Re: [BUGS] problem creating index in 6,5,3 - Mailing list pgsql-bugs

Karl DeBisschop <kdebisschop@range.infoplease.com> writes:
> The problem manifests itself as

> webusers=> CREATE INDEX zdaily_id ON daily (id);
> pqReadData() -- backend closed the channel unexpectedly.
>         This probably means the backend terminated abnormally
>         before or while processing the request.

Is a corefile dumped in the database directory when this happens?
If so, can you get a backtrace from it with gdb?

> This did work for us using 6.5.1 compiled in house.  Our problems
> started with the 6.5.3 RPM.

Hmm.  We have been seeing some reports recently of other odd behavior
with the Linux RPMs (see the "ordering RH6.1" thread on pgsql-hackers).
Would you take the time to compile 6.5.3 from source and see if you
still see the same behavior?

If you do this, please compile the backend with full debug+assert
support, which is to say
  (a) add --enable-cassert to the configure command;
  (b) do   make COPT=-g all  instead of just make all.
That should make the gdb backtrace a lot more informative, if
a coredump happens...

            regards, tom lane