.

What language is your function written in?  Unless it's a sql function
which could be inlined, I wouldn't have expected any difference between
prepared vs inline query execution.

Kris Jurka

Kris Jurka <books@ejurka.com> writes:
> On Wed, 16 Nov 2011, Maciek Sakrejda wrote:
>> Maybe, although I've never seen that happen and as I understand, the
>> driver always sends parameters along with the statement when using
>> unnamed statements, so I'm not sure what would trigger this.

> The standard example is "WHERE col LIKE ?".  If the parameter is something
> like 'abcdef%' then the query can potentially be converted to a range
> scan, but only if knows the parameter value.  When preparing this with an
> unnamed statement the server has the parameter value, but cannot assume
> that the same unnamed statement won't be used with a different parameter
> value later and must come up with a plan that is correct for all possible
> parameters even if it's not ideal for the current one.

BTW, this should be all better in 9.2 ... so if anyone is thinking of
expending lots of effort to fix it on the JDBC side, don't bother.

            regards, tom lane