Marek Peca <marek(at)duch(dot)cz> writes:
>> In any case, that complaint should be directed to your kernel vendor
>> not us. We do not control how long the TCP stack waits before declaring
>> the connection dead.
> Well, this is what I wanted to hear. So if I call PQstatus() or PQexec(),
> the libpq sends some data to the TCP pipe, and you expect, that my broken
> underlying TCP/IP subsystem pretends, that the connection is OK and the
> data have been sent. Right?
Well, if you *never* get a failure, then yeah I think you have a broken
TCP stack. The default timeouts on this sort of thing are annoyingly
long, but they aren't infinite.
regards, tom lane