FrontEnd ReadMe

TVConnection is launched from tvconnections.sh in connection docker to manage password to Frontend of HPC machine.

It creates VNC secure flux with TVSecure in secure host and a communication socket with the client browser through Flask server with action.json list of actions.

It launchs the script job.

Launch TileServer by hand on Frontend with logging : ipython3 .tiledviz/TileServer.py –debug 2>&1 |tee -a /tmp/TileServer_ One can switch DEBUG to increase verbosity and control message with clients.

Generic functions for all cases : "tileNum" is the number of a tile in a TileSet and "tileId" is a string for its Id given by containerId(tileNum+1).

Get_client_IP: return the IP of a client
tunnel : launch /opt/tunnel_ssh for activating the secure tunnel of VNC stream.
vnc : launch the vnc server with /opt/vnccommand script.
init_wmctrl : call a global wmctrl call to list all windows in the tile.
clear_VNC : send clear-all command to the vnc server if it appears problems with keyboard interaction.
changeSize : change tile resolution.
fullscreenThisApp : toggle fullscreen option for a window.
showThisGUI : toggle above a window.
clickPoint : send a click event on a given position.
kill_all_containers : this function must be defined in all case script for killing all containers.

Action sur les conteneurs : grabwin kill_ssh launch_command launch_dockers launch_tunnels launch_vncviewer launch_Wall launch_x11vnc replace_init_window rescale_docker stop_dockers

Client-serveur VMD : analyse_newNoeuds.py UDP_2_multi_TCP.py

Actions globales : build_mageia_git build_WebRTC_git all_dockers_images all_dockers_ps all_dockers_rm all_dockers_stop all_services_docker_restart mydocker-join mydocker-swarm stop_all_dockers clear_images

On GPU computation nodes :

All informations on GPU devices on the node nvidia-smi or mode detailed nvidia-smi -q
with ls hardware you can search "3D controller" : lshw -xml
we need to create /dev/nvidia# with # in GPUlist=[0-4] for example for gpu in GPUlist; do nvidia-modprobe –create-nvidia-device-file=$gpu done #mknod /dev/nvidia0 c 195 0 mknod /dev/nvidiactl c 195 255 mknod /dev/nvidia-uvm c 241 0 mknod /dev/nvidia-uvm-tools c 241 1 mknod /dev/nvidia-modset c 195 254 chmod 666 /dev/nvidiactl /dev/nvidia-uvm /dev/nvidia-uvm-tools /dev/nvidia-modset

Docker with Swarm : Must install docker on each node, build and copy mageainvidia (for example) image on each node (or pull a Xvnc system from a hub).

Build mageia docker (used in TiledTest scripts) for example.

Copy some authorized keys to your frontend in TVConnections/mageianvidia/ssh (to make secure tunneling of vnc stream)
Build container : docker build -t mageianvidia -f TVConnections/mageianvidia/Dockerfile .

Singularity : Change rights with X launch on nodes /etc/pam.d/xserver #auth required pam_console.so auth sufficient pam_permit.so account sufficient pam_permit.so because of the creation of /tmp/.X#-lock lock file and /tmp/.X11-unix/X# socket on the host for nvidia acceleration inside the container.

On the forwarding frontend if you use OpenSSH sshd server, the server's GatewayPorts option needs to be enabled.