Re: DB Connections in TIME_WAIT state

[Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>]

John Gateley <gateley(at)jriver(dot)com> writes:
> Hi, I'm using the Pg perl module to connect to Postgresql 8.1
> via localhost (127.0.0.1) from webscripts. I'm noticing a lot
> of TIME_WAIT socket connections:

> tcp        0      0 127.0.0.1:39291         127.0.0.1:5432          TIME_WAIT  
> tcp        0      0 127.0.0.1:60720         127.0.0.1:5432          TIME_WAIT  
> tcp        0      0 127.0.0.1:60735         127.0.0.1:5432          TIME_WAIT  
> tcp        0      0 127.0.0.1:60769         127.0.0.1:5432          TIME_WAIT  
> tcp        0      0 127.0.0.1:39281         127.0.0.1:5432          TIME_WAIT  
> ...

This seems like a kernel bug to me.  The TCP stack ought to know it
doesn't need any shutdown delay on a local connection.

> I think the number is high enough (200~250) that sometime the
> server runs out of sockets.

Maybe you need to stop using so many connections --- quite aside
from any kernel issues, a database session isn't exactly cheap
to launch.  Consider some form of connection pooling.

			regards, tom lane