Message-ID: <47E75F18.9060206@postnewspapers.com.au>
Date: Mon, 24 Mar 2008 16:58:16 +0900
From: Craig Ringer <craig@postnewspapers.com.au>
User-Agent: Thunderbird 2.0.0.12 (Windows/20080213)
MIME-Version: 1.0
To: firerox@centrum.cz
Cc: pgsql-performance@postgresql.org
Subject: Re: slow pg_connect()
References: <200803240840.8128@centrum.cz>
In-Reply-To: <200803240840.8128@centrum.cz>
Content-Type: text/plain; charset=windows-1250; format=flowed
Content-Transfer-Encoding: 7bit

firerox@centrum.cz wrote:
> It takes more then 0.05s :(
>
> Only this function reduce server speed max to 20request per second.
>   
If you need that sort of frequent database access, you might want to 
look into:

- Doing more work in each connection and reducing the number of 
connections required;
- Using multiple connections in parallel;
- Pooling connections so you don't need to create a new one for every job;
- Using a more efficient database connector and/or language;
- Dispatching requests to a persistent database access provider that's 
always connected

However, your connections are indeed taking a long time. I wrote a 
trivial test using psycopg for Python and found that the following script:

#!/usr/bin/env python
import psycopg
conn = pyscopg.connect("dbname=testdb")

generally took 0.035 seconds (350ms) to run on my workstation - 
including OS process creation, Python interpreter startup, database 
interface loading, connection, disconnection, and process termination.

A quick timing test shows that the connection/disconnection can be 
performed 100 times in 1.2 seconds:

import psycopg
import timeit
print timeit.Timer('conn = psycopg.connect("dbname=craig")', 'import 
psycopg').timeit(number=100);

... and this is still with an interpreted language. I wouldn't be too 
surprised if much better again could be achieved with the C/C++ APIs, 
though I don't currently feel the desire to write a test for that.

--
Craig Ringer