Re: Disconnected Backends

"Glen and Rosanne Eustace" <agree(at)godzone(dot)net(dot)nz> writes:
> I have recently allowed one of my clients to use ipsql as a means to manage
> one of his databases on my server. The server is currently running 6.5.1
> and will be upgraded to 7.0 on 1/1 so if this is version specific the
> problem may go away.

"ipsql"? Did you mean psql? If not, I'm clueless what client software
you are talking about.

> It would seem that the postmaster either doesn't shut down a remote
> connection properly or is not able to detect when the remote client has gone
> away. I get lots of postmasters for what I know to be old connections still
> hanging about.

Hm. The backend should shut down when it's informed by the kernel that
the client connection socket has been closed. We have heard about
problems of this sort before, but AFAIR all the identifiable cases
have turned out to trace to client misbehavior, namely failure to close
a connection that the client isn't using anymore. What does "netstat"
show about the state of the connections? What is your client code
exactly?

regards, tom lane