Thread: Client connection lost; still in the task list

Hello,

We've just put into production with our main application which we
ported from MSSQLserver to PostgreSQL;

We are overall very pleased with PostgreSQL speed and stability are fantastic.

We have one user which PC keeps freezing and despite rebooting the
computer several times since this morning I can find more than 24
connections for this user when there should be at most 2. So we are
experiencing 'too many users connected' errors

So my question is what makes PostgreSQL realise that a connection is
dead ? when will it drop it and free the connection ?

Doing  a  ps ax  I see many connection related to that IP address,
can I safely kill the processes which are obviously redundant ? if
yes, which signal to send ? (I do not want to kill the postmaster,
just the connection).

Going forward is there a good way of doing this ?

We are running on Centos 4, PostgreSQL 8.3

thanks in advance for your input

Didier

"Didier Gasser-Morlay" <didiergm@gmail.com> writes:
> So my question is what makes PostgreSQL realise that a connection is
> dead ? when will it drop it and free the connection ?

It'll drop it when the kernel tells it the connection is dead.
Standard TCP timeouts are measured in hours, though.  You might
be able to get a shorter timeout to be applied by fooling with
the "keepalive" parameters, if those can be adjusted on your platform.

            regards, tom lane

You're right, this morning most connections had been freed.

I have seen that the tcp_keepalive
 is 7200 second, I'll shorten that to see.

thanks

Didier


On Tue, Apr 22, 2008 at 7:29 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> "Didier Gasser-Morlay" <didiergm@gmail.com> writes:
>  > So my question is what makes PostgreSQL realise that a connection is
>  > dead ? when will it drop it and free the connection ?
>
>  It'll drop it when the kernel tells it the connection is dead.
>  Standard TCP timeouts are measured in hours, though.  You might
>  be able to get a shorter timeout to be applied by fooling with
>  the "keepalive" parameters, if those can be adjusted on your platform.
>
>                         regards, tom lane
>

Resucing the tcp_keepalive to 900 seconds did help immensely

thanks for the advice Tom

Didier

>You're right, this morning most connections had been freed.
>
>I have seen that the tcp_keepalive
> is 7200 second, I'll shorten that to see.

>thanks

>Didier


On Tue, Apr 22, 2008 at 7:29 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> "Didier Gasser-Morlay" <didiergm@gmail.com> writes:
>  > So my question is what makes PostgreSQL realise that a connection is
>  > dead ? when will it drop it and free the connection ?
>
>  It'll drop it when the kernel tells it the connection is dead.
>  Standard TCP timeouts are measured in hours, though.  You might
>  be able to get a shorter timeout to be applied by fooling with
>  the "keepalive" parameters, if those can be adjusted on your platform.
>
>                         regards, tom lane
>