Wiki - show how to start a remote server
This issue was created automatically from an original Mantis Issue. Further discussion may take place here.
Please update the wiki to show the very common case where a local client connects to a remote server. This remote server should be automatically started (not manually started by the user). Further, the server actually gets started from a login node. Use reverse connect, since this is very common. Add a second example, where ports are forwarded. (This one is complex.) Ask me for details, if needed.
The executive summary is as follows:
- A local script will ssh into the remote cluster login node, and run a server script. The local script will then start paraview, waiting for a connection.
- The remote script (running on the login node) will write an auto-generated script, written into /tmp. It then does a qsub, with this auto-gen script as input.
- The auto-gen script is now being run on the compute nodes, and starts pvserver with correct arguments.
For the case of port forwarding reverse connect, the only way I have ever gotten this to work is to have the cluster side script start an ssh back to the client machine. The issue here is that the client doesn't know what the compute node names will be when it starts - only the compute nodes know their names.
For the case of port forwarding forward connect, it is crazy. The client doesn't know the name of compute node 0, so you have to start the servers first. That cluster script then figures out it's node 0 name, and sends that back to the client side (I have done this as s scp file). After waiting an appropriate number of seconds (or some type of polling loop), have the client side read the compute node 0 name, and connect to it.