partition: Improve bidi link down detection

src/classy_partition.erl

@@ -8,7 +8,8 @@ This module contains various algorithms for calculating network partitions.
 """.
 
 %% API:
--export([ bidi_link/3
+-export([ unid_link/3
+        , bidi_link/3
         , full_meshes/1
         ]).
 
@@ -29,6 +30,27 @@ This module contains various algorithms for calculating network partitions.
 %% API functions
 %%================================================================================
 
+-doc """
+Return @code{@{ok, Value@}} if node A is alive and appears in @code{ClusterInfo}.
+Then the value indicates whether A is currently connected to B.
+
+Error tuple is returned if A is unreachable.
+""".
+-spec unid_link(classy:cluster_info(), node(), node()) -> {ok, boolean()} | {error, _}.
+unid_link(ClusterInfo, NodeA, NodeB) ->
+  maybe
+    #{infos := #{NodeA := #{peers := PeersA}}} ?= ClusterInfo,
+    case find_peer_by_nodeid(NodeB, PeersA) of
+      {_Site, #{connected := Conn}} ->
+        {ok, Conn};
+      undefined ->
+        {ok, false}
+    end
+  else
+    _ ->
+      {error, insufficient_data}
+  end.
+
 -doc """
 Return @code{@{ok, true@}} if nodes are mutually connected to each other,
 or @code{@{ok, false@}} when either node considers the other disconnected.
@@ -40,6 +62,7 @@ then an error tuple is returned.
 bidi_link(ClusterInfo, NodeA, NodeB) ->
   case ClusterInfo of
     #{infos := #{NodeA := A, NodeB := B}} ->
+      %% Simple case: both nodes are present in the results:
       #{site := SiteA, peers := PeersA} = A,
       #{site := SiteB, peers := PeersB} = B,
       maybe
@@ -51,7 +74,15 @@ bidi_link(ClusterInfo, NodeA, NodeB) ->
           {ok, false}
       end;
     _ ->
-       {error, insufficient_data}
+      %% We may still prove that bidi link *doesn't* exist if either node
+      %% reports the other as disconnected:
+      case unid_link(ClusterInfo, NodeA, NodeB) =:= {ok, false} orelse
+           unid_link(ClusterInfo, NodeB, NodeA) =:= {ok, false} of
+        true ->
+          {ok, false};
+        false ->
+          {error, insufficient_data}
+      end
   end.
 
 -doc """
@@ -122,6 +153,22 @@ sort_partitions(Partitions0) ->
        } || I <- Partitions],
   [I || {_, _, I} <- lists:sort(L)].
 
+-spec find_peer_by_nodeid(node(), #{classy:site() => classy:peer_info()}) ->
+        {classy:site(), classy:peer_info()} | undefined.
+find_peer_by_nodeid(Node, Infos) when is_map(Infos) ->
+  do_find_peer_by_nodeid(Node, maps:iterator(Infos)).
+
+do_find_peer_by_nodeid(Node, It0) ->
+  case maps:next(It0) of
+    {Site, PeerInfo, It} ->
+      case PeerInfo of
+        #{node := Node} -> {Site, PeerInfo};
+        #{}             -> do_find_peer_by_nodeid(Node, It)
+      end;
+    none ->
+      undefined
+  end.
+
 -ifdef(TEST).
 
 full_mesh_test() ->

test/classy_SUITE.erl

@@ -583,7 +583,7 @@ t_091_node_of_site(_) ->
      , fun events_on_all_sites/1
      ]).
 
-t_092_bidi_link(_) ->
+t_092_link_detect(_) ->
   S1 = <<"s1">>,
   S2 = <<"s2">>,
   Sites = [S1, S2],
@@ -595,10 +595,25 @@ t_092_bidi_link(_) ->
        N2 = create_start_site(S2, #{}),
        #{cluster := Cluster} = ?ON(S1, classy_node:hello()),
        CInfo1 = classy:info([N1, N2]),
+       %% Sites are disconnected:
+       %%   Bidi:
        ?assertEqual(
           {ok, false},
           classy_partition:bidi_link(CInfo1, N1, N2),
           CInfo1),
+       ?assertEqual(
+          {ok, false},
+          classy_partition:bidi_link(CInfo1, N2, N1),
+          CInfo1),
+       %%  Unid:
+       ?assertEqual(
+          {ok, false},
+          classy_partition:unid_link(CInfo1, N1, N2),
+          CInfo1),
+       ?assertEqual(
+          {ok, false},
+          classy_partition:unid_link(CInfo1, N2, N1),
+          CInfo1),
        %% Join sites:
        ?ON(S2, classy:join_node(N1, join)),
        wait_site_joined(Sites, Cluster, S2),
@@ -608,13 +623,45 @@ t_092_bidi_link(_) ->
           {ok, true},
           classy_partition:bidi_link(CInfo2, N1, N2),
           CInfo2),
+       ?assertEqual(
+          {ok, true},
+          classy_partition:bidi_link(CInfo2, N2, N1),
+          CInfo2),
+       %% Uni-d as well:
+       ?assertEqual(
+          {ok, true},
+          classy_partition:unid_link(CInfo2, N1, N2),
+          CInfo2),
+       ?assertEqual(
+          {ok, true},
+          classy_partition:unid_link(CInfo2, N2, N1),
+          CInfo2),
        %% Stop one of the sites:
        stop_site(S2),
        ?block_until(#{?snk_kind := classy_peer_disconnected, remote := S2}),
        CInfo3 = classy:info([N1, N2]),
+       %% We can still derive that there's no bidirectional link:
        ?assertEqual(
-          {error, insufficient_data},
+          {ok, false},
           classy_partition:bidi_link(CInfo3, N1, N2),
+          CInfo3),
+       ?assertEqual(
+          {ok, false},
+          classy_partition:bidi_link(CInfo3, N2, N1),
+          CInfo3),
+       %% Unid:
+       ?assertEqual(
+          {ok, false},
+          classy_partition:unid_link(CInfo3, N1, N2),
+          CInfo3),
+       ?assertEqual(
+          {error, insufficient_data},
+          classy_partition:unid_link(CInfo3, N2, N1),
+          CInfo3),
+       %% Both sites are missing:
+       ?assertEqual(
+          {error, insufficient_data},
+          classy_partition:bidi_link(CInfo3, 'missing1@badhost', 'missing2@badhost'),
           CInfo3)
      end,
      [ fun no_unexpected_events/1