Re: postmaster dies on EOF?

"Denis A. Doroshenko" <d(dot)doroshenko(at)omnitel(dot)net> writes:
> one day i decided to switch password authetntication in pg_hba.conf
> for local tcp/unix connection as well, ran psql and saw 'Password:' for
> the first time :-) But, when i pressed ^D in answer it showed many
> lines with 'Password:' at the begining and then postmaster died.

This is a previously reported bug in psql. It gets into a loop wherein
it keeps opening new connections to the postmaster and then leaving them
hanging at the postmaster's demand for password. That part is fixed in
current sources; also the postmaster doesn't elog(FATAL) on running out
of open file slots anymore.

But there's still a potential for denial of service by a broken client,
because the postmaster will just sit there waiting for password
responses on all these open connections, and there'll be no way for
other clients to get in. There's been some talk of instituting a
timeout on the connection request protocol to fix that, but nothing's
been done about it yet AFAIK.

regards, tom lane