[mpich-discuss] On gethostbyname() and related issues on Macs

[Nicolas Rosner]

Dear Pavan,

Maybe I wasn't too clear about this point (should have included a
cmdline) -- please re-read the following, just in case:

>> a) it suffices to feed Hydra an invalid target hostname (via -f or
>> -hosts, but also via env var, right?) to reproduce the exact same
>> errors**,
>>
>>   [mpiexec at fiona] HYDU_sock_is_local (./utils/sock/sock.c:515):
>>     unable to get host address (No such file or directory)
>>   [mpiexec at fiona] main (./ui/mpich/mpiexec.c:344):
>>     unable to check if fiona is local
>
> I'm surprised with the error message above. Hydra should be hardcoding the
> hostname as "localhost" in this case (see ui/mpich/mpiexec.c:277). I'm not
> sure where the "fiona" hostname is coming from.

I had told Hydra about `fiona' myself in this case (!).

What I tried to say up there ("it suffices...") was: "Here's what
happens when you DO tell Hydra explicitly about a hostname that isn't
valid, via -hosts, -f, etc."

In other words:

* I found it easy to reproduce Alex's exact error messages by lying to
Hydra directly (i.e. by explicitly requesting a host that I knew it
wouldn't be able to resolve).

    [Details: tried -hosts and -f with identical results, and IIRC,
there was also a way to specify a machine file through an env variable
as well (was just wondering whether Alex might have done something
like that at some point; the cmdline options seem unlikely, since he
said he was using exactly "mpiexec -n 2 ./a.out", right?).]

* However, despite many tries, I failed to run into any problem
*without* resorting to such a direct lie (that is, merely by altering
system settings, TCP/IP config, etc, in every odd-yet-plausible way I
could imagine).

    [Details: the goal was to hit some context (possibly a rather
broken, yet not totally useless one, e.g. web browsing must still be
possible) under which Hydra would choke on "mpiexec -n 2 ./a.out"
given no further info. But it ran just fine every time, and at least
according to the cpi example's output, never even resorted to using
`localhost', or so it seemed.]


> I added some more debug messages to Hydra. Can you [...]

Sure. I'll post the results soon.

Regards,
Nicolás