add_instance_to_interface improvement: sending a response via websockets after just shutting down/restarting doesn't work

[Connoropolous]

holochain-rust/crates/conductor_lib/src/interface.rs

Lines 664 to 676 in 9564458

	self.io
	.add_method("admin/interface/add_instance", move |params| {
	let params_map = Self::unwrap_params_map(params)?;
	let interface_id = Self::get_as_string("interface_id", &params_map)?;
	let instance_id = Self::get_as_string("instance_id", &params_map)?;
	let alias = Self::get_as_string("alias", &params_map).ok();
	conductor_call!(|c| c.add_instance_to_interface(
	&interface_id,
	&instance_id,
	&alias
	))?;
	Ok(json!({"success": true}))
	});

add_instance_to_interface causes clients of the interface who called this function to lose their connection, and thus lose their response to their call. I wonder whether you could add_instance_to_interface, send response, then restart interface outside of add_instance_to_interface instead?

for context, I am dealing with this here, and you will notice I have to do some pretty major workarounding
https://github.com/h-be/acorn-ui/blob/master/src/routes/Dashboard/Dashboard.js#L195-L228

This is where, within add_instance_to_interface, it restarts the interface

holochain-rust/crates/conductor_lib/src/conductor/admin.rs

Lines 369 to 371 in 9564458

	let _ = self.stop_interface_by_id(interface_id);
	sleep(Duration::from_millis(SWEET_SLEEP));
	self.start_interface_by_id(interface_id)?;

[Connoropolous]

I think Holoscape works around this by just refreshing a whole browser client page, but that is a very non-ideal solution

[StaticallyTypedAnxiety]

Wondering why the interface has to be restarted anywway,

• If the interface is already exists, it will now hit the restart subsection because of the return error
• is there a reason why an already running instance should be restarted? to apply changes maybe?

[Connoropolous]

@AshantiMutinta thanks :)

I am looking closer into it. The issue, (why it needs to restart the interface), is because the interface JSON-RPC calls are statically defined during interface start up, versus dynamically checked during JSON-RPC calls...
see here, traced manually along this call stack...

holochain-rust/crates/conductor_lib/src/conductor/admin.rs

Line 371 in 9564458

self.start_interface_by_id(interface_id)?;

holochain-rust/crates/conductor_lib/src/conductor/base.rs

Line 497 in 9564458

.and_then(|config| self.start_interface(&config))

holochain-rust/crates/conductor_lib/src/conductor/base.rs

Line 1268 in 9564458

let handle = self.spawn_interface_thread(config.clone());

holochain-rust/crates/conductor_lib/src/conductor/base.rs

Line 1342 in 9564458

let dispatcher = self.make_interface_handler(&interface_config);

holochain-rust/crates/conductor_lib/src/conductor/base.rs

Line 1338 in 9564458

conductor_api_builder.spawn()

holochain-rust/crates/conductor_lib/src/interface.rs

Line 115 in 9564458

self.setup_call_api();

holochain-rust/crates/conductor_lib/src/interface.rs

Lines 205 to 215 in 9564458

	fn setup_call_api(&mut self) {
	let instances = self.instances.clone();
	let instance_ids_map = self.instance_ids_map.clone();
	self.io.add_method("call", move |params| {
	let instances = instances.clone();
	let instance_ids_map = instance_ids_map.clone();
	let response = Self::method_call(params, instances, instance_ids_map)?;
	Ok(Value::String(response.to_string()))
	});
	}

Notice how the instances are stored as a hard reference outside of the JSON-RPC call.
Those assumptions would need to be re-wired differently in order to prevent the stopping and restarting of the interface