Re: Global Named Prepared Statements

[Samba <saasira@xxxxxxxxx>]

If Stored Procedures are equivalent to prepared statements [ as far as preparing the query plan is concerned], then what i'm looking for is perhaps a Global Prepared Statements at the client/driver side.

Specifically, It wold be good if the JDBC driver prepares all the queries for invoking stored procedures at once per JVM so that each connection need not incur the cost of preparing [parsing and storing] those queries per connection.

Thus we can put all the queries [stored procedure calls] at a single place, and prepare those queries during boot of the server [or deployment of the web application], and then execute those queries endless times by closing just the resultset object while keeping the statement open for ever.

I know this is not form to discuss the JDBC related questions but put my thoughts here to complete the question i raised. If folks think this idea is valid then i will take it up with the JDBC Driver team.

Thanks and Regards,

Samba

=====================================================

On Tue, May 15, 2012 at 6:46 PM, Merlin Moncure <mmoncure@xxxxxxxxx> wrote:

On Tue, May 15, 2012 at 1:21 AM, Martijn van Oosterhout
<kleptog@xxxxxxxxx> wrote:
> On Tue, May 15, 2012 at 05:38:27AM +0530, Samba wrote:
>> Hi,
>>
>> Does postgresql support Global Prepared Statements, which are prepared only
>> once per server and not per every connection?
>
> As pointed out, no.
>
>> Problem with per-connection prepared statements is that the onus of
>> preparing those statements for each connection lies with the client which
>> makes those connections. Ideally, the performance of an application must be
>> decided by the Server that hosts and not by the client that uses the
>> service.
>
> How is this different from using CREATE FUNCTION to create a function
> which has the desired effect?  This is a well understood and commonly
> used paradigm.  When using a connection pooler any query plan caching
> will happen automatically.

this is not necessarily true, right?  for example, 'sql' language
functions don't cache plans while plpgsql functions mostly (no
EXECUTE) do.  other languages will typically have ability to save
plans  (or not).  but yeah, functions generally cover this case quite
nicely.

i rarely use prepared statements anymore but if you're counting
microseconds of latency for trivial queries, they still have a niche
role...but to really see the benefit you'd want to be coding directly
against the C api and making the appropriate calls (PQexecPrepared,
etc).

merlin