Re: URGENT HELP about 'duration' stats - Mailing list pgsql-hackers

From Camilo Porto
Subject Re: URGENT HELP about 'duration' stats
Date
Msg-id BLU111-W5283EBA5222878EBF17393BC8C0@phx.gbl
Whole thread Raw
In response to Re: URGENT HELP about 'duration' stats  ("Filip Rembiałkowski" <plk.zuber@gmail.com>)
List pgsql-hackers


[Camilo Porto]



> Date: Tue, 30 Oct 2007 16:09:01 +0000
> From: plk.zuber@gmail.com
> To: camiloporto@hotmail.com
> Subject: Re: [HACKERS] URGENT HELP about 'duration' stats
> CC: pgsql-hackers@postgresql.org
>
> 2007/10/30, Camilo Porto <camiloporto@hotmail.com>:
>
> > > > I am simulating only 1 client with the Benchmark. Can 1 Client submit
> > > > parallel queries, in single-processor enviroment?
> > > If this client uses two connections, you can run two queries in paralell.
> > The client uses only 1 connection. In this situation is possible that
> > the EXECUTOR's duration time become greater than the time period which the
> > Query was observed? (as stated in my first topic)?
>
> I guess it's possible under some circumstances...
> 2007-10-30 16:07:00 GMT [123] LOG: duration: 99000.000 ms select longfunc()
> 2007-10-30 16:07:01 GMT [123] LOG: duration: 1000.000 ms select shortfunc()
> interval is 1 second, sum of durations 100 seconds :)
>
>
> AFAIK, timestamps in the front of each line are assigned by log
> writer, ie. *in the moment of writing* to the log. I'd better trust
> "duration: xxx ms " messages. they are calculated in backend directly.
>
>
> In this log sample you showed us, the sum of durations is circa 625
> ms. and the interval between first and last log entry is circa 822 ms.
> If you have a test case which shows that much difference you speak of,
> could you please present it here, along with your logging settings?

I tried to generate a log file that could present the above situation, but was not sucessfully! This case happens when I wa using version 7.4 os PostgreSQL. Using 8.2.5, the problem (with 1 client and 1 connection) appear not to happen anymore! I've just tried with more than one client (1 connection per client) and the problem turn to appear. But now this make sense (considering the comments submitted to this topic).

Well, I will think about how can I calculate the % of time the executor passing executing queries along a period of time.

I would be grateful if someone could help me to resolve this...

Many Thanks!!!

Camilo
>
>
>
> --
> Filip Rembiałkowski


Conheça o Windows Live Spaces, o site de relacionamentos do Messenger! Crie já o seu!

pgsql-hackers by date:

Previous
From: Tom Lane
Date:
Subject: Re: Postgresql 8.3 beta crash
Next
From: Jeff Amiel
Date:
Subject: Re: [GENERAL] AutoVacuum Behaviour Question