Bc_desktop XFCE session fails: “No protocol specified / xfce4-session: cannot open display” (OOD 4.2.4, TurboVNC 3.3.1, ParallelCluster 3.16.1)

Get Help
1

Hello,

Interactive Desktop (XFCE) sessions launch through Slurm, and Xvnc and websockify start, but xfce4-session exits immediately with “cannot open display”. vncserver run manually on the compute node works.

Environment

  • AWS ParallelCluster 3.16.1, Slurm 25.11.8
  • OS: Ubuntu 22.04.5 LTS
  • Open OnDemand 4.2.4, on a dedicated host
  • TurboVNC 3.3.1, /usr/local/turbovnc/bin/vncserver
  • Compute node dftq-dy-dnode-1 is not resolvable by name from the OOD host, so IP addresses must be used.

Configuration

/etc/ood/config/apps/bc_desktop/testcluster.yml

title: "HPC Slurm Cluster"
cluster: "my_slurm_cluster"
attributes:
  desktop: "xfce"
  bc_queue: null
  bc_num_slots: 1
  bc_account: null

/var/www/ood/apps/sys/bc_desktop/submit.yml.erb

script:
  copy_environment: true
batch_connect:
  template: "vnc"
  script_wrapper: |
    export PATH="/usr/local/turbovnc/bin:$PATH"
    export WEBSOCKIFY_CMD="/usr/local/python3/bin/websockify"
    %s

Modifications made

  1. In vncserver, so the node is reachable by IP:
    chop($host = \uname -n`);changed tochop($host = `hostname -i`);`
  2. In vncserver, the startup message:
    warn "\nDesktop '$desktopName' started on display $host:$displayNumber\n"; changed to warn "\nDesktop '$host' started on display $host:$displayNumber\n";
  3. In template/desktops/xfce.sh, commented out the two lines below to work around Failed to init libxfconf: Could not connect: No such file or directory:
   xfconf-query -c xfce4-session -p /startup/ssh-agent/enabled -n -t bool -s false
   xfconf-query -c xfce4-session -p /startup/gpg-agent/enabled -n -t bool -s false

Running /usr/local/turbovnc/bin/vncserver -wm xfce -verbose manually on the compute node works as expected.

xauth list output

deftq-dy-node-1.testcluster.pcluster:1 MIT-MAGIC-COOKIE-1 0614abe0a
192.168.1.6:1 MIT-MAGIC-COOKIE-1 0614abe0a

output.log

Setting VNC password...
Starting VNC server...
Desktop '192.168.1.6' started on display 192.168.1.6:1
Successfully started VNC server on 192.168.1.6:5901...
[websockify]: pid: 17117 (proxying 55635 ==> localhost:5901)
[websockify]: log file: ./websockify.log
[websockify]: waiting ...
No protocol specified
xfce4-session: cannot open display: .
Type 'xfce4-session --help' for usage.
Desktop 'xfce' ended with 1 status...
[websockify]: started successfully (proxying 55635 ==> localhost:5901)
Scanning VNC log file for user authentications...
Generating connection YAML file...
Cleaning up...
Killing Xvnc process ID 29438

Questions

  1. Where does the OOD VNC template set DISPLAY before xfce.sh runs, and what value should it have? The error shows ..

  2. Could changing $host to an IP in vncserver cause the “No protocol specified” error, because the xauth cookie is keyed by IP while the client looks up the real hostname? If so, what is the supported way to use IPs for compute nodes without patching vncserver?

  3. Is copy_environment: true appropriate for bc_desktop, or can it leak DBUS_SESSION_BUS_ADDRESS from the OOD host and cause the libxfconf error?

  4. For websockify, how to change from

    proxying 55635 ==> localhost:5901

    to

    proxying 55635 ==> 192.168.1.6:5901