Why is `net_kernel:hidden_connect_node/1` undocumented?

Questions / Help
1

I’m using peer:start_link/1 to start a node, and I want to connect to it as a hidden node via shell. The net_kernel:hidden_connect_node/1 function does exactly what I want, but it’s undocumented. Why? Is there any other alternative?

2

There were some relevant discussions in the past:

3

Thanks, José!

It does not sound good to me. Is there a way to call peer:start_link/1 and pass the -hidden option to it?

A fully connected mesh is not an option for me. I need a “master” node connected to “slave” nodes, but “slave” nodes should not be connected.

My current implementation is something like this:

start(Name, Host, Port, Apps) when
    is_atom(Name), is_list(Host), is_integer(Port), is_list(Apps)
->
    {ok, Ref, Node} = peer:start_link(#{
        name => Name,
        host => Host,
        connection => Port
    }),
    ok = peer:call(Ref, code, add_paths, [code:get_path()]),
    [
        {ok, _} = peer:call(Ref, application, ensure_all_started, [App])
     || App <- Apps
    ],
    true = peer:call(Ref, erlang, set_cookie, [erlang:get_cookie()]),
    true = net_kernel:hidden_connect_node(Node),
    Ref.

It works as intended, but can I achieve this another way without the net_kernel:hidden_connect_node/1?

Any help is appreciated :slight_smile:

4

Ok, so I ended up starting all nodes via the terminal.

My goal is to have a structure like this:

foo * * master * * baz
           *
           *     
          bar

And not a mesh, like:

foo * * master * * baz
   *       *       *  
     *     *     *
       *  bar  *

And this is something I have ended up with:

  1. Start the slave nodes:
$ erl -sname foo@localhost -setcookie foo -connect_all false
$ erl -sname bar@localhost -setcookie bar -connect_all false
  1. Start and setup the master node:
$ erl -sname master@localhost -setcookie master
(master@localhost)1> erlang:set_cookie(foo@localhost, foo).
true
(master@localhost)2> net_kernel:connect_node(foo@localhost).
true
(master@localhost)3> erlang:set_cookie(bar@localhost, bar).
true
(master@localhost)4> net_kernel:connect_node(bar@localhost).
true
(master@localhost)5> nodes().
[foo@localhost,bar@localhost]
  1. Result in slave nodes:
(foo@localhost)1> nodes().
[master@localhost]

(bar@localhost)1> nodes().
[master@localhost]

The reason for keeping the slave nodes unconnected is security. I’ve no idea if this can be really secure since cookies intent is not security.

I would appreciate any good article, book, suggestion, video, or anything related to this topic.
I’m a newbie in distribution. It’s a challenging subject for me. Thanks!

Edit

I’ve just realized I can achieve the same result by using peer in the master node shell:

erl -sname master@localhost
Erlang/OTP 27 [erts-15.1] [source] [64-bit] [smp:8:8] [ds:8:8:10] [async-threads:1] [jit:ns]

Eshell V15.1 (press Ctrl+G to abort, type help(). for help)
(master@localhost)1> {ok, Foo, _} = peer:start_link(#{name => foo@localhost, args => ["-connect_all", "false"], connection => 0}).
{ok,<0.97.0>,foo@localhost}
(master@localhost)2> {ok, Bar, _} = peer:start_link(#{name => bar@localhost, args => ["-connect_all", "false"], connection => 0}).
{ok,<0.99.0>,bar@localhost}
(master@localhost)3> erlang:set_cookie(foo@localhost, foo).
true
(master@localhost)4> erlang:set_cookie(bar@localhost, bar).
true
(master@localhost)5> peer:call(Foo, erlang, set_cookie, [foo]).
true
(master@localhost)6> peer:call(Bar, erlang, set_cookie, [bar]).
true
(master@localhost)7> net_kernel:connect_node(foo@localhost).
true
(master@localhost)8> net_kernel:connect_node(bar@localhost).
true
(master@localhost)9> nodes().
[foo@localhost,bar@localhost]
(master@localhost)10> peer:call(Foo, erlang, nodes, []).
[master@localhost]
(master@localhost)11> peer:call(Bar, erlang, nodes, []).
[master@localhost]
1 Like
5

It seems that the Erlang distribution is not what I need in that case. I can always access the master node and all information about other nodes via RPC, so I’m considering moving to a gen_tcp interface. BTW, I’m still accepting suggestions :slight_smile:

Relevant sources:

6

Some time ago I created a MeshxNode package. In short it is a custom carrier protocol implementation using service mesh for transport and discovery. It would be possible to leverage for example Consul intentions to control communication between nodes. I didn’t update this for a long time, possibly it will not even work with current Consul version.

1 Like
7

Yes. In fact, you can start the peer node with any distribution option you like, or even with no distribution. You can also use it to test the distribution itself! Take a look at connection argument, specifically, standard_io control channel option.

With that option, you can still use peer:call to make RPC calls via control channel (not distribution).

It appears similar to structure created by this test case in the OTP codebase: otp/lib/kernel/test/pg_SUITE.erl at master · erlang/otp · GitHub

That’s exactly what standard_io was designed for. If you look at peer documentation, there is a Docker example that goes even further - it starts peer nodes in a separate container! Potentially, it could be a remote host just as well.

4 Likes
8

Thanks a lot! The example in the pg_SUITE is exactly what I was looking for. I don’t need net_kernel:hidden_connect_node/1 anymore.