Feature: enable externally started nodes to connect to running simulation

[EskoDijk]

As discussed in #742 , it could be useful sometimes to (manually or scripted) launch a new node that connects to an already-running simulation. OTNS currently gets the node's startup event already. What it could do in that case:

• check if the node ID used is still 'free'
• if it is free, create a simulation node of a default type e.g. 'external'
• connect to the virtual-UART CLI of the new node

To consider in the design whether:

1. The new node is auto-configured with the current OTNS startup script, so that the new node gets the network key and other params.
2. The new node does not get any configuration - treated as setup type 'raw'.
3. There's a setting in the running simulation (OTNS) whether 1 or 2 above is applied. YAML file can specify the startup script for node types 'external' just like for other types like 'br'.

[gmarcosb]

This is critical simply because the flow for OTNS right now is broken

We have the OTNS class in python, which is great, this makes the user the controller & we can setup the environment: create nodes, setup test parameters, etc.

But now (as the controller) we want to actually run the test, control its inputs interactively, process its outputs, etc. - this can't be done currently because OTNS process creation is a blackbox & we lose control

To have proper tests based on OTNS, we need some hook into the thread devices, without having to do gymnastics (like setting up a pipe into that process)

Another reason it's critical: attaching a GDB debugger to the node (I'm guessing current AddNode method is going to have a hard time with debugger attachment, i.e. timeout)

[EskoDijk]

PR for this is in progress. Some design details:

• when a new node reports, and the claimed node ID is already taken, log a warning and close the socket.
• when exiting the simulation, or deleting the node, ensure to close sockets of external nodes.
• more robust handling of malformed incoming events: warn and close the socket, and delete the node, but don't stop the entire simulation. (See also Feature/change: keep a node that died in the simulation until manually removed #754 for the aspect of auto-deletion.)
• a new node type "EXT" (external) is defined

[gmarcosb]

Sweet, thank you! That sounds awesome

[EskoDijk]

PR now added: #755

[EskoDijk]

PR merged now, closing.