Opened 3 years ago

Last modified 2 years ago

#961 qa defect

GalaxSee, Life X11 display non-functional

Reported by: skylar Owned by:
Priority: major Milestone: 3.4.0
Component: Both Version: 3.4.0
Keywords: Cc:
Blocked By: Blocking:
Estimated Hours: 4 Total Hours: 3

Description (last modified by skylar)

Problem of DISPLAY propagating through MPI

Attachments (2)

mpich2_errors_r5330 (1.2 KB) - added by skylar 3 years ago.
openmpi_errors_r5330 (1.1 KB) - added by skylar 3 years ago.

Download all attachments as: .zip

Change History (20)

Changed 3 years ago by skylar

Changed 3 years ago by skylar

comment:1 Changed 3 years ago by skylar

added snippet to test which host each rank is on:

#include <errno.h> #include <strerror.h> ...

if((gethostname(hostname,HOSTNAME_LENGTH)) == -1) {

fprintf(stderr,"Failed to get hostname: %s\n",

strerror(errno)

);

}

fprintf(stderr,"Rank %d is on %s\n",g_mpi.rank,hostname);

comment:2 Changed 3 years ago by skylar

confirmed that rank 0 is on node000

comment:3 Changed 3 years ago by skylar

confirmed that xhost access control is disabled (i.e. xhost +)

comment:4 Changed 3 years ago by skylar

replacing display name 0 (integer) with ":0" (string) in XOpenDisplay works!

dpy = XOpenDisplay(":0");

NB: XCreateGC still expects an integer

comment:5 Changed 3 years ago by skylar

In 5365//cluster/svnroot:

fixing XOpenDisplay for GalaxSee? re #961

comment:6 Changed 2 years ago by mmludin08

This seem to be fixed with 3.4.0-rc.5368.

comment:7 Changed 2 years ago by mmludin08

  • Status changed from new to qa

comment:8 Changed 2 years ago by skylar

  • Description modified (diff)
  • Total Hours changed from 0 to 3

comment:9 Changed 2 years ago by skylar

Problem is this:

===

#define NIL (0) ... dpy = XOpenDisplay(NIL);

===

this works if you run locally:

dpy = XOpenDisplay(":0");

but does not work if you run through SSH forwarding

printing getenv("DISPLAY") reveals that mpirun changes DISPLAY to this:

Display is node000:0.0

echo'ing DISPLAY from a terminal reveals that local DISPLAY is

:0

while ssh forwarded DISPLAY is

localhost:10.0

Hardcoding XOpenDisplay's display parameter works; how can we get DISPLAY passed through un-munged by mpirun?

comment:10 Changed 2 years ago by skylar

In 5481//cluster/svnroot:

use DISPLAY variable, works for mpich2 re #961

comment:11 Changed 2 years ago by skylar

mpich2 now works:

mpirun -genv DISPLAY ${DISPLAY} -np 8 -f ~/machines-mpich2 ~/GalaxSee/GalaxSee?.cxx-mpi 1000 500 400

comment:12 Changed 2 years ago by skylar

In 5482//cluster/svnroot:

use fourth display argument to allow passing in of DISPLAY variable re #961

comment:13 Changed 2 years ago by skylar

In 5483//cluster/svnroot:

remove debug, add comments re #961

comment:14 Changed 2 years ago by skylar

In 5484//cluster/svnroot:

merge addresses #961

comment:15 Changed 2 years ago by skylar

In 5485//cluster/svnroot:

error message needs newline re #961

comment:16 Changed 2 years ago by skylar

In 5486//cluster/svnroot:

updating GalaxSee? README re #961

comment:17 Changed 2 years ago by skylar

In 5498//cluster/svnroot:

merging in galaxsee changes re #961

comment:18 Changed 2 years ago by skylar

In 5499//cluster/svnroot:

life attempt re #961

Note: See TracTickets for help on using tickets.