Hi there,

I doubt very much that is a problem with the GT, more my trying
to understand what I must have missed.

I have a client machine that's been installed with a number of
VDT-packaged RPMs (GT 4.0.7) from here:

 http://vdt.cs.wisc.edu/vdt_rpms/1.10.1/release-1/x86_rhap_5/

The client has been successfully used to:

1) Interact with an iRODS server
2) Do GridFTP with a pacman/VDT-2 GT4 job submission gateway using UberFTP
3) Access a pacman/VDT-2 GT4 job submission gateway (JSG) using
    globusrun-ws in passive mode

Note that I had to set $GLOBUS_TCP_PORT_RANGE to get "active" GridFTP to
work for the UberFTP client.

The working globusrun-ws command was:

globusrun-ws -passive -submit -s -S -F jsg.dom.ain -Ft Fork -c /usr/bin/id

Ok, here's where it starts to get slightly messy:

if I remove the "-passive" then a "reverse" GridFTP connection made
from the JSG tries to connect to a random port on the client and falls
foul of the firewall.

If set GLOBUS_TCP_PORT_RANGE to match the firewall, akin to what was
needed for the UberFTP, I get an almost immediate seg. fault.

If I unset GLOBUS_TCP_PORT_RANGE but open up all incoming ports on
the client I get an almost immediate seg. fault.

Setting GLOBUS_TCP_SOURCE_RANGE whilst :fully open" makes no
difference either.

If I then restrict the firewall back the original settings, I get
back to having the attempted return GridFTP connection fail.


I have tried turning on what debugging there is and looking in
various log but no clues.


Questions thus are:

Why would the segfault only happen when the firewall was opened up
and not when I specify port ranges in the environment ?

What's likely to be causing the segfault ?

Do GLOBUS_TCP_PORT_RANGE/GLOBUS_TCP_SOURCE_RANGE have any client-side
influence on globusrun-ws submissions ?

Is there a way to turn on even more debug/logging info ?

I appreciate that this is an old VDT/GT but the fact that the passive
submission works has me believing that this should be a simple way
to get a test client operational without needing a full pacman/VDT2
install.

Unless of course the issue is of the interation with the VDT1 client
accessing the VDT2 "server".

Open to suggestions,
Kevin

-- 
Kevin M. Buckley                                  Room:  CO327
School of Engineering and                         Phone: +64 4 463 5971
 Computer Science
Victoria University of Wellington
New Zealand

Reply via email to