rexec hangs or fails on attempt to access OpenVMS Alpha

[Bill Pedersen]

Wim:

Did some testing with my I64 system...

With TCPIP disabled the attempt to use rexec times out with connection refused.

With rexec service disabled the same thing happens.

rexec is implemented by the tcpip$inetacp so there is no process running except when the service is requested. Sort of like how DECnet does the start of a job.

So I can not see that having it disabled is not the issue.

I also checked by stopping the service rather than disabling, same result.

I still suspect there is something either with my firewall or some other firewall in the path...

Thanks,

Bill.

[Steven Schweda]

> With TCPIP disabled the attempt to use
> rexec times out with connection refused.

It "times out", or fails promptly? Normally,
"connection refused" is a quick failure, and
is what's expected when you can contact the
server system, but the particular service
(server) program is disabled (or absent).

> I still suspect there is something either
> with my firewall or some other firewall in
> the path...

Sounds plausible to me.

[Bill Pedersen]

It is not immediate. It is more like 60 to 90 seconds in all cases except where it works, which is immediate.

Still looking into other possible issues on the paths.

Continuing to work on the alternate solution.

Bill.

[Steven Schweda]

I wrote a Q&D program on a Solaris system
which calls rexec(). (There is no obvious
"rexec" program there.) At least there, you
get a choice about opening a second channel
to handle stderr. I pointed the program at
my main VMS system. Without the second
channel, I see a thing like this from TCPIP
SHOW DEVICE /HOST = sol (which is 10.0.0.38):

bg5619 STREAM 512 32794 REXEC 10.0.0.38

With the second channel, I see things like
these:

bg5717 STREAM 512 32795 REXEC 10.0.0.38
bg5718 STREAM 1021 32796 10.0.0.38

Port 1021 seems to be the first free one if
you start at 1023 and search down. ("rsh"
sessions seem to be eating 1023 and 1022 at
the moment.)

So, if your rexec stuff is doing the
two-channel thing, then opening port 512 at
the firewall would seem to be necessary but
not sufficient.

All of which would seem to be consistent with
WVdW's first reply.

Rexec (with that stderr channel) would seem
to be a sub-ideal choice for communication
through a firewall (unless you don't mind
opening all the ports 1-1023).