Gossip Strategy not working

[techitdeveloper]

• Configuration Used

config :libcluster,
  topologies: [
    example_cluster: [
      strategy: Cluster.Strategy.Gossip,
      config: [
        port: 45892
      ]
    ]
  ]

• Errors/Incorrect Behaviour Encountered

Application file_processor exited: FileProcessor.Application.start(:normal, []) returned an error: shutdown: failed to start child: Cluster.Supervisor
    ** (EXIT) shutdown: failed to start child: :example_cluster
        ** (EXIT) an exception was raised:
            ** (MatchError) no match of right hand side value: {:error, :eaddrinuse}
                (libcluster 3.3.3) lib/strategy/gossip.ex:127: Cluster.Strategy.Gossip.init/1
                (stdlib 6.1) gen_server.erl:2229: :gen_server.init_it/2
                (stdlib 6.1) gen_server.erl:2184: :gen_server.init_it/6
                (stdlib 6.1) proc_lib.erl:329: :proc_lib.init_p_do_apply/3

Description of issue

I cant get why its not working, if you want the complete code then this is it

https://github.com/techitdeveloper/File_ProcessingDistributed_System-Elixir

[Ind-Pandey]

I am also facing the same issue, Gossip is not working, i tried emp strategy and its working but its not as good as goosip.

If you fixed it or figured out the solution pls let me know

[keathley]

@techitdeveloper Based on your error message, you have two BEAMs both trying to bind the same port or address. Libcluster should be setting the socket flags for reuseport and reuseaddr. But there may be an issue on your specific OS or system config. Can you provide more information?

[keathley]

@Ind-Pandey Please provide information on what specifically is going wrong, under what circumstances, the OS you're using, your topology configuration, etc.

[infinityfye]

@techitdeveloper @Ind-Pandey I had this problem on windows.

The first search results point to this erlang/otp#4819 which in turn points to this erlang/otp#6522. Specifically:

Change behavior on Windows so that SO_REUSEADDR is only set if both reusaddr and reuseport have been set. This since SO_REUSEADDR on Windows behaves as BSD behaves if both SO_REUSEADDR and SO_REUSEPORT have been set.

I do a local patch for libcluster and it works like a charm. I changed reuse_port/0 in Cluster.Strategy.Gossip by adding a single clause to the case:

defp reuse_port() do
  case :os.type() do
    {:unix, os_name} ->
      cond do
        os_name in [:darwin, :freebsd, :openbsd, :linux, :netbsd] ->
          [{:raw, @sol_socket, @so_reuseport, <<1::native-32>>}]

        true ->
          []
      end

    {:win32, :nt} ->
      [reuseport: true]

    _ ->
      []
  end
end

However, there are a few catches. This behaviour is ONLY available on OTP 26+ AND only on windows. I believe even alpine supports this out of the box. So like @keathley said, without further information it's hard to pinpoint the issue.

If anyone is interested I can provide a minimal PR like the above where I check if String.to_integer(System.otp_release()) >= 26 then I add [reuseport: true] and otherwise keep the same flow.