gethostbyname problem

[Pete Wyckoff]

vanw at tticluster.com wrote on Tue, 07 Feb 2006 21:58 -0500:
> I'm getting "p4_error: Could not gethostbyname..." error. Although
> /etc/hosts on all nodes has all the correct entries for all nodes. I'm at
> a loss. Also, I get the same error if I use mpirun. The mpich program is a 
> FORTRAN90 program compiled with pgf90 -Bstatic on a 2.4 kernel, though 
> the execution nodes are 2.6.
[..]
> p0_3152:  p4_error: Could not gethostbyname for host blue92; may be 
> invalid name
> : 61

Generally this is a problem with /etc/hosts or /etc/resolv.conf on
your compute nodes.  Your use of a statically-compiled fortran
binary from a different GLIBC version complicates things, though:
name resolution happens via a dynamically loaded library whether you
compile statically or not.  You may want to debug with some C codes
and see if you can figure it out.  "strace" is also helpful to watch
what it's trying to do.

And as you point out, since this fails both with mpiexec and mpirun,
you may get better advice asking the MPICH developers or some
generic beowulf list, showing them the mpirun failure.

		-- Pete