Обсуждение: psql & "unexpected EOF on client connection"

Hi Y'all.

I have been getting the dreaded "unexpected EOF on client
connection" in my system logs, but the surprising thing is that
it comes from psql (I think). I am executing psql
via a 'system' system call, and about 1 time out of 20 I get
the "unexpected EOF" message. Unfortunately, this
also causes the 'system' call to return an error, even though
all SQL code was executed properly. (After searching
postgresql.org I found out that the message normally
means that PGfinish wasn't called.)

the exact code in question is:
res=system("/usr/bin/psql <a-file-name");

Gory details available on request.

Thanks for any help,

John Gateley
gateley@jriver.com, J. River, Inc.

gateley@jriver.com writes:
> I have been getting the dreaded "unexpected EOF on client
> connection" in my system logs, but the surprising thing is that
> it comes from psql (I think). I am executing psql
> via a 'system' system call, and about 1 time out of 20 I get
> the "unexpected EOF" message. Unfortunately, this
> also causes the 'system' call to return an error, even though
> all SQL code was executed properly. (After searching
> postgresql.org I found out that the message normally
> means that PGfinish wasn't called.)

So it does.  I didn't think there were any paths in psql that would
lead to such an ungraceful exit, especially not if no SQL error occurs.

> Gory details available on request.

Please.  For starters, what platform are you on, and which PG version?

            regards, tom lane

Tom Lane wrote:
>
> gateley@jriver.com writes:
> > I have been getting the dreaded "unexpected EOF on client
> > connection" in my system logs, but the surprising thing is that
> > it comes from psql (I think). I am executing psql
> > via a 'system' system call, and about 1 time out of 20 I get
> > the "unexpected EOF" message. Unfortunately, this
> > also causes the 'system' call to return an error, even though
> > all SQL code was executed properly. (After searching
> > postgresql.org I found out that the message normally
> > means that PGfinish wasn't called.)
>
> So it does.  I didn't think there were any paths in psql that would
> lead to such an ungraceful exit, especially not if no SQL error occurs.
>
> > Gory details available on request.
>
> Please.  For starters, what platform are you on, and which PG version?

RedHat Linux 7.0, with kernel 2.2.17
Postgresql version 7.0.3, as well as the stock
version shipped with RedHat 7 (7.0.1???)

My app gathers a bunch of data, and then
dumps it into the database. It does not
retrieve any data from the database (other
apps do that). About 1 time out of 20, the call
to PGconnectdb fails with a segfault signal. Because
of this, I run the PGconnectdb/PGexec etc. in a
forked process. If the forked process dies because
of the signal, then I try using system("psql ....")
as described above. About 1 time out of 20 this
fails.

Thanks,

j

gateley@jriver.com writes:
> About 1 time out of 20, the call
> to PGconnectdb fails with a segfault signal.

Well, you've already got some pretty severe problems then.  Seems to me
you ought to track down the reason for this failure before worrying
about cosmetic issues like an ungraceful disconnect.

I suggest compiling both libpq and your app with -g and extracting a
backtrace from the core file produced by the segfault ...

            regards, tom lane